idnits 2.17.1 draft-iab-covid19-workshop-01.txt: -(6): Line appears to be too long, but this could be caused by non-ascii characters in UTF-8 encoding Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- == There are 3 instances of lines with non-ascii characters in the document. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- ** The document seems to lack an IANA Considerations section. (See Section 2.2 of https://www.ietf.org/id-info/checklist for how to handle the case when there are no actions for IANA.) Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (22 February 2021) is 1149 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- No issues found here. Summary: 1 error (**), 0 flaws (~~), 2 warnings (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group J. Arkko 3 Internet-Draft Ericsson 4 Intended status: Informational S. Farrell 5 Expires: 26 August 2021 Trinity College Dublin 6 M. Kühlewind 7 Ericsson 8 C. Perkins 9 University of Glasgow 10 22 February 2021 12 Report from the IAB COVID-19 Network Impacts Workshop 2020 13 draft-iab-covid19-workshop-01 15 Abstract 17 The COVID-19 pandemic caused changes in Internet user behavior, 18 particularly during the introduction of the initial quarantine and 19 work-from-home arrangements. These behavior changes drove changes in 20 Internet traffic. 22 The Internet Architecture Board (IAB) held a workshop to discuss 23 network impacts of the pandemic on November 9-13, 2020. The workshop 24 was held to convene interested researchers, network operators, 25 network management experts, and Internet technologists to share their 26 experiences. The meeting was held online given the on-going travel 27 and contact restrictions at that time. 29 Discussion Venues 31 This note is to be removed before publishing as an RFC. 33 Source for this draft and an issue tracker can be found at 34 https://github.com/intarchboard/covid19-workshop. 36 Status of This Memo 38 This Internet-Draft is submitted in full conformance with the 39 provisions of BCP 78 and BCP 79. 41 Internet-Drafts are working documents of the Internet Engineering 42 Task Force (IETF). Note that other groups may also distribute 43 working documents as Internet-Drafts. The list of current Internet- 44 Drafts is at https://datatracker.ietf.org/drafts/current/. 46 Internet-Drafts are draft documents valid for a maximum of six months 47 and may be updated, replaced, or obsoleted by other documents at any 48 time. It is inappropriate to use Internet-Drafts as reference 49 material or to cite them other than as "work in progress." 51 This Internet-Draft will expire on 26 August 2021. 53 Copyright Notice 55 Copyright (c) 2021 IETF Trust and the persons identified as the 56 document authors. All rights reserved. 58 This document is subject to BCP 78 and the IETF Trust's Legal 59 Provisions Relating to IETF Documents (https://trustee.ietf.org/ 60 license-info) in effect on the date of publication of this document. 61 Please review these documents carefully, as they describe your rights 62 and restrictions with respect to this document. Code Components 63 extracted from this document must include Simplified BSD License text 64 as described in Section 4.e of the Trust Legal Provisions and are 65 provided without warranty as described in the Simplified BSD License. 67 Table of Contents 69 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 70 2. Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 71 3. Workshop Topics and Discussion . . . . . . . . . . . . . . . 4 72 3.1. Measurement-based Observations on Network Traffic 73 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . 5 74 3.1.1. Overall Traffic Growth . . . . . . . . . . . . . . . 5 75 3.1.2. Changes in Application Use . . . . . . . . . . . . . 6 76 3.1.3. Mobile Networks and Mobility . . . . . . . . . . . . 8 77 3.1.4. A Deeper Look at Interconnections . . . . . . . . . . 9 78 3.1.5. Cloud Platforms . . . . . . . . . . . . . . . . . . . 9 79 3.1.6. Last-Mile Congestion . . . . . . . . . . . . . . . . 10 80 3.1.7. User Behaviour . . . . . . . . . . . . . . . . . . . 10 81 3.2. Operational Practises and Architectural Considerations . 11 82 3.2.1. Digital Divide . . . . . . . . . . . . . . . . . . . 11 83 3.2.2. Applications . . . . . . . . . . . . . . . . . . . . 12 84 3.2.3. Observability . . . . . . . . . . . . . . . . . . . . 13 85 3.2.4. Security . . . . . . . . . . . . . . . . . . . . . . 14 86 3.2.5. Discussion . . . . . . . . . . . . . . . . . . . . . 14 87 3.3. Conclusions . . . . . . . . . . . . . . . . . . . . . . . 15 88 4. Feedback on Meeting Format . . . . . . . . . . . . . . . . . 16 89 5. Position Papers . . . . . . . . . . . . . . . . . . . . . . . 16 90 6. Workshop participants . . . . . . . . . . . . . . . . . . . . 17 91 7. Program Committee . . . . . . . . . . . . . . . . . . . . . . 19 92 8. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 20 93 9. Informative References . . . . . . . . . . . . . . . . . . . 20 94 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 23 96 1. Introduction 98 The Internet Architecture Board (IAB) held a workshop to discuss 99 network impacts of the COVID-19 pandemic, on November 9-13, 2020. 100 The workshop was held to convene interested researchers, network 101 operators, network management experts, and Internet technologists to 102 share their experiences. The meeting was held online given the on- 103 going travel and contact restrictions at that time. 105 COVID-19 has caused changes in user behavior, which in turn drove 106 change to Internet traffic. These changes appeared rather abruptly 107 and were significant, in particular during the introduction of the 108 initial quarantine and work-from-home arrangements. The changes 109 relate to traffic volumes, location of traffic, as well as the types 110 of traffic and applications used. 112 Announcement for the workshop was sent out in July 2020, requesting 113 interested parties to submit position papers for the workshop program 114 committee. A total of 15 position papers were received from 115 altogether 33 authors. The papers are listed in Section 5. In 116 addition, several other types of contributions and pointers to 117 existing work were provided. A number of position papers referred to 118 parallel work being published in measurement-related academic 119 conferences. 121 Invitations for the workshop were sent out based on the position 122 papers and other expressions of interest. On the workshop conference 123 calls were 45 participants, listed in Section 6. 125 The workshop was held over one week hosting three sessions covering 126 i) measurements and observations, ii) operational issue, and iii) 127 future consideration and conclusions. As these three sessions were 128 scheduled Monday, Wednesday, and Friday a positive side effect was 129 that the time in between could be used for mailing list discussion 130 and compilation of additional workshop material. 132 2. Scope 134 The COVID-19 pandemic has had a tremendous impact on people's lives 135 and the societies and economies around the globe. But it also had a 136 big impact on networking. With large numbers of people working from 137 home or otherwise depending on the network for their daily lives, 138 network traffic volume has surged. Internet service providers and 139 operators have reported a 20% traffic growth or more in a matter of 140 weeks. Traffic at Internet Exchange Points (IXPs) is similarly on 141 the rise. Most forms of network traffic have seen an increase, with 142 conversational multimedia traffic growing in some cases more than 143 200%. And user time spent on conferencing services has risen by an 144 order of magnitude on some conferencing platforms. 146 In general, the Internet has coped relatively well with this traffic 147 growth, albeit not without some issues. For instance, some outages, 148 video quality reduction, and other issues were reported. 149 Nevertheless, it is interesting to see how the technology, operators 150 and service providers have been able to respond to large and sudden 151 changes in traffic patterns. 153 Understanding what actually happened with Internet traffic is of 154 course interesting by its own right. How that impacted user 155 experience or the intended function of the services is equally 156 interesting. Measurements and reports of the traffic situation from 157 2020 are therefore valuable. But it would also be interesting to 158 understand what types of network management and capacity expansion 159 actions were taken in general. Anecdotal evidence points to Internet 160 and service providers tracking how their services are used, and in 161 many cases adjusting services to accommodate the new traffic 162 patterns, from dynamic allocation of compute resources to more 163 complex changes. 165 The impacts of this crisis are also a potential opportunity to 166 understand the impact of traffic shifts and growth more generally, or 167 to prepare for future situations -- crises or otherwise - that impact 168 networking. Or even allow us to adjust the technology to be even 169 better suited to respond to changes. 171 The scope of this workshop included: 173 * measurements about traffic changes, user experience, service 174 performance, and other relevant aspects 176 * discussion about the behind the scenes network management and 177 expansion activities 179 * experiences in the fields of general Internet connectivity, 180 conferencing, media/entertainment, and Internet infrastructure 182 * lessons learned for preparedness and operations 184 * lessons learned for Internet technology and architecture 186 3. Workshop Topics and Discussion 187 3.1. Measurement-based Observations on Network Traffic Dynamics 189 The workshop started with a focus on measurements. A large portion 190 of the submitted papers presented and discussed measurement data and 191 these submissions provided a good basis get a better understanding of 192 the situation, covering different angles and aspects of network 193 traffic and kind of networks. 195 Changes in Internet traffic due to the COVID-19 pandemic affected 196 different networks in various ways. Yet all networks observed some 197 form of change, be it a reduction in traffic, an increase in traffic, 198 a change in working days and weekend days patterns, or a change in 199 traffic classes. Traffic volume, directionality ratios, and its 200 source and destination are radically different than from before 201 COVID-19. 203 At a high level, while traffic from home networks increased 204 significantly, the traffic in mobile networks decreased as a result 205 of reduced population mobility. The observed behavior in mobile 206 networks is antagonistic, yet complementary, to the one observed in 207 residential ISPs. In residential networks there was a strong 208 increase in video conferencing and remote learning application 209 traffic due to the shift for working and learning at home. With that 210 shift, the typical diurnal usage patterns in network traffic also 211 changed, with peak times occuring earlier in the day and lasting 212 longer over the day - reflecting the start of the work or school day 213 from home. This behavior is antagonistic, yet complementary, to the 214 one observed in residential ISPs. 216 While diurnal congestion at interconnect point as well in certain 217 last mile network was reported, mainly in March, no persitent 218 congestion was observed. Further, a downward trends in download 219 throughput to certain cloud regions was measured, which can probably 220 explained with the increase use of cloud services. This gives 221 another indication that the scalng of shared resources in the 222 Internet is working reasonably well enough to handle even larger 223 changes in traffic as experience during the first nearly global 224 lockdown of the COVID-19 pandemic. 226 3.1.1. Overall Traffic Growth 228 The global pandemic has significantly accelerated the growth of data 229 traffic worldwide. Based on the measurement data of one ISP, three 230 IXPs, a metropolitan educational network, and a mobile operator, it 231 was observed at the beginning of the workshop [Feldmann2020] that 232 overall the network was able to handle the situation well, despite a 233 significant and sudden increase in traffic growth rate in March and 234 April. That is, after the lockdown was implemented in March, a 235 traffic increase of 15-20% at the ISP as well as the three IXPs was 236 observed. That represents the traffic growth expected in a typical 237 year which now took place in the matter of a few weeks only---a 238 substantial increase. At DE-CIX Frankfurt, the world's largest 239 Internet Exchange Point in terms of data throughput, the year 2020 240 has seen the largest increase in peak traffic within a single year 241 since the IXP was founded in 1995. Additionally, mobile traffic has 242 slightly receded. In access networks, the growth rate of upstream 243 traffic also exceeded the growth in downstream traffic, reflecting 244 increased adoption and use of video conferencing and other remote 245 work and school applications. 247 Most traffic increases happened during non-traditional peak hours: 248 Before the first COVID-19 lockdowns, the main time of use was in the 249 evening hours during the week, whereas since March it has been spread 250 more equally across the day. That is, the increase in usage has 251 mainly occurred outside the previous peak usage times (e.g. during 252 the day while working from home). This means that, for the first 253 time, network utilization on weekdays resembled that on weekends. 254 The effects of the increased traffic volume could easily be absorbed: 255 either by using existing reserve capacity, or by quickly switching 256 additional bandwidth. This is one reason why the Internet was able 257 to cope well with the pandemic during the first lockdown period. 259 Some of the lockdowns were lifted or relaxed around May 2020. As 260 people were allowed to perform some of their daily habits outside of 261 their home again, as expected, there was a decrease of the traffic 262 observed at the IXPs and the ISP; instead mobile traffic began to 263 grow again. 265 3.1.2. Changes in Application Use 267 The composition of data traffic has changed since the beginning of 268 the pandemic: the use of videoconferencing services and virtual 269 private networks (VPNs) for access to company resources from the home 270 environment has risen sharply. In ISP and IXP network it was 271 observed [Feldmann2020] that traffic associated with web 272 conferencing, video, and gaming increased largely in March 2020 as a 273 result of the increasing user demand for solutions like Zoom or 274 Microsoft Teams. For example, the relative traffic share of many 275 "essential" applications like VPN and conferencing tools increased by 276 more than 200%. 278 Also, as people spent more hours at home, they tended to watch videos 279 or play games, thus increasing entertainment traffic demands. At the 280 same time, the traffic share for other traffic classes decreased 281 substantially, e.g., traffic related to education, social media, and 282 ---for some periods---CDNs. In April and June, web conferencing 283 traffic was still high compared to the pre-pandemic scenario, while a 284 slight decrease in CDN and social media traffic was observed. During 285 these months many people were still working from home, but 286 restrictions had been lifted or relaxed, which likely led to an 287 increase in in-person social activities and a decrease in online 288 ones. 290 3.1.2.1. Example Campus Networks 292 Changes in traffic have been observed at University campus networks 293 as well, especially due to the necessary adoption of remote teaching. 294 The Politecnico di Torino University (Italy) deployed its in-house 295 solution for remote teaching, which caused the outgoing traffic to 296 grow by 2.5 times, driven by more than 600 daily online classes. 297 Incoming traffic, instead, decreased by a factor of 10 due to the 298 cessation of any in-person activity. Based on their measurements, 299 this change in traffic and network usage did however not lead to 300 noticeable performance impairments, nor have significantly poor 301 performance been observed for students in remote regions of Italy. 302 Outgoing traffic also increased due to other remote working 303 solutions, such as collaboration platforms, VPNs, and remote 304 desktops. 306 Similar changes were observed by measuring REDIMadrid [Feldmann2020], 307 a European educational and research network, which connects 16 308 independent universities and research centers in the metropolitan 309 region of Madrid. A drop of up to 55% in traffic volume on working 310 days during the pandemic was observed. Similar to findings for ISP/ 311 IXP networks, it was observed that working days and weekend days are 312 becoming more similar in terms of total traffic. The hourly traffic 313 patterns reveal a traffic increase between 9 pm and 7 am. This could 314 be due to users working more frequently at unusual times, but also 315 potentially caused by overseas students (mainly from Latin America 316 and East Asia as suggested by the AS numbers from which these 317 connections came from) who accessed university network resources from 318 their home countries. 320 Given the fact that the users of the academic network (e.g., students 321 and research staff) had to leave the campus as a response to lockdown 322 measures, also the traffic in and out (i.e., ingress and egress) 323 ratio changed drastically. Prior to the lockdown, the incoming 324 traffic was much larger then the outgoing traffic. This changed to a 325 more balanced ratio. This change of traffic asymmetry can be 326 explained by the nature of remote work. On the one end, users 327 connected to the network services mainly to access resources, hence 328 the increase in outgoing traffic. On the other end, all external 329 (i.e., Internet-based) resources requested during work were no longer 330 accessed from the educational network but from the users' homes. 332 3.1.3. Mobile Networks and Mobility 334 Mobile network data usage appeared to decline following the 335 imposition of localized lockdown measures, as these reduced typical 336 levels of mobility and roaming. 338 [Lutu2020] measured the cellular network of O2 UK to evaluate how the 339 changes in people's mobility impacted traffic patterns. By analyzing 340 cellular network signalling information regarding users' device 341 mobility activity, they observed a decrease of 50% in mobility 342 (according to different mobility metrics) in the UK during the 343 lockdown period. As they found no correlation between this reduction 344 in mobility and the number of confirmed COVID-19 cases, only the 345 enforced government order was effective in significantly reducing 346 mobility and this reduction was more significant in densely populated 347 urban areas than in rural areas. For London, specifically, it could 348 be observed from the mobile network data that approximately 10% of 349 the residents temporarily relocated during the lockdown. 351 These mobility changes had immediate implications in traffic patterns 352 of the cellular network. The downlink data traffic volume aggregated 353 for all bearers (including conversational voice) decreased for all UK 354 by up to 25% during the lockdown period. This correlates with the 355 reduction in mobility that was observed country-wide, which likely 356 resulted in people relying more on broadband residential Internet 357 access to run download intensive applications such as video 358 streaming. The observed decrease in the radio cell load, with a 359 reduction of approximately 15% across the UK after the stay-at-home 360 order, further corroborates the drop in cellular connectivity usage. 362 The total uplink data traffic volume, on the other hand, experienced 363 little changes (between -7% and +1,5%) during lockdown. This was 364 mainly due to the increase of 4G voice traffic (i.e., VoLTE) across 365 the UK that peaked at 150% after the lockdown compared to the 366 national medial value before the pandemic, thus compensating for the 367 decrease in data traffic in the uplink. 369 Finally, it was also observed that mobility changes have a different 370 impact on network usage in geodemographic area clusters. In densely 371 populated urban areas, a significantly higher decrease of mobile 372 network usage (i.e., downlink and uplink traffic volumes, radio load 373 and active users) was observed than in rural areas. In the case of 374 London, this was likely due to geodemographics of the central 375 districts, which include many seasonal residents (e.g., tourists), 376 business and commercial areas. 378 3.1.4. A Deeper Look at Interconnections 380 Traffic at points of network interconnection noticeably increased, 381 but most operators reacted quickly by rapidly adding additional 382 capacity [Feldmann2020]. The amount of increases varied, with some 383 networks that hosted popular applications such as video conferencing 384 experiencing traffic growth of several hundred to several thousand 385 percent. At the IXP-level, it was observed that port utilization 386 increased. This phenomenon is mostly explained by a higher traffic 387 demand from residential users. 389 Measurements of interconnection links at major US ISPs by CAIDA and 390 MIT found some evidence of diurnal congestion around the March 2020 391 timeframe [Clark2020], but most of this congestion disappeared in a 392 few weeks, which suggests that operators indeed took steps to add 393 capacity or otherwise mitigate the congestion. 395 3.1.5. Cloud Platforms 397 Cloud infrastructure played a key role in supporting bandwidth- 398 intensive video conferencing and remote learning tools to practice 399 social distancing during the COVID-19 pandemic. Network congestion 400 between cloud platforms and access networks could impact the quality 401 of experience of these cloud-based applications. CAIDA leveraged 402 web-based speed test servers to perform download and upload 403 throughput measurements from virtual machines in public cloud 404 platforms to various access ISPs in the United States [Mok2020]. 406 The key findings included: 408 * Persistent congestion events were not widely observed between 409 cloud platforms and these networks, particular for large-scale 410 ISPs, but we could observe large diurnal download throughput 411 variations in peak hours from some locations to the cloud. 413 * There was evidence of persistent congestion in the egress 414 direction to regional ISPs serving suburban areas in the U.S. 415 Their users could have suffered from poor video streaming or file 416 download performance from the cloud. 418 * The macroscopic analysis over 3 months (June-August, 2020) 419 revealed downward trends in download throughput from ISPs and 420 educational networks to certain cloud regions. We believe that 421 increased use of the cloud in the pandemic could be one of the 422 factors that contributed to the decreased performance. 424 3.1.6. Last-Mile Congestion 426 The last mile is the centerpiece of broadband connectivity, where 427 poor last-mile performance generally translates to poor quality of 428 experience. In a recent IMC'20 research paper Fontugne et al. 429 investigated last-mile latency using traceroute data from RIPE Atlas 430 probes located in 646 ASes and looked for recurrent performance 431 degradation [Fontugne2020-1]. They found that in normal times Atlas 432 probes in only 10% ASes experience persistent last-mile congestion, 433 but they recorded 55% more congested ASes during the COVID-19 434 outbreak. This deterioration caused by stay-at-home measures is 435 particularly marked in networks with a very large number of users and 436 certain parts of the world. They found Japan to be the most impacted 437 country in their study looking specifically at NTT OCN, but noting 438 similar observations for several Japanese networks, including IIJ 439 (AS2497). 441 From mid-2020 onwards, they however observed better performance than 442 before the pandemic. In Japan, this was partly due to the 443 deployments originally planned for accommodating the Tokyo Olympics, 444 and more generally, it reflects the efforts of network operators to 445 cope with these exceptional circumstances. The pandemic has 446 demonstrated that its adaptive design and proficient community can 447 keep the Internet operational during such unprecedented events. 448 Also, from the numerous research and operational reports recently 449 published, the pandemic is apparently shaping a more resilient 450 Internet, as Nietzsche wrote, "What does not kill me makes me 451 stronger". 453 3.1.7. User Behaviour 455 The type of traffic needed by the users also changed in 2020. 456 Upstream traffic increased due the use of video conferences, remote 457 schooling, and similar applications. The NCTA and Comcast reported 458 that while downstream traffic grew 20%, upstream traffic grew as much 459 as 30% to 37% [NCTA2020] [Comcast2020]. Vodafone reported that 460 upstream traffic grew 100% in some markets [Vodafone2020]. 462 Ericsson's Consumer Lab surveyed users for their usage and 463 experiences during the crisis. Some of the key findings in 464 [ConsumerlabReport2020] were: 466 * 9 in 10 users increased Internet activities, and time spent 467 connected increased. In addition, 1 in 5 started new online 468 activities, many in the older generation felt that they were 469 helped by video calling, parents felt that their children's 470 education was helped, and so on. 472 * Network performance was, in general, found satisfactory. 6 in 10 473 were very satisfied with fixed broadband, and 3 in 4 felt that 474 mobile broadband was the same or better compared to before the 475 crisis. Consumers valued resilience and quality of service as the 476 most important task for network operators. 478 * Smartphone application usage changed, with fastest growth in apps 479 related to COVID-19 tracking and information, remote working, 480 e-learning, wellness, education, remote health consultation, and 481 social shared experience applications. Biggest decreases were in 482 travel and booking, ride hailing, location, and parking 483 applications. 485 Some of the behaviours are likely permanent changes 486 [ConsumerlabReport2020]. The adoption of video calls and other new 487 services by many consumers, such as the older generation, is likely 488 going to have a long-lasting effect. Surveys in various 489 organizations point to a likely long-term increase in the number of 490 people interested in remote work [WorkplaceAnalytics2020] 491 [McKinsey2020]. 493 3.2. Operational Practises and Architectural Considerations 495 The second and third day of the workshop were held based on more open 496 discussions focussed on operational issues and the architectural 497 issues arising or other conclusions that could be reached. 499 3.2.1. Digital Divide 501 Measurements from Fastly confirmed that Internet traffic volume, in 502 multiple countries, rose rapidly at the same time as COVID cases 503 increased and lockdown policies came into effect. Download speeds 504 also decreased, but in a much less dramatic fashion than overall 505 bandwidth usage increased. School closures led to a dramatic 506 increase in traffic volume in many regions, and other public policy 507 announcements triggered large traffic shifts. This suggests that 508 governments might usefully coordinate with operators to allow time 509 for pre-emptive operational changes, in some cases. 511 Measurements from the US showed that download rates correlate with 512 income levels. However, download rates in the lowest income zip 513 codes increased as the pandemic progressed, closing the divide with 514 higher income areas. One possible reason for this in the data is 515 decisions by some ISPs, such as Comcast and Cox, that increased 516 speeds for users on lower-cost certain plans and in certain areas. 517 This suggests that network capacity was available, and that the 518 correlation between income and download rates was not necessarily due 519 to differences in the deployed infrastructure in different regions, 520 although it was noted that certainly access link technologies provide 521 more flexibility than others in this regard. 523 3.2.2. Applications 525 The web conferencing systems (e.g., Microsoft Teams, Zoom, Webex) saw 526 incredible growth, with overnight traffic increases of 15-20% in 527 response to public policy changes, such as lockdowns. This required 528 significant and rapid changes in infrastructure provisioning. 530 Major video providers (YouTube, etc.) reduced bandwidth by 25% in 531 some regions. It was suggested that this had a huge impact on 532 quality of videoconferencing systems until networks could scale to 533 handle full bit-rate, but other operators of some other services saw 534 limited impact. 536 Updates to popular games has a significant impact on network load. 537 Some discussions were reported between ISPs, CDNs, and the gaming 538 industry on possibly coordinating various high-bandwidth update 539 events, similar to what was done for entertainment/video download 540 speeds. There was an apparently difficult interplay between bulk 541 download and interactive real-time applications, potentially due to 542 buffer bloat and queuing delays. 544 It was noted that operators have experience of rapid growth of 545 Internet traffic. New applications with exponential growth are not 546 that unusual in the network, and the traffic spike due to the 547 lockdown was not that unprecedented for many. Many operators have 548 tools and mechanisms to deal with this. Ensuring that knowledge if 549 shared is a challenge. 551 Following these observations traffic prioritisation was discussed, 552 starting from DSCP marking, basically wondering if a minimal priority 553 marking scheme would have helped during the pandemic, e.g. by 554 allowing marking of less-than-best-effort traffic. That discussion 555 quickly devolved into a more general QoS and observability 556 discussion, and as such also touching on the effects of increased 557 encryption. The group was not, unsurprisingly, able to resolve the 558 different perspectives and interests involved in that, but the 559 discussion demonstrated that progress is made (and being less 560 heated). 562 3.2.3. Observability 564 It is clear that there is a contrast in experience. Many operators 565 reported few problems, in terms of metrics such as measured download 566 bandwidth, while video conferencing applications experienced 567 significant usability problems running on those networks. The 568 interaction between application providers and network providers 569 worked very smoothly to resolve these issues, supported by strong 570 personal contacts and relationships. But it seems clear that the 571 metrics used by many operators to understand their network 572 performance don't fully capture the impact on certain applications, 573 and there is an observability gap. Do we need more tools to figure 574 out the various impacts on user experience? 576 These types of applications use surprising amounts of Forward Error 577 Correction (FEC). Applications hide lots of loss to ensure a good 578 user experience. This makes it harder to observe problems. The 579 network can be behaving poorly, but experience can be good enough. 580 Resiliency measures can improve the user experience but hide severe 581 problems. There may be a missing feedback loop between application 582 developers and operators. 584 It's clear that it's difficult for application providers and 585 operators to isolate problems. Is a problem due to the local WiFi, 586 the access network, cloud network, etc.? Metrics from access points 587 would help, but in general lack of observability into the network as 588 a whole is a real concern when it comes to debugging performance 589 issues. 591 Further, it's clear that it can be difficult to route problem reports 592 to the person who can fix them, across multiple networks in the 593 Internet. COVID-enhanced cooperation made it easier to debug 594 problems; lines of communication are important. 596 3.2.4. Security 598 It was noted that there is a shift to home working generally, and in 599 the way people use the network, with IT departments rolling out new 600 equipment quickly and using technologies like VPNs for the first 601 time. 603 There are reports of a strong rise in phishing, fraud, and scams 604 related to COVID [Kirsty2020]. It's unclear if there was an increase 605 in fraud overall but there was certainly a shift in activity. New 606 types of attacks, for example on vaccine research labs, health 607 services, and home working were reported. 609 It's unclear how to effectively detect and counter these attacks at 610 scale. Approaches such as crowd-sourced flagging of suspicious 611 emails help, and others noted that observing DNS to detect malicious 612 use is popular. The use of DNS over HTTPS offers privacy benefits 613 but is also observed to bypasses some protective measures. 615 It was also noted that when everyone moves to performing their job 616 online, lack of understanding of security becomes a bigger issue. 617 Who is ultimately responsible for security? Do we expect every user 618 of the Internet to take password training? Or is there a fundamental 619 problem here with a technical solution. Technologies such as Zoom 620 are not new: many people have used them for years; nobody attacked it 621 until it was the front line. What's the next vulnerable service? 623 Overall, it may be that the pandemic caused fewer security changes, 624 with many people suddendly working from home, than one might have 625 guessed prior to the pandemic. However, existing security problems 626 and challenges may have become even more obvious and acute with an 627 increased use of Internet-based services. 629 3.2.5. Discussion 631 There is a concern that we're missing observability for the network 632 as a whole. Each application provider and operator has their own 633 little lens. No-one has the big-picture view of the network. 635 How much of a safety margin do we need? Some of the resiliency comes 636 from us not running the network too close to its limit. This allows 637 traffic to shift, and gives headroom for the network to cope. The 638 best effort nature of the network may help here. Techniques to run 639 the network closer to its limits improve performance in the usual 640 case, but highly optimised networks may be less robust. 642 Finally, it was observed that we get what we measure. There may be 643 an argument for operators to shift their measurement focus perhaps 644 away from pure capacity, to rather measure QoE or resilience. The 645 Internet is a critical infrastructure, and people are realising that 646 now. We should use this as a wake-up-call to improve resilience, 647 both in protocol design and operational practise, not necessarily to 648 optimise for absolute performance or quality of experience. 650 3.3. Conclusions 652 There is a wealth of data about the performance of the Internet 653 during the crisis. The main conclusion from the various measurements 654 is that fairly large shifts occurred. And those shifts were not 655 merely about changing one application for another, they actually 656 impacted traffic flows and directions, and caused in many cases a 657 significant traffic increase. Early reports also seem to indicate 658 that the shifts have gone relatively smoothly from the point of view 659 of overall consumer experience. 661 An important but not so visible factor that led to this was that many 662 people and organizations where highly motivated to ensure good 663 experience. A lot of collaboration happened in the background, 664 problems were corrected, many providers significantly increased their 665 capacity, and so on. 667 In general, the Internet also seems well suited for adapting to new 668 situations, at least within some bounds. The Internet is designed 669 for any application and situation, rather than optimized for today's 670 particular traffic. This makes it possible to use it for many 671 applications, in many deployment situations, and make changes as 672 needed. The generality is present in many parts of the overall 673 system, from basic Internet technology to browsers, from name servers 674 to content delivery networks and cloud platforms. When needs change, 675 what is needed is often merely different services, perhaps some re- 676 allocation of resources, but not fundamental technology or hardware 677 changes. 679 On the other hand, this is not to say that no improvements are 680 needed: 682 * Better understanding of the health of the Internet: Going forward, 683 the critical nature that the Internet plays in our lives means 684 that the health of the Internet needs to receive significant 685 attention. Understanding how well networks work is not just a 686 technical matter, it is also of crucial importance to the people 687 and economy of the societies using it. Projects and research that 688 monitor Internet and services performance in a broad scale and 689 across different networks are therefore important. 691 * The pandemic has shown how the effects of the digital divide can 692 be amplified during a crisis. More attention is needed to ensure 693 that broadband is available to all, and that Internet services 694 equally serve different groups. 696 * We need to continue to work on all the other improvements that are 697 seen as necessary anyway, such as further improvements in 698 security, ability for networks and applications to collaborate 699 better, etc. 701 * Informal collaboration between different parties needs to continue 702 and be strengthened. 704 4. Feedback on Meeting Format 706 While there are frequently virtual participants in IAB workshops, the 707 IAB had no experience running workshops entirely virtually. 709 Feedback on this event format was largely positive, however. It was 710 particularly useful that as the three sessions were scheduled Monday, 711 Wednesday, and Friday, the time in between could be used for mailing 712 list discussion and compilation of additional workshop material. The 713 positive feedback was likely at least partly due to the fact that 714 many of the workshop participants knew one another from previous 715 face-to-face events (primarily IETF meetings). 717 The process for sending invitations to the workshop should be 718 improved for next time, however, as a few invitations were initially 719 lost, and in a virtual meeting it may be more reasonable to invite 720 not just one person but all co-authors of a paper, for instance. At 721 least for this workshop, we did not appear to suffer from too many 722 participants, and in many cases there may be some days when a 723 particular participant may not be able to attend a session. 725 5. Position Papers 727 The following position papers were received, in alphabetical order: 729 * Afxanasyev, A., Wang, L., Yeh, E., Zhang, B., and Zhang, L.: 730 Identifying the Disease from the Symptoms: Lessons for Networking 731 in the COVID-19 Era [Afxanasyev2020] 733 * Arkko, Jari: Observations on Network User Behaviour During 734 COVID-19 [Arkko2020] 736 * Bronzino, F., Culley, E., Feamster, N. Liu. S., Livingood. J., 737 and Schmitt, P.: IAB COVID-19 Workshop: Interconnection Changes in 738 the United States [Bronzino2020] 740 * Campling, Andrew and Lazanski, Dominique: Will the Internet Still 741 Be Resilient During the Next Black Swan Event? [Campling2020] 743 * Cho, Kenjiro: On the COVID-19 Impact to broadband traffic in Japan 744 [Cho2020] 746 * Clark, D.: Measurement of congestion on ISP interconnection links 747 [Clark2020] 749 * Favale, T., Soro, F., Trevisan, M., Drago, I., and Mellia, M.: 750 Campus traffic and e-Learning during COVID-19 pandemic 751 [Favale2020] 753 * Feldmann, A., Gasser, O., Lichtblau, F., Pujol, E., Poese, I., 754 Dietzel, C., Wagner, D., Wichtlhuber, M., Tapiador, J., Vallina- 755 Rodriguez, N., Hohlfeld, O., and Smaragdakis, G.: A view of 756 Internet Traffic Shifts at ISP and IXPs during the COVID-19 757 Pandemic [Feldmann2020] 759 * Fontugne, R., Shah, A., and Cho, K.: The Impact of COVID-19 on 760 Last-mile Latency [Fontugne2020] 762 * Gillmor, D.: Vaccines, Privacy, Software Updates, and Trust 763 [Gillmor2020] 765 * Gu, Y. and Li, Z. Covid 19 Impact on China ISP's Network Traffic 766 Pattern and Solution Discussion [Gu2020] 768 * Jennings, C. and Kozanian, P.: WebEx Scaling During Covid 769 [Jennings2020] 771 * Lutu, A., Perino, D., Bagnulo, M., Frias-Martinez, E., and 772 Khangosstar, J.: A Characterization of the COVID-19 Pandemic 773 Impact on a Mobile Network Operator Traffic [Lutu2020] 775 * Mok, Ricky, and claffy, kc: Measuring the impact of COVID-19 on 776 cloud network performance [Mok2020] 778 * Kirsty P: IAB COVID-19 Network Impacts [Kirsty2020] 780 6. Workshop participants 782 The following is an alphabetical list of participants in the 783 workshop. 785 * Jari Arkko (Ericsson/IAB) 787 * Ben Campbell (Independent/IAB) 788 * Andrew Campling (419 Consulting) 790 * Kenjiro Cho (IIJ) 792 * kc Claffy (CAIDA) 794 * David Clark (MIT CSAIL) 796 * Chris Dietzel (DE-CIX) 798 * Idilio Drago (University of Turin) 800 * Stephen Farrell (Trinity College Dublin/IAB) 802 * Nick Feamster (University of Chicago) 804 * Anja Feldmann (Max Planck Institute for Informatics) 806 * Romain Fontugne (IIJ Research Lab) 808 * Oliver Gasser (Max Planck Institute for Informatics) 810 * Daniel Kahn Gillmor (ACLU) 812 * Yunan Gu (Huawei) 814 * Oliver Hohlfeld (Brandenburg University of Technology, BTU) 816 * Jana Iyengar (Fastly) 818 * Cullen Jennings (Cisco/IAB) 820 * Mirja Kuhlewind (Ericsson/IAB) 822 * Franziska Lichtblau (Max Planck Institute for Informatics) 824 * Dominique Lazanski 826 * Zhenbin Li (Huawei/IAB) 828 * Jason Livingood (Comcast) 830 * Andra Lutu (Telefonica Research) 832 * Vesna Manojlovic (RIPE NCC) 834 * R Martin EC (?) 835 * Matt Matthis (Google) 837 * Larry Masinter (Retired) 839 * Jared Mauch (Akamai/IAB) 841 * Deep Medhi (NSF) 843 * Marco Mellia (Politecnico di Torino) 845 * Ricky Mok (CAIDA) 847 * Karen O'Donoghue (Internet Society) 849 * Kirsty P (NCSC) 851 * Diego Perino (Telefonica Research) 853 * Colin Perkins (University of Glasgow/IRTF/IAB) 855 * Enric Pujol (Benocs) 857 * Anant Shah (Verizon Media Platform) 859 * Francesca Soro (Politecnico di Torino) 861 * Brian Trammell (Google) 863 * Gergios Tselentis (European Commission) 865 * Martino Trevisan 867 * Lan Wang (University of Memphis) 869 * Rob Wilton (Cisco) 871 * Jiankang Yao (CNNIC) 873 * Lixia Zhang (UCLA) 875 7. Program Committee 877 The workshop Program Committee members were Jari Arkko, Stephen 878 Farrell, Cullen Jennings, Colin Perkins, Ben Campbell, and Mirja 879 Kuehlewind. 881 8. Acknowledgments 883 The authors would like to thank the workshop participants, the 884 members of the IAB, the program committee, the participants in the 885 architecture discussion list for interesting discussions, and Cindy 886 Morgan for the practical arrangements. 888 Further special thanks to those participants who also contributed to 889 this report: Romain Fontugne provided text based on his blog post at 890 https://eng-blog.iij.ad.jp/archives/7722; Ricky Mok for text on cloud 891 platform; Martino Trevisan for text on campus networks; David Clark 892 on congestion measurements at interconnects; Oliver Hohlfeld for the 893 text on traffic growth, changes in traffic shifts, campus networks, 894 and interconnections; Andra Lutu on mobile networks; And thanks to 895 Jason Livingood for his review and additions. 897 9. Informative References 899 [Afxanasyev2020] 900 Afxanasyev, A., Wang, L., Yeh, E., Zhang, B., and L. 901 Zhang, "Identifying the Disease from the Symptoms: Lessons 902 for Networking in the COVID-19 Era", Position paper in the 903 2020 IAB COVID-19 Network Impacts workshop. , October 904 2020. 906 [Arkko2020] 907 Arkko, J., "Observations on Network User Behaviour During 908 COVID-19", Position paper in the 2020 IAB COVID-19 Network 909 Impacts workshop. , October 2020. 911 [Bronzino2020] 912 Bronzino, F., Culley, E., Feamster, N., Liu, S., 913 Livingood, J., and P. Schmitt, "IAB COVID-19 Workshop: 914 Interconnection Changes in the United States", Position 915 paper in the 2020 IAB COVID-19 Network Impacts workshop. , 916 October 2020. 918 [Campling2020] 919 Campling, A. and D. Lazanski, "Will the Internet Still Be 920 Resilient During the Next Black Swan Event?", Position 921 paper in the 2020 IAB COVID-19 Network Impacts workshop. , 922 October 2020. 924 [Cho2020] Cho, K., "On the COVID-19 Impact to broadband traffic in 925 Japan", Position paper in the 2020 IAB COVID-19 Network 926 Impacts workshop. , October 2020. 928 [Clark2020] 929 Clark, D., "Measurement of congestion on ISP 930 interconnection links", Position paper in the 2020 IAB 931 COVID-19 Network Impacts workshop. , October 2020. 933 [Comcast2020] 934 Comcast, ., "COVID-19 Network Update", 935 https://corporate.comcast.com/covid-19/network/may-20-2020 936 , May 2020. 938 [ConsumerlabReport2020] 939 Ericsson Consumer & IndustryLab, ., "Keeping consumers 940 connected in a COVID-19 context", 941 https://www.ericsson.com/en/reports-and- 942 papers/consumerlab/reports/keeping-consumers-connected- 943 during-the-covid-19-crisis , June 2020. 945 [Favale2020] 946 Favale, T., Soro, F., Trevisan, M., Drago, I., and M. 947 Mellia, "Campus traffic and e-Learning during COVID-19 948 pandemic", Position paper in the 2020 IAB COVID-19 Network 949 Impacts workshop. , October 2020. 951 [Feldmann2020] 952 Feldmann, A., Gasser, O., Lichtblau, F., Pujol, E., Poese, 953 I., Dietzel, C., Wagner, D., Wichtlhuber, M., Tapiador, 954 J., N Vallina-Rodriguez, ., Hohlfeld, O., and G. 955 Smaragdakis, "A view of Internet Traffic Shifts at ISP and 956 IXPs during the COVID-19 Pandemic", Position paper in the 957 2020 IAB COVID-19 Network Impacts workshop. , October 958 2020. 960 [Fontugne2020] 961 Fontugne, R., Shah, A., and K. Cho, "The Impact of 962 COVID-19 on Last-mile Latency", Position paper in the 2020 963 IAB COVID-19 Network Impacts workshop. , October 2020. 965 [Fontugne2020-1] 966 Fontugne, R., Shah, A., and K. Cho, "Persistent Last-mile 967 Congestion: Not so Uncommon", Proceedings of the ACM 968 Internet Measurement Conference (IMC '20) , October 2020. 970 [Gillmor2020] 971 Gillmor, D., "Vaccines, Privacy, Software Updates, and 972 Trust", Position paper in the 2020 IAB COVID-19 Network 973 Impacts workshop. , October 2020. 975 [Gu2020] Gu, Y. and Z. Li, "Covid 19 Impact on China ISP's Network 976 Traffic Pattern and Solution Discussion", Position paper 977 in the 2020 IAB COVID-19 Network Impacts workshop. , 978 October 2020. 980 [Jennings2020] 981 Jennings, C. and P. Kozanian, "WebEx Scaling During 982 Covid", Position paper in the 2020 IAB COVID-19 Network 983 Impacts workshop. , October 2020. 985 [Kirsty2020] 986 Kirsty P, ., "IAB COVID-19 Network Impacts", Position 987 paper in the 2020 IAB COVID-19 Network Impacts workshop. , 988 October 2020. 990 [Lutu2020] Lutu, A., Perino, D., Bagnulo, M., Frias-Martinez, E., and 991 J. Khangosstar, "A Characterization of the COVID-19 992 Pandemic Impact on a Mobile Network Operator Traffic", 993 Position paper in the 2020 IAB COVID-19 Network Impacts 994 workshop. , October 2020. 996 [McKinsey2020] 997 Boland, B., De Smet, A., Palter, R., and A. Sanghvi, 998 "Reimagining the office and work life after COVID-19", htt 999 ps://www.mckinsey.com/~/media/McKinsey/Business%20Function 1000 s/Organization/Our%20Insights/Reimagining%20the%20office%2 1001 0and%20work%20life%20after%20COVID%2019/Reimagining-the- 1002 office-and-work-life-after-COVID-19-final.pdf , June 2020. 1004 [Mok2020] Mok, R. and . kc claffy, "Measuring the impact of COVID-19 1005 on cloud network performance", Position paper in the 2020 1006 IAB COVID-19 Network Impacts workshop. , October 2020. 1008 [NCTA2020] NCTA, ., "COVID-19: How Cable's Internet Networks Are 1009 Performing: Metrics, Trends & Observations", 1010 https://www.ncta.com/COVIDdashboard , 2020. 1012 [Vodafone2020] 1013 Vodafone, ., "An update on Vodafone's networks", 1014 https://www.vodafone.com/covid19/news/update-on-vodafone- 1015 networks , April 2020. 1017 [WorkplaceAnalytics2020] 1018 Lister, K., "Work-At-Home After Covid-19—Our Forecast", 1019 https://globalworkplaceanalytics.com/work-at-home-after- 1020 covid-19-our-forecast , 2020. 1022 Authors' Addresses 1024 Jari Arkko 1025 Ericsson 1027 Email: jari.arkko@ericsson.com 1029 Stephen Farrell 1030 Trinity College Dublin 1032 Email: stephen.farrell@cs.tcd.ie 1034 Mirja Kühlewind 1035 Ericsson 1037 Email: mirja.kuehlewind@ericsson.com 1039 Colin Perkins 1040 University of Glasgow 1042 Email: csp@csperkins.org