WEBVTT

1
00:00:21.150 --> 00:00:34.440
Manish Parashar: Good morning, everybody, welcome to the first sighs distinguished lecture of 2022 it's my absolute pleasure to introduce our speaker today, Dr Julio ibarra.

2
00:00:35.280 --> 00:00:51.630
Manish Parashar: he's a research professor in the knight foundation school of computer science in the college of engineering and computing and is the assistant Vice President in technology augmented research in the division of it at Florida international university.

3
00:00:52.950 --> 00:01:01.200
Manish Parashar: Dr bar is responsible for furthering the mission of the Center for Internet augmented research and assessment of Sierra.

4
00:01:01.500 --> 00:01:08.340
Manish Parashar: To contribute to the pace and the quality of research at fsu to the applications of advanced cyber infrastructure.

5
00:01:08.790 --> 00:01:21.630
Manish Parashar: under his leadership and stewardship nsf is funded the empath international exchange point and america's light paths am like network in its portfolio of international science infrastructure.

6
00:01:22.380 --> 00:01:30.420
Manish Parashar: Both these projects have been tremendously impactful a empath provides international research and education network connectors.

7
00:01:30.840 --> 00:01:40.890
Manish Parashar: with access to US production experimental backbone networks such as Internet to and, yes, net to facilitate international science, research and education collaborations.

8
00:01:41.310 --> 00:01:51.330
Manish Parashar: Am light is an international network backbone and interconnect the research and education networks in the US with peer networks in Latin America, the Caribbean and Africa.

9
00:01:51.900 --> 00:02:06.210
Manish Parashar: Dr Bowers research interests include software defined networks autonomic network architectures network automation and network control and management so without delaying any further, let me turn it over to Dr bar over to Julio.

10
00:02:08.040 --> 00:02:15.870
Julio Ibarra: diminish, thank you for having me and for the opportunity to present at the nsf distinguished lecture series.

11
00:02:17.010 --> 00:02:17.550
Julio Ibarra: My.

12
00:02:19.830 --> 00:02:25.380
Julio Ibarra: The topic of my talk is the family expressive protect network and.

13
00:02:26.640 --> 00:02:44.820
Julio Ibarra: me expressing protect operates as an international production network and platform for network innovation supporting research and education, and as you see here my introductory slide here, I am the principal investigator for the vm light project.

14
00:02:49.950 --> 00:03:02.310
Julio Ibarra: This is the outline of my talk and my objective is for you to gain an understanding of first what am I, is a production, research and education network.

15
00:03:03.300 --> 00:03:18.600
Julio Ibarra: How we're using emulate resources for network, innovation and how am I responding to challenges from science applications i'll start by giving an introduction to the home of emulated fit you followed by some history about Emily.

16
00:03:21.600 --> 00:03:31.740
Julio Ibarra: So Emma is managed from the Center for Internet augmented research and assessment Sierra in the division of it edify you as as minish mentioned.

17
00:03:32.880 --> 00:03:43.620
Julio Ibarra: Sierra is an interdisciplinary Center it supports and conducts research and education through the application of advanced cyber infrastructure.

18
00:03:44.550 --> 00:03:56.910
Julio Ibarra: Sierra consists of a small group of network engineers software developers and a support team to the right of my slide you can see a recent photo of the Sierra team.

19
00:03:58.320 --> 00:04:19.530
Julio Ibarra: Some of the goals of Sierra art to bridge technology gaps between researchers and it practitioners, so one of my roles is to leverage the expertise in the division of it and the College of engineering and computing to better inform our community about em light and projects at Sierra.

20
00:04:20.700 --> 00:04:36.180
Julio Ibarra: Another goal is to reinvigorate scholarship for undergraduate and graduate students engaging students to participate in projects is one of our core activities at Sierra in 2021 19 students participated in projects at Sierra.

21
00:04:37.470 --> 00:04:47.310
Julio Ibarra: Finally, Sierra aligns with fit goals as a public research university contributing to its research scholarship and technology development.

22
00:04:53.760 --> 00:05:00.780
Julio Ibarra: Some history about Emily Emily it was established in 2020 2010 under an iron see award.

23
00:05:03.210 --> 00:05:14.760
Julio Ibarra: From the office of cyber of advanced cyber infrastructure and like consists of a 20 year build out that includes connections to research and education in Latin America.

24
00:05:15.600 --> 00:05:36.690
Julio Ibarra: fit you lead the effort to link the US research and education networks to Latin America, in collaboration with Internet to global crossing and industry collaborators this collaboration resulted in the creation of an path as an international exchange point in Miami it back in 2000.

25
00:05:38.250 --> 00:06:00.780
Julio Ibarra: Emily built upon the accomplishments of the project we call read laila it's an iron see award that was made to fit you and scenic that proceeded emulate also part of the rnc portfolio Emily was one of the first to use optical spectrum, combined with least capacity on it's backbone.

26
00:06:01.860 --> 00:06:09.270
Julio Ibarra: Long term leases on optical spectrum was a key accomplishment we have these leases until 2032.

27
00:06:10.380 --> 00:06:16.620
Julio Ibarra: They provide value to us science facilities in Chile, South America can potentially Antarctica.

28
00:06:18.930 --> 00:06:33.570
Julio Ibarra: Emily was one of the first to deploy and operate its production network with software defined networking since 2014 SDN enable dynamic service provisioning and has significantly increased operations efficiency.

29
00:06:35.580 --> 00:06:55.920
Julio Ibarra: We also establish a committee, called the South American astronomy coordination committee or the sack sack provides a venue for the exchange of information and coordination between the US astronomy projects in Chile and the am late network operators sack is now in its 11th year.

30
00:06:57.060 --> 00:07:13.530
Julio Ibarra: We had a sack meeting in 2021 we hosted 61 participants each year that number appears to be growing with more interest in the work with astronomy in in South America and also branching out to Africa.

31
00:07:14.550 --> 00:07:18.840
Julio Ibarra: The 2021 sack report is available at the M like project website.

32
00:07:23.190 --> 00:07:26.310
Julio Ibarra: Some key factors that have enabled and like success.

33
00:07:27.540 --> 00:07:37.260
Julio Ibarra: First, support from nsf the office of advanced cyber infrastructure and the rnc program the rnc program has been critical to the success of Emily.

34
00:07:38.460 --> 00:07:44.790
Julio Ibarra: Support for you, if I you has encouraged our participation and nsf programs, such as the rnc.

35
00:07:46.050 --> 00:08:04.830
Julio Ibarra: And our partnerships with research and education network scan Latin America, the Caribbean and Africa have been key to emulate success they have been built upon layers of trust, over time, for example, sharing operations resources resources such as network bandwidth.

36
00:08:05.850 --> 00:08:09.090
Julio Ibarra: Colocation facilities network and compute resources.

37
00:08:10.950 --> 00:08:17.700
Julio Ibarra: The sharing of human resources, I can emphatically say that emulates accomplishments rest upon its people.

38
00:08:18.900 --> 00:08:28.320
Julio Ibarra: collaboration and cooperation amongst some of the most talented network engineers in the global r&d Community participate in the am like project.

39
00:08:33.090 --> 00:08:37.170
Julio Ibarra: Nuts let's let's take a look at MIT as it is today.

40
00:08:38.310 --> 00:08:43.170
Julio Ibarra: Am like operates as an international production, research and education network.

41
00:08:46.470 --> 00:08:55.710
Julio Ibarra: This is the production network operating today the network map shows and light footprint in the US, Latin America and Africa.

42
00:08:56.730 --> 00:09:15.030
Julio Ibarra: The light ring showing green consists of multiple 200 gigabit segments from boca raton to San Paulo boca raton to fortaleza some pollo to fortaleza boca raton to Cape Town and Sunday, I go to Porto Alegre.

43
00:09:17.010 --> 00:09:18.150
Julio Ibarra: The ring.

44
00:09:19.320 --> 00:09:32.010
Julio Ibarra: The protect ring in red carries primary traffic, as well as serves to protect the express network it consists of the following city pairs Miami fortaleza fortaleza some pollo.

45
00:09:33.210 --> 00:09:43.140
Julio Ibarra: Some pollo Santiago Santiago Panama Panama some point one and someone back to Miami that's the red ring you see here.

46
00:09:44.910 --> 00:09:47.850
Julio Ibarra: It aggregate we have 600 gigabits of capacity.

47
00:09:49.470 --> 00:10:00.150
Julio Ibarra: To the US, we also have a president ECHO presence it's open exchange points in Miami fortaleza some pollo Santiago and Cape Town.

48
00:10:04.050 --> 00:10:05.010
Julio Ibarra: This map here.

49
00:10:06.750 --> 00:10:17.220
Julio Ibarra: represents what am late envisions for the next five years in the current iron see award the goal of this plan network infrastructure is to increase capacity and resiliency.

50
00:10:18.600 --> 00:10:24.000
Julio Ibarra: it's a very busy slide on busy diagram showing the complexity of the network.

51
00:10:25.020 --> 00:10:43.170
Julio Ibarra: My reasons for showing it our first, the number of network segments and facilities operated by am like and its partners, you see there, and all of these solid and dash and dotted lines and let operate segments shown in three colors you can see here in this legend above.

52
00:10:45.840 --> 00:11:05.940
Julio Ibarra: All the other colors are network segments operated by am like partners, this is one of the successes of the rnc program is attracting participation among all the networks in many different countries to exchange and link to the US for supportive research and education.

53
00:11:06.960 --> 00:11:11.880
Julio Ibarra: And the logos on the right identify all of our partners in Emily.

54
00:11:13.380 --> 00:11:19.050
Julio Ibarra: The ellipses as you see here represent data centers for submarine cable landings.

55
00:11:20.340 --> 00:11:31.560
Julio Ibarra: The clouds represent external collaborators, such as Internet to an s nets lines represent network connections between facilities, the solid lines here.

56
00:11:32.820 --> 00:11:48.930
Julio Ibarra: The solid lines represent active connections as well dashed lines represent connections and dotted lines represent goals for the next four years, the blue rectangles represents cities, for example, San Paulo fortaleza Brazil Santiago, Chile.

57
00:11:51.660 --> 00:11:58.830
Julio Ibarra: And the light green rectangle above represents the empath open exchange port facilities in Florida and Georgia.

58
00:12:00.780 --> 00:12:11.670
Julio Ibarra: Planned increases in bandwidth capacity are MIT is adding 200 gigabits of capacity from some pollo the boca raton it's shown in this Green dashed line here.

59
00:12:12.330 --> 00:12:22.740
Julio Ibarra: That scheduled to be completed operating by 2023 and rb is adding 200 gigabits of capacity from fortaleza to Apollo that's this this blue.

60
00:12:23.700 --> 00:12:48.210
Julio Ibarra: dashed line that you see here, so all of this is increasing the capacity of within the region on the light network with our partners, contributing to to the the capacity so by 2023 Am I scheduled to have 800 gigabits about true aggregate capacity between the US and and South America.

61
00:12:52.560 --> 00:13:05.400
Julio Ibarra: Up to this point i've presented the MIT network infrastructure it's links aggregation points and bandwidth capacity, as it is today, and the enhancements plan to increase capacity and resilience.

62
00:13:06.510 --> 00:13:12.510
Julio Ibarra: Next, I will introduce embed network telemetry referred to as I empty.

63
00:13:13.560 --> 00:13:20.610
Julio Ibarra: it's a technology that is currently in production service, but we also consider it one of the most significant innovations.

64
00:13:24.060 --> 00:13:29.400
Julio Ibarra: So i'll start by describing the challenge and reasons for it.

65
00:13:31.620 --> 00:13:42.120
Julio Ibarra: So isolating and detecting faults of data transfers in long home that works with high latency such as as the light network is complex and time consuming.

66
00:13:43.050 --> 00:13:51.780
Julio Ibarra: it's both a technical challenge and a social challenge because isolating and detective detecting false is constrained by lack of visibility.

67
00:13:52.500 --> 00:13:59.880
Julio Ibarra: And oftentimes it's not just the technology that's needed but also the support and collaboration with peer networks which.

68
00:14:00.480 --> 00:14:16.470
Julio Ibarra: Sometimes, is not not easy when they have policies to not expose all the necessary data from their networks as a result, detecting what events cause performance degradation often result in questions that have incomplete answers.

69
00:14:17.520 --> 00:14:33.870
Julio Ibarra: For example, we don't really know where there is a packet loss and why it's happening or which path the packet took or how much time a packet queued at each switch effectively there's a lack of tools that provide enough visibility to the tech faults.

70
00:14:36.840 --> 00:14:50.550
Julio Ibarra: The slide shows network monitoring pain points when attempting to detect network printing events common network management tools fail to detect network transit events tools we've used for years i've just too limited.

71
00:14:51.930 --> 00:15:11.340
Julio Ibarra: Network transit events are short term and sporadic degradation in network performance they're caused by conditions that can lead to failures, over time, for example, attenuation on an optical channel, they often go undetected such as micro bursts, so what causes a microburst.

72
00:15:12.630 --> 00:15:17.430
Julio Ibarra: A data transfer that so short and time in the time domain, the tools cannot detect it.

73
00:15:18.660 --> 00:15:28.110
Julio Ibarra: It can be malicious or not, but it's it's not easy to detect the time scale can be as low as 100 milliseconds.

74
00:15:29.970 --> 00:15:31.470
Julio Ibarra: To hundreds of microseconds.

75
00:15:33.600 --> 00:15:39.240
Julio Ibarra: They can have a high impact, causing packet loss and long haul network latency such as emulate.

76
00:15:41.040 --> 00:15:49.170
Julio Ibarra: So what is an online solution to this to this condition and challenge with monitoring and limitations with tools.

77
00:15:50.940 --> 00:15:52.260
Julio Ibarra: inbound network telemetry.

78
00:15:53.310 --> 00:16:10.920
Julio Ibarra: I believe we are creating new methods to see deeper into the phenomena in our networks, similar to the instruments astronomers and high energy physicists create to see deeper into phenomena, it is a new method by which to instrument, the network for more granular visibility.

79
00:16:17.460 --> 00:16:20.970
Julio Ibarra: This slide describes what I empty is and what it does.

80
00:16:21.990 --> 00:16:29.580
Julio Ibarra: So it records network telemetry information in the packet while the packet to versus a path between two points in the network.

81
00:16:31.230 --> 00:16:38.250
Julio Ibarra: What is network telemetry information basically it's a snapshot of the state of the network provided as metadata.

82
00:16:40.260 --> 00:16:48.330
Julio Ibarra: telemetry reports are exported directly from the data plane, with no impact to the control plane, this is very important because typically.

83
00:16:49.410 --> 00:16:59.430
Julio Ibarra: there's with our traditional monitoring tools, the cpu can be severely impacted and oftentimes resources have to be dedicated for the function.

84
00:17:01.200 --> 00:17:10.680
Julio Ibarra: It tracks contract, monitor and evaluate every single packet at line rate and in real time, and this is unprecedented.

85
00:17:11.790 --> 00:17:22.560
Julio Ibarra: Some examples of network telemetry information collected are timestamp ingress port eagerness port keeper for utilization and many others.

86
00:17:23.880 --> 00:17:35.430
Julio Ibarra: As a result, visibility on the network is unprecedented the telemetry data is available to detect throughput issues due to bottlenecks failures or configuration errors.

87
00:17:40.110 --> 00:17:45.390
Julio Ibarra: So, how does it work, this is an example of a network instrumented with iot.

88
00:17:46.620 --> 00:17:49.350
Julio Ibarra: The diagram here on the on the right shows you.

89
00:17:50.850 --> 00:17:58.290
Julio Ibarra: The the source node on the left and a sink note on the right and the network is.

90
00:17:59.430 --> 00:18:04.140
Julio Ibarra: A five switches with iot enabled in the switches.

91
00:18:05.190 --> 00:18:15.960
Julio Ibarra: So the source node at the left of the diagram sends a packet and the source node is unaware, it is enabled in the network it doesn't need to know this information.

92
00:18:17.190 --> 00:18:28.620
Julio Ibarra: And this is illustrated here with number one in the path the job of the iot source switch this first one here is the portion that I empty header and metadata into the packet.

93
00:18:30.120 --> 00:18:33.510
Julio Ibarra: that's the first telemetry a header here.

94
00:18:36.780 --> 00:18:40.620
Julio Ibarra: Every it switch in the network path and push it as metadata into the packet.

95
00:18:42.000 --> 00:18:51.780
Julio Ibarra: Sony your number three you see all of these headers being put in here with telemetry information in there note the stacking of the IT meditate.

96
00:18:53.310 --> 00:19:02.370
Julio Ibarra: If I switch is not and I empty switch it just ignores the IMC content in the packet the I empty sinks, which.

97
00:19:03.390 --> 00:19:08.610
Julio Ibarra: This one here at the end of the network path and.

98
00:19:09.750 --> 00:19:17.190
Julio Ibarra: Its job is to extract the telemetry information, then, to forward the original packet to the sink node which is here.

99
00:19:18.630 --> 00:19:20.610
Julio Ibarra: The same it sinks, which.

100
00:19:22.230 --> 00:19:39.540
Julio Ibarra: Then forwards telemetry the telemetry report to the telemetry collector all of these headers and metadata here form the telemetry report it's all collected and forward it to the telemetry collected here on this computer.

101
00:19:41.880 --> 00:19:53.910
Julio Ibarra: The telemetry collector then receives parses processes and generates operational telemetry reports and there you have the ABC of of it, how it simply works.

102
00:19:55.800 --> 00:19:56.880
Julio Ibarra: it's not that complicated.

103
00:19:58.380 --> 00:20:11.460
Julio Ibarra: So here's a representative representation of a telemetry report with metadata and its data, so you can see here all of these variables are metadata identifiers.

104
00:20:12.510 --> 00:20:35.610
Julio Ibarra: A purse which level, and these are the metadata on a telemetry port level this here is an example of what the Meta data variable is such as uptime in time per switch switch 1234 and five, as illustrated in the previous slide and all of this data is now available.

105
00:20:37.680 --> 00:20:39.120
Julio Ibarra: For us to be able to.

106
00:20:41.010 --> 00:20:45.240
Julio Ibarra: understand what is going on in the network capturing.

107
00:20:47.460 --> 00:20:48.660
Julio Ibarra: Events per packet.

108
00:20:50.400 --> 00:21:00.870
Julio Ibarra: Now this this thick set of slides provide examples of how we're using the metadata and the granular visibility provided and what has with iot.

109
00:21:01.710 --> 00:21:10.380
Julio Ibarra: So interface utilization measured in the switch per packet, this is the from ingress egress all the way through this which.

110
00:21:11.310 --> 00:21:25.470
Julio Ibarra: it's useful for detecting micro bursts and i'll get to more detail about that, but as we define microburst earlier, these are bursts that occur within a very short time domain that are very difficult to detect unless you have the visibility.

111
00:21:26.520 --> 00:21:30.540
Julio Ibarra: We can see the start and the end of the micro bursts here in this graph.

112
00:21:32.070 --> 00:21:40.590
Julio Ibarra: The Spikes or 40 gigabytes I mean gigabits per second are classified as microburst these, these are the are the first few of these five.

113
00:21:42.660 --> 00:21:43.110
Julio Ibarra: and

114
00:21:44.130 --> 00:21:54.450
Julio Ibarra: The Spikes below 20 gigabits are classified as normal traffic, these are the ones shown here below the 20 gigabit line in green and seven orange.

115
00:21:55.680 --> 00:22:07.170
Julio Ibarra: The telemetry allows us to monitor bandwidth per interface and Q, so this gives us the ability to see exactly what's going on within the switch and how long that Pack is in there.

116
00:22:10.800 --> 00:22:13.410
Julio Ibarra: instantaneous egress interface buffer utilization.

117
00:22:14.520 --> 00:22:27.210
Julio Ibarra: we're now able to measure buffer utilization report of every switch, which is something we've never been able to do before we're using this metadata to evaluate que es policies and to detect sources of packet drops.

118
00:22:28.590 --> 00:22:34.290
Julio Ibarra: note the difference in the vertical scale between the normal buffer utilization and under congestion.

119
00:22:35.430 --> 00:22:45.120
Julio Ibarra: The graph on the left the vertical scale here is in kilobytes versus on the right that's in megabytes So you can see just.

120
00:22:46.290 --> 00:22:54.360
Julio Ibarra: The difference in scale between when buffers are operating at normal levels of utilization versus under congestion.

121
00:22:55.860 --> 00:23:05.910
Julio Ibarra: So what are the thresholds for normal and under congestion buffers We found that anything above 200 kilobytes becomes a problem that leads to congestion.

122
00:23:10.320 --> 00:23:11.460
Julio Ibarra: The specs Meta data.

123
00:23:12.600 --> 00:23:22.650
Julio Ibarra: allows us to monitor perhaps per packet forwarding delay it's useful for finding sources of gender along the path deter refers to variation and packet to link.

124
00:23:23.610 --> 00:23:34.860
Julio Ibarra: The graph on the Left shows normal buffer utilization the graph on the right shows congested buffers the vertical axis represents delay that you see here.

125
00:23:35.760 --> 00:23:51.060
Julio Ibarra: Note that under normal buffer utilization delay was measured in microseconds and if you can see that, clearly, but the scale here is in microseconds under congested buffers the delay was measured in milliseconds that's here on the on the right graph.

126
00:23:52.770 --> 00:23:59.310
Julio Ibarra: And you can see here some of the Spikes that were generated, for example, this first year occurred.

127
00:24:01.050 --> 00:24:13.320
Julio Ibarra: And you had these the the buffers congested reflecting that at about 65 milliseconds so to put things in perspective, consider the degree of delay in the right track here.

128
00:24:14.040 --> 00:24:26.340
Julio Ibarra: Some packets with buffer for like I said, approximately 65 milliseconds that's about how long a packet takes to transit from some pollo to Atlanta, to give you an idea of how long.

129
00:24:27.600 --> 00:24:31.590
Julio Ibarra: Under congestion conditions, these buffers and the being.

130
00:24:34.200 --> 00:24:35.490
Julio Ibarra: delayed within cubes.

131
00:24:40.920 --> 00:24:52.770
Julio Ibarra: To sum up, this is a use case of an experiment, we did to compare it and SNP we producing an event of micro bursts on a 100 gigabit link.

132
00:24:54.510 --> 00:24:59.760
Julio Ibarra: The experiment transmits five bursts of 40 to 50 gigabits.

133
00:25:03.270 --> 00:25:12.660
Julio Ibarra: of traffic for five seconds, the top graph represents an iot switch exporting metadata in real time per packet.

134
00:25:13.770 --> 00:25:16.530
Julio Ibarra: The anti graph shows five Spikes.

135
00:25:17.760 --> 00:25:24.000
Julio Ibarra: Lasting five seconds each generating 38 to 50 gigabits per second of traffic.

136
00:25:25.380 --> 00:25:29.010
Julio Ibarra: And these these Spikes are classified as microburst.

137
00:25:30.450 --> 00:25:38.520
Julio Ibarra: The bottom graph represents an ethernet switch it's pulling with SMP at its maximum rate of every 15 seconds.

138
00:25:39.900 --> 00:25:47.910
Julio Ibarra: The s&p graph shows to Spikes lasting 30 plus seconds with peaks of utilization of 13 gigabits per second.

139
00:25:50.310 --> 00:25:54.930
Julio Ibarra: they're not report they're not reported as microbus given that they're there.

140
00:25:56.190 --> 00:25:58.860
Julio Ibarra: Within a 15 second time scale.

141
00:26:02.220 --> 00:26:07.710
Julio Ibarra: So this clearly shows that the identity graph is a more realistic representation of network events.

142
00:26:08.880 --> 00:26:19.800
Julio Ibarra: And then find an important finding, though, is that a legacy protocols, such as s&p is not enough to characterize the microburst and to determine their impact.

143
00:26:21.990 --> 00:26:30.480
Julio Ibarra: From another perspective as an MP reported that he goes ation of the hundred gigabit link was at most 13 gigabits per second.

144
00:26:31.230 --> 00:26:47.790
Julio Ibarra: Giving a false impression there's more available bandwidth and it really is so if you're using us an MP other tools will be needed in addition to accurately characterize events legacy protocols are just not going to be sufficient to do this.

145
00:26:49.380 --> 00:26:57.900
Julio Ibarra: As a result, observing microburst becomes straightforward when you're monitoring instrument provides full visibility of network events such as I empty.

146
00:27:03.480 --> 00:27:06.060
Julio Ibarra: So let's review what was previously presented.

147
00:27:10.110 --> 00:27:10.380
Julio Ibarra: We.

148
00:27:12.600 --> 00:27:16.140
Julio Ibarra: recovered reasons why amway was an early adopter of it.

149
00:27:17.370 --> 00:27:28.410
Julio Ibarra: how it works what it metadata uses for deeper visibility into the network to improve operations and support it's already communities.

150
00:27:30.090 --> 00:27:44.010
Julio Ibarra: Comparison of it is an MP for measuring micro bursts, and this next section opposite how am I operates as a platform for network innovation alongside operating as a production, research and education network.

151
00:27:47.340 --> 00:27:51.930
Julio Ibarra: This slide represents a snapshot of the M light SDN architecture.

152
00:27:53.490 --> 00:28:13.110
Julio Ibarra: SDN has been essential to foster innovation on me like the blue boxes represent southbound interfaces southbound interfaces introduce an abstraction between the traditional forwarding control planes the yellow boxes represents the ketosis SDN platform.

153
00:28:14.730 --> 00:28:23.820
Julio Ibarra: He does performs a function of the control it's responsible for interfacing own network applications and southbound protocols.

154
00:28:25.920 --> 00:28:35.550
Julio Ibarra: The Green boxes represent the keto Spyker applications I keep those my application performs a specific task for the control plane.

155
00:28:36.630 --> 00:28:51.090
Julio Ibarra: Green yellow and blue collectively form the control plane, the pink layer represents higher layer applications to support business services, such as routing troubleshooting and provisioning etc.

156
00:28:52.440 --> 00:29:06.180
Julio Ibarra: And the ellipses you see there at the top, our applications or interfaces for users to make service requests for example the ether that virtual circuit manager the ABC manager, you see here in this ellipse.

157
00:29:08.070 --> 00:29:11.520
Julio Ibarra: leveraging multiple applications from the pink layer

158
00:29:12.690 --> 00:29:16.860
Julio Ibarra: which then call upon the the green layer for specific.

159
00:29:18.030 --> 00:29:19.740
Julio Ibarra: operations with the control plane.

160
00:29:22.050 --> 00:29:34.560
Julio Ibarra: The red box is what we refer to as the optical and packet telemetry collector over here on the right it collects streaming telemetry from multiple sources, including the optical later.

161
00:29:35.760 --> 00:29:43.320
Julio Ibarra: The job of the op etc, is to detect network events at the optical and packet layers that could result in failures.

162
00:29:44.610 --> 00:29:50.790
Julio Ibarra: The purple box is the behavior anomaly and performance management bpm.

163
00:29:52.530 --> 00:29:55.050
Julio Ibarra: bpm interprets network status updates.

164
00:29:56.280 --> 00:30:13.380
Julio Ibarra: And network telemetry reports from the op etc and other sources, the bpm applies learning algorithms searches for specific patterns, then notifies the SDN controller when nonconformity is detected and by nonconformity I mean.

165
00:30:14.460 --> 00:30:32.640
Julio Ibarra: Not compliance or out of range with policies that have been already programmed into the resources of the network, so this is how we're able to use it to detect if if if things are operating normally or not normally.

166
00:30:34.110 --> 00:30:43.620
Julio Ibarra: be a PM leveraging that we're calamity to learn the current state of the network and then respond if network anomalies are potentially impacting science applications.

167
00:30:45.000 --> 00:30:49.380
Julio Ibarra: The op etc, and the bpm form the management plane of the architecture.

168
00:30:53.370 --> 00:31:08.370
Julio Ibarra: This slide shows how we're applying the the architecture to the IT infrastructure that MIT is deploying at each of the sites, you can see here in this figure the.

169
00:31:10.680 --> 00:31:16.380
Julio Ibarra: Each MIT site for too late, for example, the US and Brazil, I empty switches.

170
00:31:17.550 --> 00:31:27.090
Julio Ibarra: generate telemetry reports here in the in the data plane a telemetry collector to process telemetry reports is also part of.

171
00:31:27.630 --> 00:31:38.580
Julio Ibarra: insight and applications and database to interpret can display telemetry reports is also part of the the infrastructure, so what this.

172
00:31:39.120 --> 00:31:50.460
Julio Ibarra: What this figure here shows is that every packet becomes a telemetry report and each telemetry collector can parse up to 2 million packets per second of telemetry reports.

173
00:31:51.060 --> 00:32:00.360
Julio Ibarra: And this is unprecedented because before I empty network devices and never management tools to about process to limit reports on a per packet level.

174
00:32:01.200 --> 00:32:12.840
Julio Ibarra: And we're doing this and each one of these sites, which is our ad exchange points to be able to track every single packet as it goes through every I empty switch in the network.

175
00:32:14.460 --> 00:32:25.980
Julio Ibarra: Our goal is for me to be fully instrumented with I empty by the second quarter of 2022.

176
00:32:27.990 --> 00:32:28.470
Julio Ibarra: So.

177
00:32:29.790 --> 00:32:36.540
Julio Ibarra: To look at where we are at this point um we've covered the am light SDN architecture.

178
00:32:37.800 --> 00:32:49.800
Julio Ibarra: I described the optical and packet telemetry collector and the behavior anomaly and performance manager as intimations leveraging network telemetry and described it appointment to instrument, the light network.

179
00:32:51.270 --> 00:32:58.830
Julio Ibarra: The next innovation to present is a subset of autonomic networking that we refer to as closed loop orchestration.

180
00:33:03.060 --> 00:33:04.290
Julio Ibarra: So let me take.

181
00:33:06.420 --> 00:33:09.480
Julio Ibarra: A moment here for a brief review of autonomic networking.

182
00:33:12.210 --> 00:33:17.820
Julio Ibarra: So autonomic networking autonomic systems were first described in 2001.

183
00:33:19.080 --> 00:33:25.950
Julio Ibarra: There is a paper by gephardt and chess in 2003 on autonomic computing that.

184
00:33:27.180 --> 00:33:29.640
Julio Ibarra: describes the concepts and what they mean.

185
00:33:31.230 --> 00:33:37.830
Julio Ibarra: And they were used by IBM for many of their systems so autonomic computing is not a new idea.

186
00:33:39.420 --> 00:33:53.460
Julio Ibarra: autonomic networking adopted many of those concepts and they have been documented in the Iit pipe by the ETF in rfc documents such as 7575 as well as other rfc.

187
00:33:54.960 --> 00:34:06.630
Julio Ibarra: The fundamental goal with autonomic networking and system is self management it's comprised of several self star properties.

188
00:34:08.160 --> 00:34:14.250
Julio Ibarra: One of its goals is to reduce dependencies on your administrators or centralized management systems.

189
00:34:16.920 --> 00:34:20.130
Julio Ibarra: And to adapt to a changing environment.

190
00:34:21.960 --> 00:34:22.350
Julio Ibarra: The.

191
00:34:23.730 --> 00:34:38.850
Julio Ibarra: The other the other important characteristic about condom applicant is a closed loop control, this is a mechanism for self management functions that includes typically collecting analyzing deciding and acting on.

192
00:34:40.920 --> 00:34:49.230
Julio Ibarra: An acting process, it does this in a forever loop, and so we refer to this closed loop control as.

193
00:34:50.730 --> 00:34:52.020
Julio Ibarra: closed loop orchestration.

194
00:34:53.400 --> 00:34:58.230
Julio Ibarra: So, so why is am like developing armored functions into its production operation.

195
00:34:59.700 --> 00:35:08.280
Julio Ibarra: One reason is to increase operational efficiency by reducing dependency on knock operators and central management systems.

196
00:35:09.570 --> 00:35:23.970
Julio Ibarra: But an even bigger reason is a technology pool that's coming from science applications with a requirement for SLA grade network resilience and I have an example coming up to explain what I mean.

197
00:35:26.550 --> 00:35:30.360
Julio Ibarra: But first let's look at where we are today with.

198
00:35:32.220 --> 00:35:47.190
Julio Ibarra: Our networking and what we call closed loop orchestration so this table represents categories of self management as a as a continuum towards the left side our categories, with more human dependency.

199
00:35:48.750 --> 00:35:52.920
Julio Ibarra: The right side describes categories, with less human dependency.

200
00:35:54.000 --> 00:36:05.220
Julio Ibarra: User input towards the left requires more prescriptive information as we move to the right user inputs consists of responding to conditions and triggers.

201
00:36:05.730 --> 00:36:22.950
Julio Ibarra: and responding to us or policies, as opposed to prescriptive information and light has evolved from the left to the right since its adoption of software defined networking and emulate is currently around the middle of closed loop orchestration around here.

202
00:36:24.390 --> 00:36:35.610
Julio Ibarra: And so, with our current project, we want to get as close to autonomic as possible, for example, having autonomic provisioning and operation of layer to vpn.

203
00:36:36.870 --> 00:36:45.510
Julio Ibarra: is a goal for the current project layer to vpn is a very common service that research and education networks provide.

204
00:36:47.100 --> 00:36:51.330
Julio Ibarra: So our goal is to move more services towards autonomic over time.

205
00:36:55.620 --> 00:37:09.120
Julio Ibarra: This slide here represents a use case to self optimize the M light network or reason for self optimizing is to respond to network transit events in near real time if you recall.

206
00:37:10.140 --> 00:37:21.210
Julio Ibarra: These transit moments occur in a very short time scale, so we have to be able to do this as close to real time as possible today optimization is a manual and iterative process.

207
00:37:22.860 --> 00:37:29.430
Julio Ibarra: The diagram shows three sites in the ham like domain, we have Florida Chile and Brazil.

208
00:37:30.570 --> 00:37:37.260
Julio Ibarra: This is representative of the slide that I showed earlier of where we will have the.

209
00:37:38.280 --> 00:37:42.990
Julio Ibarra: The light SDN infrastructure and telemetry infrastructure deploy.

210
00:37:44.670 --> 00:37:46.410
Julio Ibarra: The data plane and green.

211
00:37:47.430 --> 00:37:54.210
Julio Ibarra: That you see here exports it reports to the telemetry collector represented here in blue.

212
00:37:55.680 --> 00:37:59.220
Julio Ibarra: In turn, the telemetry collectors export telemetry summaries.

213
00:38:00.960 --> 00:38:08.910
Julio Ibarra: To the learning systems, which includes the behavior anomaly and performance manager the bpm.

214
00:38:11.430 --> 00:38:14.670
Julio Ibarra: What when when nonconformity is detected.

215
00:38:15.870 --> 00:38:18.270
Julio Ibarra: The learning systems will send alarms to.

216
00:38:19.470 --> 00:38:32.520
Julio Ibarra: The the Doc where we have human intervention, if necessary, as well as to the SDN orchestrator where there's less dependency on human intervention.

217
00:38:34.350 --> 00:38:45.000
Julio Ibarra: And the SDN orchestrator sends instruction then to the control plane, which are shown here in red segments, the program the data plane.

218
00:38:46.200 --> 00:39:04.290
Julio Ibarra: If the controls are there to to allow that so all of that is is under operator control it's not like it's something that the network will now be self driven there's always a way to have operators.

219
00:39:05.370 --> 00:39:17.070
Julio Ibarra: First, determine if it's necessary to intervene or through allow for self driving of the network, so this this is representative of the closed loop orchestration system.

220
00:39:18.780 --> 00:39:35.040
Julio Ibarra: So, essentially, just to recap we're going from the bottom up here into the learning systems with telemetry summaries the interpretation occurs here and actions, then and Sue to be able to do this in a closed loop fashion fashion.

221
00:39:36.630 --> 00:39:40.740
Julio Ibarra: Now this is, this is the approach that we're taking for self self optimization.

222
00:39:41.760 --> 00:39:47.970
Julio Ibarra: To be able to respond to the the networking events in real time.

223
00:39:49.260 --> 00:39:50.490
Julio Ibarra: So i'm.

224
00:39:51.840 --> 00:40:13.020
Julio Ibarra: The roadmap to self optimizing shows the improvements plan as autonomic functions and learning algorithm arm are improved, so today where where we're working around five seconds by year three we we plan to improve to to your four to one and your five to 500 milliseconds.

225
00:40:18.450 --> 00:40:36.180
Julio Ibarra: So we've covered autonomic networking and now we've also compared closed loop orchestration with automatic automation and autonomic and presented a use case to self optimize am like that applies autonomic functions and closed loop orchestration.

226
00:40:37.890 --> 00:40:45.480
Julio Ibarra: Next let's let's take a look at how we are applying these innovations on em light as a production network to support science.

227
00:40:48.840 --> 00:40:56.220
Julio Ibarra: let's start with a service level agreement SLA agreement driven use case.

228
00:40:57.240 --> 00:40:58.620
Julio Ibarra: The verb Observatory.

229
00:41:00.630 --> 00:41:07.410
Julio Ibarra: very rude, but it is a large aperture wide field ground based optical telescope under construction in northern Chile.

230
00:41:10.230 --> 00:41:21.030
Julio Ibarra: it's an 8.4 meter telescope it will take a picture of the southern sky every 27 seconds and will produce a 30 gigabyte data set.

231
00:41:23.250 --> 00:41:36.510
Julio Ibarra: Every 27 seconds each dataset must be transferred to the US Data facility at slack in menlo park California within five seconds inside that 27 second transfer window.

232
00:41:37.650 --> 00:41:45.900
Julio Ibarra: So and white, has, along with all the networks participating in this very reuben network we have five seconds to get the.

233
00:41:46.980 --> 00:41:51.210
Julio Ibarra: The data set from northern chilly here.

234
00:41:52.230 --> 00:41:58.230
Julio Ibarra: In a town called la serena with the base station is all the way to slack in menlo park California.

235
00:42:01.080 --> 00:42:06.870
Julio Ibarra: Some of the challenges that we are are facing here with with this use case is.

236
00:42:08.310 --> 00:42:10.440
Julio Ibarra: The the high propagation delay.

237
00:42:11.610 --> 00:42:33.780
Julio Ibarra: In the end, to end path the round trip time from the base station and loss arena she led to the US Data facility and slack is approximately 180 plus milliseconds we've computed that a point 00 1% packet loss will compromise the Ruby Observatory application.

238
00:42:35.250 --> 00:42:38.040
Julio Ibarra: So this is very, very time sensitive.

239
00:42:42.300 --> 00:42:47.400
Julio Ibarra: So this slide represents the very reuben workflow cadence and.

240
00:42:48.480 --> 00:42:52.740
Julio Ibarra: we're trying to represent here, the what happens in the 27 seconds.

241
00:42:54.120 --> 00:42:59.880
Julio Ibarra: That I described in the previous slide so the 27 seconds is for data.

242
00:43:02.160 --> 00:43:11.100
Julio Ibarra: is for the data set transfer window 22 seconds is to gather and process the image shown here in white and.

243
00:43:12.540 --> 00:43:23.130
Julio Ibarra: five seconds for the transfer to the US Data facility shown in blue over here Okay, at least 40 gigabits per second of dedicated bandwidth is the estimated.

244
00:43:24.510 --> 00:43:27.810
Julio Ibarra: level to achieve the data set transfer within five seconds.

245
00:43:29.550 --> 00:43:45.000
Julio Ibarra: So what if a condition occurs that results in packet loss shown here at this point the network must react within 22 seconds of the next data set process window in order to not miss the data transfer window.

246
00:43:46.500 --> 00:43:55.710
Julio Ibarra: Without proper instrumentation because of the complexity and troubleshooting a distributed multi domain network topology such as their rubens.

247
00:43:57.210 --> 00:43:58.050
Julio Ibarra: is likely.

248
00:43:59.400 --> 00:44:08.220
Julio Ibarra: This this condition is likely to impact the following data transfer window and that's shown here with this Red Square and this X.

249
00:44:11.430 --> 00:44:13.650
Julio Ibarra: And the following wants to, but you see here.

250
00:44:14.760 --> 00:44:19.080
Julio Ibarra: So, depending on the type of issue, for example, soft issues such as a packet loss.

251
00:44:20.160 --> 00:44:26.010
Julio Ibarra: It can take hours days and possibly weeks to mitigate this problem, unlike a hard.

252
00:44:27.570 --> 00:44:28.980
Julio Ibarra: Issues such as a fiber cut.

253
00:44:30.000 --> 00:44:37.110
Julio Ibarra: It could take quite a bit of time with with packet loss as I covered previously it's hard to know where it's happening.

254
00:44:38.340 --> 00:44:39.870
Julio Ibarra: In in a short amount of time.

255
00:44:41.220 --> 00:44:49.710
Julio Ibarra: So it looks like the very reuben Observatory needs a network that's properly instrumented I can respond to transit events in real time.

256
00:44:51.300 --> 00:44:53.730
Julio Ibarra: So let's uh let's take a look.

257
00:44:57.150 --> 00:45:09.510
Julio Ibarra: emulators instrumented for SLA grade network resilience for the rubber very reuben Observatory the expressive protect paths are instrumented with iot and persona.

258
00:45:10.110 --> 00:45:21.060
Julio Ibarra: From Santiago, Chile to Atlanta Georgia, where Am I will hand off to the ies that network, you can see here all of this instrumentation is in place.

259
00:45:22.170 --> 00:45:38.640
Julio Ibarra: from Chile, all the way to to Georgia, we have the it switches we have dual connectivity here, with the primary express path and the protect path, and we have our closed loop orchestration here with the iot collector per site.

260
00:45:40.770 --> 00:45:47.760
Julio Ibarra: So instrumented with iot and lights management plane is processing telemetry reports.

261
00:45:48.450 --> 00:45:58.440
Julio Ibarra: Isolating and detecting traffic anomalies validating performance thresholds and computing risk profiles of optical links and IP layer metrics.

262
00:45:59.340 --> 00:46:13.470
Julio Ibarra: So if a transit event were to occur during the data set transfer window packet loss by impact the transfer the event will be detected in the iot reports so we may lose one image.

263
00:46:14.490 --> 00:46:15.330
Julio Ibarra: Transfer window.

264
00:46:18.810 --> 00:46:29.730
Julio Ibarra: The the instructions will program the data plane to provision an alternate path before the next data transfer window is is is set.

265
00:46:30.450 --> 00:46:37.530
Julio Ibarra: The the closed loop orchestration repeat this process for every packet as it traverses the network so we're able to.

266
00:46:38.100 --> 00:46:50.160
Julio Ibarra: Have the detection mechanism in place to be able to not miss the next window so much and not miss the next day that's made us a transfer window, as a result of the instrumentation that's in place.

267
00:46:52.050 --> 00:47:04.860
Julio Ibarra: So, finally, our metric for success is to not miss that data transfer window, and this is what we're instrument thing the network to be able to achieve from Aruba as well as for other SLA driven applications.

268
00:47:08.580 --> 00:47:10.200
Julio Ibarra: Emily also supports fabric.

269
00:47:11.880 --> 00:47:13.620
Julio Ibarra: And let will be providing a.

270
00:47:14.850 --> 00:47:16.800
Julio Ibarra: Dedicated optical.

271
00:47:18.000 --> 00:47:30.660
Julio Ibarra: path between the fabric notify you and the Atlantic or note, if I, you will be hosting one we were expecting it to arrive within the next few months.

272
00:47:31.530 --> 00:47:46.380
Julio Ibarra: We will have multiple hundred gigabits stitching points to these to these fabric nodes in Atlanta stitching points to yes, that and Internet to as well as Pam path in Miami.

273
00:47:48.000 --> 00:47:48.660
Julio Ibarra: We have.

274
00:47:49.920 --> 00:48:01.830
Julio Ibarra: This will be this will be available in Miami as well as an other locations that you see in the footprint in we saw on the footprint in South America, and even in in Cape Town.

275
00:48:03.570 --> 00:48:20.250
Julio Ibarra: and up to 50 gigabits per seconds of available bandwidth capacity over am light links during experiments to support the goal for reproducibility on fabric experiments will have access to per packet telemetry in real time as well.

276
00:48:22.980 --> 00:48:35.430
Julio Ibarra: This is a network diagram to show the the network as planned for for fabric, you see the fabric edge notify you year and the fabric or note in Atlanta.

277
00:48:36.060 --> 00:48:47.100
Julio Ibarra: there's multiple hundred gigabit path between the the fabric nodes and if I you and the coordinate in Atlanta that leverages the infrastructure that's in place.

278
00:48:47.820 --> 00:48:55.680
Julio Ibarra: compute and storage devices services at open exchange points will be available and path and these exchange points here.

279
00:48:56.520 --> 00:49:05.160
Julio Ibarra: These exchange points will be operating as software defined exchanges, the blue rectangles they will support close the orchestration.

280
00:49:05.640 --> 00:49:19.710
Julio Ibarra: For exchange point in across multiple network domains, so my main takeaway is to note that the M like backbone is instrumented with real time in bed network telemetry that fabric can leverage for its experiments.

281
00:49:22.050 --> 00:49:37.680
Julio Ibarra: Other science Community supported on emulate the large hadron collider open network environment polizzi one the open science grid partnership to advance supercomputing the event horizon telescope.

282
00:49:38.910 --> 00:49:47.580
Julio Ibarra: Ground based telescopes in Chile and South Africa, the event horizon telescope actually it's an interesting story, because that involves will eat all the.

283
00:49:48.900 --> 00:49:53.280
Julio Ibarra: The projects in the rnc program not just me right.

284
00:49:54.750 --> 00:50:02.640
Julio Ibarra: So this gives you an idea of all the different communities that can benefit from the the instrumentation that we will be operating on.

285
00:50:05.640 --> 00:50:07.020
Julio Ibarra: The MIT team.

286
00:50:09.900 --> 00:50:27.390
Julio Ibarra: Is is here, we have myself is principal investigator Toronto bizarro is the chief network architect he leads the and the engineering teams and also the software development teams, for all of the software that I described in this presentation.

287
00:50:28.950 --> 00:50:48.210
Julio Ibarra: But soca check out over is our lead for outreach and education, Heidi Morgan at usc is the lead for research engagement chip Cox of vanderbilt university is our coordinator for operations.

288
00:50:49.290 --> 00:50:56.760
Julio Ibarra: We say Professor least Lopez at the University of San Paulo and if I you, these are chair for the Research Committee.

289
00:50:58.110 --> 00:51:05.280
Julio Ibarra: And Eduardo was n D R amp D he's our Chair for for the engineering committee.

290
00:51:07.680 --> 00:51:10.590
Julio Ibarra: And I left for the last slide.

291
00:51:12.150 --> 00:51:15.540
Julio Ibarra: The the Julio in one slide.

292
00:51:16.590 --> 00:51:25.620
Julio Ibarra: description, and so I will just briefly mentioned or answer the question of when how and why did that I decided to go to pursue a research career.

293
00:51:26.970 --> 00:51:38.190
Julio Ibarra: I will give credit to much encouragement I received from a Vice President at nyu many, many years ago and and also a family member, I must say that.

294
00:51:39.750 --> 00:51:42.120
Julio Ibarra: Entering graduate school and then.

295
00:51:43.680 --> 00:51:47.340
Julio Ibarra: Pursuing my PhD was was a journey and.

296
00:51:48.360 --> 00:51:57.870
Julio Ibarra: There was inspiration from colleagues and TEAM members and also motivation for my PhD professor, I will say the experience was transformational.

297
00:51:59.910 --> 00:52:05.070
Julio Ibarra: It has really helped me to be more mindful and effective of.

298
00:52:06.180 --> 00:52:09.480
Julio Ibarra: The the importance of working with.

299
00:52:10.500 --> 00:52:23.640
Julio Ibarra: Research faculty and it practitioners and the importance of them working together effectively and I give much credit to the the programs nsf has produced to enable us to do this.

300
00:52:27.000 --> 00:52:31.920
Julio Ibarra: Some references if you're interested in more information about the topic site I covered.

301
00:52:33.090 --> 00:52:36.270
Julio Ibarra: And thank you very much for your time and the.

302
00:52:37.320 --> 00:52:40.980
Julio Ibarra: The opportunity to present this to to all of you, thank you.

303
00:52:43.200 --> 00:52:50.670
Manish Parashar: Thank you, Julio for absolutely brilliant talk on the skiff virtual round of applause TULIO.

304
00:52:54.270 --> 00:52:54.990
Elements of.

305
00:52:56.220 --> 00:53:02.400
Manish Parashar: it's just amazing to see the tremendous impact and broad impact your research is having.

306
00:53:03.600 --> 00:53:20.790
Manish Parashar: You know, both in the networking research community, but also in the science broader science community by the research and enables to these remote instruments so with that, let me open it up for questions, I know there are two lined up in the Q amp a.

307
00:53:22.500 --> 00:53:37.680
Manish Parashar: box i'll start with that the first question is from Sally o'connor and she's asking our redundancies addressed in the network, why was boca raton selected instead of fit campus.

308
00:53:39.720 --> 00:53:41.760
Julio Ibarra: The reason for boca raton.

309
00:53:42.810 --> 00:53:57.870
Julio Ibarra: Is that the the undersea cable system lands in boca raton and there is a data Center that the the the cable is is terminated X.

310
00:53:58.320 --> 00:54:13.080
Julio Ibarra: And it was just more effective for us to establish a point of presence there to minimize the latency for the Vera ruben observed the variable Observatory If you recall.

311
00:54:14.460 --> 00:54:31.290
Julio Ibarra: Where we have we have five seconds to get that 13 gigabyte data set to to the data facility at slack so by by going North directly from boca raton where the cable lands were able to.

312
00:54:32.790 --> 00:54:41.280
Julio Ibarra: Keep the the latency to a minimum, as opposed to coming all the way down to Miami and then going North again and that's mainly the reason why.

313
00:54:44.280 --> 00:54:54.300
Manish Parashar: Thank Thank you and again a reminder to everybody if you have questions, please enter them in the Q a box and i'll read them out.

314
00:54:55.020 --> 00:54:58.230
Manish Parashar: So we have another question from Jennifer schaaf.

315
00:54:59.220 --> 00:55:12.570
Manish Parashar: She asks on slide 28 our experiences with other astronomy applications, such as the hd has shown that the problem with transferring the files isn't the long haul networks such as.

316
00:55:13.230 --> 00:55:18.240
Manish Parashar: The ones you're supporting, but the network between instruments and the first exchange point.

317
00:55:18.630 --> 00:55:36.120
Manish Parashar: Which is generally less than one gigabit per second network or a controller computer with less than one gigabit transfer rate how have you dealt with these high level bottlenecks, which are more common and perform performance impacting then micro events.

318
00:55:37.800 --> 00:55:40.170
Julio Ibarra: Thank you Jennifer for your question.

319
00:55:41.340 --> 00:55:52.440
Julio Ibarra: That is is a reality that we have to to to work with that I classify that more as a social engineering challenge than a technology challenge because.

320
00:55:53.340 --> 00:56:02.250
Julio Ibarra: Many of these instruments are just not connected at the level that they should be, and Jennifer has been leading this effort for the IMC.

321
00:56:03.390 --> 00:56:06.090
Julio Ibarra: group she has a very.

322
00:56:07.290 --> 00:56:18.930
Julio Ibarra: Important program called epic that works with researchers and makes it very easy for them to understand where there are bottlenecks and what needs to be done.

323
00:56:20.040 --> 00:56:33.030
Julio Ibarra: Network telemetry essentially is providing visibility and much needed visibility that we don't have today with traditional tools, so it doesn't solve a bottleneck problem.

324
00:56:34.050 --> 00:56:40.860
Julio Ibarra: That has to be that that typically has a cost, and it just has to be addressed by the.

325
00:56:42.180 --> 00:56:44.220
Julio Ibarra: Projects or the the operators.

326
00:56:47.130 --> 00:56:47.670
Manish Parashar: Thank you.

327
00:56:49.650 --> 00:56:52.020
Manish Parashar: Our next question is from deep Medi.

328
00:56:53.070 --> 00:56:53.730
Manish Parashar: He says.

329
00:56:53.790 --> 00:56:54.090
Julio Ibarra: i'd be.

330
00:56:54.960 --> 00:56:55.770
Manish Parashar: Great talk.

331
00:56:57.150 --> 00:57:10.230
Manish Parashar: How are planning how are you planning to incorporate coordinate the need of event follow the horizon telescope an action to meet their data transfer in incoming incoming is.

332
00:57:12.240 --> 00:57:28.230
Julio Ibarra: Well, the the event horizon telescope is is, as I mentioned a social challenge because many of their sites just are not well connected for the kind of data movement that that they require.

333
00:57:29.760 --> 00:57:31.920
Julio Ibarra: There was a I think it was an.

334
00:57:33.360 --> 00:57:37.590
Julio Ibarra: That the science, the science digest nsf.

335
00:57:38.760 --> 00:57:48.870
Julio Ibarra: publish that the the the way that they are, they are getting the data to their their analysis centers essentially is via.

336
00:57:49.830 --> 00:57:57.990
Julio Ibarra: Via hard drives they drive up they take the hard drives and drive them down a mountain to whether then shipped to a an analysis Center.

337
00:57:58.920 --> 00:58:08.580
Julio Ibarra: And we have been meeting with them, working with them to adopt the research and education networks as their approach instead of using.

338
00:58:09.540 --> 00:58:23.670
Julio Ibarra: Physical disk drives from to move their data many of a number of the sites are properly connected, but quite a few others are not and, and so this is this is part of the work that we we we've.

339
00:58:25.050 --> 00:58:40.590
Julio Ibarra: we've taken on with the the sack committee to to show them what's at least available in South America, but they can leverage, but there is this is a, this is a global instrument involving many.

340
00:58:41.880 --> 00:58:44.490
Julio Ibarra: Many instruments, and so the.

341
00:58:45.570 --> 00:58:59.940
Julio Ibarra: Each of the iron CP is is working to better inform these these these astronomers about what's available to them and for them to to use these these resources.

342
00:59:02.970 --> 00:59:03.600
Manish Parashar: Thank you.

343
00:59:05.580 --> 00:59:05.970
Think.

344
00:59:08.460 --> 00:59:11.280
Manish Parashar: Jennifer thanks you for the shoutout T book.

345
00:59:13.440 --> 00:59:14.250
Julio Ibarra: you're welcome Jen.

346
00:59:15.360 --> 00:59:17.250
Manish Parashar: Other questions for Julio.

347
00:59:43.740 --> 00:59:44.130
Manish Parashar: well.

348
00:59:45.810 --> 00:59:47.520
Manish Parashar: we'll wait for another minute.

349
00:59:49.260 --> 00:59:52.800
Manish Parashar: see if there are any further questions.

350
00:59:54.600 --> 01:00:01.650
Manish Parashar: I will remind that for the nsf program officers there's a session later this afternoon.

351
01:00:04.200 --> 01:00:12.660
Manish Parashar: The boss live lecture officer it's between three and four, and so that's another opportunity to chat with Julio.

352
01:00:15.360 --> 01:00:18.630
Manish Parashar: Seeing no more questions Oh, there is one.

353
01:00:19.950 --> 01:00:29.250
Manish Parashar: Thank you Sylvia this is from Sylvia spangler do you see the network as being eXtensible to other disciplines within South America.

354
01:00:30.540 --> 01:00:33.750
Julio Ibarra: Yes, absolutely I I didn't.

355
01:00:35.940 --> 01:00:45.750
Julio Ibarra: I didn't cover other disciplines as well, but yes, we definitely are working to attract more disciplines.

356
01:00:47.190 --> 01:01:14.070
Julio Ibarra: Soviet, thank you for your question it's good to hear from you certainly the the bio sciences community is one of interest, as well as ecology and geosciences we as as a as a program the rnc P eyes communities week we share a lot of experiences and try to figure out how we can get these.

357
01:01:15.240 --> 01:01:16.080
Julio Ibarra: These different.

358
01:01:17.160 --> 01:01:18.210
Julio Ibarra: Communities of.

359
01:01:19.320 --> 01:01:34.380
Julio Ibarra: Researchers to engage with the research networks, so we have been leveraging leveraging some types of cyber infrastructure, for example, data transfer notes that that they can use to least.

360
01:01:35.580 --> 01:01:42.180
Julio Ibarra: copy their data to these nodes and be able to move them rapidly across the network to where they they are analyzed.

361
01:01:43.890 --> 01:01:45.660
Julio Ibarra: In some cases, this is a.

362
01:01:46.680 --> 01:01:56.640
Julio Ibarra: This has been very, very effective, the results have been very good in other cases it's just like I said earlier, it's a social engineering challenge to.

363
01:01:58.260 --> 01:02:11.790
Julio Ibarra: get some of the the researchers to adopt new methods and to use the research intuition networks so for sure we were certainly interested in every opportunity to engage with.

364
01:02:12.840 --> 01:02:18.750
Julio Ibarra: The different disciplines and have them benefit from the networks that are available to them.

365
01:02:23.220 --> 01:02:23.700
Thank you.

366
01:02:37.290 --> 01:02:37.770
Manish Parashar: well.

367
01:02:40.950 --> 01:02:43.740
Manish Parashar: We have a question from Patrick Smith.

368
01:02:46.290 --> 01:02:48.180
Manish Parashar: Sylvia says lovely talk.

369
01:02:48.870 --> 01:02:49.650
Julio Ibarra: Thank you, Sophia.

370
01:02:50.520 --> 01:02:57.780
Manish Parashar: and Patrick says fascinating presentation, thank you for your insights linear talk you mentioned Antarctica.

371
01:02:58.200 --> 01:03:09.570
Manish Parashar: Yes, are you interested in the recent announcement by sub tell Chile, for a subsea cable between Tierra del Fuego and King George island.

372
01:03:10.440 --> 01:03:16.890
Julio Ibarra: Yes, absolutely we we have been pursuing and following all the activities.

373
01:03:18.510 --> 01:03:40.050
Julio Ibarra: With nsf in the rnc program I I follow pat smith's presentations as much as possible, and I know he has one coming up, I think this this Saturday, as part of T PR E and we will we will certainly be be there to enjoy your talk Pat, but we are for sure interested in.

374
01:03:41.070 --> 01:03:53.010
Julio Ibarra: in Antarctica and the opportunities for Antarctica to benefit from the research networks, we have close to them, so the MIT network is is is a is a resource that we would like to.

375
01:03:54.630 --> 01:04:06.360
Julio Ibarra: make available for for the instruments in Antarctica I believe the the telemetry capability that we're instrument thing on emulate would be beneficial.

376
01:04:06.870 --> 01:04:21.330
Julio Ibarra: For a lot of that science and we would welcome the opportunity to show that so, but you know clearly this is a an opportunity for the entire irc program and nsf to to really.

377
01:04:22.800 --> 01:04:28.620
Julio Ibarra: be able to get the instruments connected to a high speed optical network.

378
01:04:29.760 --> 01:04:30.060
Julio Ibarra: and

379
01:04:31.440 --> 01:04:35.850
Julio Ibarra: As opposed to just using satellite communications as as they are today.

380
01:04:38.790 --> 01:04:46.920
Manish Parashar: Thank you Sylvia has another question, he asked should domains work with your science person at usc.

381
01:04:48.000 --> 01:04:53.070
Julio Ibarra: Yes, absolutely for sure hi amy hi amy Morgan is.

382
01:04:54.150 --> 01:05:07.110
Julio Ibarra: A excellent person to the contact and and inform them of pinterest for for me like thank you very much for for doing that.

383
01:05:09.750 --> 01:05:15.990
Manish Parashar: Q and we have a comment for for me the chose me she says no question just thanks for a great presentation.

384
01:05:16.380 --> 01:05:18.390
Julio Ibarra: Thank you so much, I really appreciate that.

385
01:05:30.450 --> 01:05:30.870
Okay.

386
01:05:32.250 --> 01:05:39.810
Manish Parashar: So with that, let us then close the session reminder again, there is a post lecture office are.

387
01:05:40.440 --> 01:05:56.520
Manish Parashar: between three and four, this afternoon, where we have another opportunity to talk to Julio can Julio, thank you for all your work impactful work and for this great presentation and with that i'm close the session, thank you.

388
01:05:56.670 --> 01:05:58.170
Julio Ibarra: Welcome Thank you so much.