Source-linked AI summary
Campus Traffic and e-Learning during COVID-19 Pandemic
Thomas Favale, Francesca Soro, Martino Trevisan, Idilio Drago, Marco Mellia
TL;DR
COVID-19 restrictions shifted learning and work onto Internet services, creating unprecedented demand and raising questions about network resilience. The paper analyzes PoliTO campus traffic and service logs, including its internally hosted teaching platform. It finds sharply reduced incoming traffic, increased outgoing traffic driven by online teaching, and robust operation at the reported scale.
Problem
Lockdown-driven shifts to online learning and remote work created unprecedented Internet load, motivating analysis of their effects on campus traffic and services.
Method
The study analyzes PoliTO campus traffic, collaboration and remote-working platforms, and the internally hosted BigBlueButton teaching infrastructure using aggregate and service-level measurements.
Results
Incoming traffic drastically decreased, while outgoing traffic more than doubled as PoliTO served over 600 daily classes to more than 16 000 students and generated peaks of 1.5 Gbit/s.
Takeaways & Limitations
The campus and Internet proved robust enough to maintain university operations despite widespread online teaching, remote working, and collaboration.
Takeaways & Limitations
User counts assume client IP addresses are unique daily, and Microsoft Teams use was not compulsory for PoliTO staff.
Abstract
from arXiv · showhide
The COVID-19 pandemic led to the adoption of severe measures to counteract the spread of the infection. Social distancing and lockdown measures modifies people's habits, while the Internet gains a major role to support remote working, e-teaching, online collaboration, gaming, video streaming, etc. All these sudden changes put unprecedented stress on the network. In this paper we analyze the impact of the lockdown enforcement on the Politecnico di Torino campus network. Right after the school shutdown on the 25th of February, PoliTO deployed its own in-house solution for virtual teaching. Ever since, the university provides about 600 virtual classes daily, serving more than 16,000 students per day. Here, we report a picture of how the pandemic changed PoliTO's network traffic. We first focus on the usage of remote working and collaborative platforms. Given the peculiarity of PoliTO in-house online teaching solution, we drill down on it, characterizing both the audience and the network footprint. Overall, we present a snapshot of the abrupt changes on campus traffic and learning due to COVID-19, and testify how the Internet has proved robust to successfully cope with challenges and maintain the university operations.
1. Introduction
COVID-19 restrictions sharply changed mobility and increased reliance on Internet-based work, learning, and collaboration. PoliTO studies these changes through its campus network, highlighting rapid adoption of online teaching and the network’s ability to sustain operations.
- Motivation: Lockdown measures drastically reduced mobility while driving greater use of remote working, e-learning, collaboration, gaming, and streaming services.Internet traffic volume increased by about 40%, sometimes alongside reduced download performance.
- Paper scope: PoliTO deployed an in-house BigBlueButton e-learning platform to support all second-semester classes during the lockdown.The platform was designed, installed, and tested during the first week of March, going live one week later.
- Main findings: 10 times decrease in incoming traffic accompanied a 2.5 times increase in outgoing traffic, driven by more than 600 daily online classes serving around 16 000 students per day.These figures summarize the paper’s central observation of an inversion in campus traffic patterns.
- Main findings: More than 17 000 chat messages and over 1 000 calls per day indicate an explosion in online collaboration among faculty and staff.The paper also reports no major connectivity problems, apart from a few possibly 3G/4G-related poor-performance cases.
- Conclusion: The paper concludes that remote working, e-learning, and online collaboration were viable responses to social-distancing policies and that the Internet coped with the sudden connectivity demand.It suggests that online collaboration may remain relevant beyond the pandemic.
2. Datasets and methodology
The study combines campus-wide traffic measurements with fine-grained flow, service, application, and security logs. These sources characterize traffic changes, remote-working adoption, online teaching, and unsolicited activity during February–April 2020.
- Data sources: The dataset covers February 1 to April 5, 2020, using multiple sources to measure traffic and campus services.The study examines a medium-sized engineering and architecture university with about 35 000 students and two 10 Gbit/s Internet links.
- Aggregate traffic: GARR edge-router statistics provide five-minute traffic-volume time series for PoliTO, compared with Politecnico di Milano and Università di Torino.These data support cross-campus comparison of aggregate incoming and outgoing traffic.
- Fine-grained measurements: Tstat passively measures campus traffic at flow level, exposing TCP/UDP fields together with RTT and packet-loss metrics.The infrastructure provides finer-grained measurements than GARR’s high-level aggregate data.
- Application and service logs: Service logs track VPN and RDP access, Microsoft Teams collaboration, and email, antispam, and security services.Management-console logs reveal how service adoption changed during the epidemic.
- Teaching and security: BigBlueButton application logs and legacy teaching-platform logs measure virtual-classroom consumption, performance, and access to teaching materials.Firewall and darknet logs additionally support analysis of SSH, RDP, SIP, and unsolicited traffic patterns.
3. Impact on campus traffic and remote working solutions
Lockdown sharply reduced campus incoming traffic but increased PoliTO’s outgoing traffic because its internally hosted online-teaching service generated substantial uploads. Remote-working and collaboration services also saw rapid, sustained adoption.
- 3.1. Aggregate traffic volume: Incoming traffic fell to about one tenth of its pre-lockdown level at PoliTO and UniTO, while PoliTO’s outgoing traffic increased because of its in-house teaching platform.PoliTO’s online-teaching service produced about 2.8 times baseline outgoing traffic during peak time.
- 3.1. Aggregate traffic volume: PoliTO’s in-house online-teaching system caused a massive inversion from incoming to upload traffic during lockdown.Unlike PoliTO, the comparison campuses used cloud-based solutions and did not show relevant outgoing-traffic changes.
- 3.2. Smart working adoption: VPN and RDP usage increased simultaneously after the March 11 lockdown, with SSL VPN growing more than IPSec VPN.The authors suggest non-expert users preferred SSL VPN because it was easier to configure.
- 3.2. Smart working adoption: About 500 users contacted RDP services daily after lockdown, while weekend RDP accesses exceeded 200 per day.Several-hour sessions suggest regular work activity rather than occasional file access.
- 3.3. Remote collaboration: Microsoft Teams adoption expanded to more than 700 daily private-chat users, over 17 000 messages per day, and more than 1 500 calls per day.Video calls and meetings reached 400 users per day, while team-chat usage declined after an initial surge.
- 3.3. Remote collaboration: The lockdown pushed smart working into widespread use through VPN, RDP, VoIP, and online conferencing services.The paper describes these services as becoming regular tools for PoliTO users working from home.
4. Online teaching
PoliTO’s emergency BigBlueButton deployment achieved high daily participation while generating substantial, schedule-driven campus traffic. Live classes created sharp peaks, whereas on-demand material remained heavily accessed beyond lecture hours, with modest bandwidth requirements for students.
- Audience: More than 16,000 students joined at least one of around 700 virtual classrooms daily, with participation remaining stable over time.Over 70% of student feedback rated session technical quality 4 or 5 stars.
- Audience: 47% of classes were followed by students outside Piedmont, while most participants used Windows 10 PCs or Mac OS X browsers.Android and iOS accounted for 6.2% and 2.4% of students, respectively.
- Network workload: Weekday outbound teaching traffic exceeded 1 Gbit/s, with live streaming responsible for slightly more than one third of the total.Weekend traffic still reached 750 Mbit/s because students downloaded on-demand lectures and teaching material.
- Network workload: Live-classroom connections peaked above 4,500 when scheduled lectures began, while on-demand access was distributed throughout the day and continued after midnight.Lectures started every 90 minutes between 8:30 AM and 7 PM.
- Session characteristics: All virtual classes carried audio at 40 kbit/s, while video commonly used 100 kbit/s and only 10% of streams reached 300 kbit/s or higher.Students needed no more than 0.5-1.0 Mbit/s to enjoy a lecture.
- Session characteristics: Half of multimedia streams lasted less than 30 minutes, but bumps at 1.5 and 3 hours matched typical PoliTO class durations.Short streams cannot be directly interpreted as student abandonment because servers may reset streams for other reasons.
- Network workload: Live classrooms imposed large scheduled peaks, yet bandwidth was not a major problem; non-live media remained heavily consumed during evenings and weekends.The paper’s takeaway is that students required less than 1 Mbit/s downstream bandwidth for live lectures.
5. Network performance
The study evaluates teaching-content download performance by ISP and Italian region using flow-level measurements. Most connections performed adequately for online teaching, but access technology produced important differences in throughput, delay, and variability.
- Internet Service Providers breakdown: Most flows achieved average throughput above 5 Mb/s, while Fastweb’s median reached 10 Mbit/s.Violin plots represent per-flow average-throughput probability densities, with white dots marking medians.
- Internet Service Providers breakdown: 3G/4G-only providers had throughput below 5 Mbit/s, indicating possible client-side impairment such as congestion.The analysis used TCP downloads of HTTP objects at least 10 MB in size.
- Internet Service Providers breakdown: Fastweb connections had a condensed RTT distribution with a median around 20 ms, whereas wireless-only providers had medians above 40 ms and peaks above 100 ms.RTT reflects physical distance, access-technology delays, congestion, and Internet routing.
- Internet Service Providers breakdown: The vast majority of Italian ISPs could meet the roughly 1 Mbit/s needed for live-streaming classes, but some 3G/4G customers might experience delays and limited throughput.These limitations had potential to disturb user experience.
- Geographical characteristics: Students connected from all Italian regions, with the largest number of connections coming from Piedmont.The broad regional participation enabled comparisons across locations.
- Geographical characteristics: Latium connections typically had lower RTT than Piedmont despite geography, because Internet routing through Rome or Milan creates region-dependent detours.Northern Veneto and Emilia-Romagna also showed substantial RTT variability, with some connections exceeding Sicily’s median.
- Geographical characteristics: Median average throughput in the best-performing regions was twice that of the lowest-performing regions.Sicily and Friuli-Venezia-Giulia combined relatively large median RTTs with good throughput, consistent with distance and routing effects.
- Geographical characteristics: Overall access performance remained good for online teaching, while access technology was more critical to flow performance than physical or routing distance.The paper notes only limited Internet-performance problems in the broader lockdown context.
6. Security events and unsolicited traffic
PoliTO’s security monitoring found little evidence that lockdown-related remote-working changes produced new malicious-traffic trends. RDP events temporarily increased after new servers were deployed, while other observed variations were consistent with normal activity or unrelated trends.
- Firewall events: The firewall monitored SSH, RDP, and SIP scan/brute-force events to assess whether attack patterns changed during lockdown.The first week served as the reference for weekly comparisons.
- Firewall events: RDP events more than doubled during the first lockdown week after new RDP servers were enabled for smart working.The increase reflected deployment of new systems, which exposed additional servers to scanning and brute-force activity.
- Firewall events: Subsequent RDP-event declines followed firewall changes intended to limit scanning activity.The reported decrease was explained by changes in the firewall setup rather than by a broader lockdown effect.
- Darknet traffic: Darknet traffic showed the noisy variations normally associated with large-scale Internet scans, including sudden increases during lockdown weeks.The paper characterizes this pattern as typical rather than exceptional for the lockdown period.
- Unsolicited traffic: SPAM emails decreased during March, but the decline had begun months earlier and was not attributed to lockdown.Together with other security observations, this provided no clear lockdown-related signal in unsolicited traffic.
7. Related Work
Prior work documented how catastrophic events and the COVID-19 outbreak affected Internet usage, mobility, and traffic. Reports from operators and exchanges consistently described substantial shifts in broadband, upstream, downstream, voice, and data traffic.
- Industry reports: Business and web-analytics reports recorded reduced access in travel-related sectors and changed user habits during the outbreak.SimilarWeb reported shifts across sectors including air travel, hotels, and car rentals.
- Industry reports: Operator and exchange measurements reported major traffic growth, including over 50% higher fixed broadband usage in Italy and Spain and 10–40% increases at European Internet Exchange Points.Other reported changes included upstream, downstream, mobile voice, and mobile data growth.
- Prior research: COVID-19-specific research papers had not yet been published, so the paper situated its study alongside earlier work on Internet behavior during disasters and unrest.Examples included earthquake-related traffic and routing changes, inter-domain rerouting, and the Internet’s role during the Egyptian Revolt.
8. Conclusions
The study examines how lockdown and the shift to online collaboration and e-learning changed traffic at the Politecnico di Torino campus. Incoming traffic fell sharply while outgoing traffic more than doubled to support large-scale online teaching, and the university found no clear evidence of lockdown-driven changes in major security events.
- 8. Conclusions: Lockdown measures forced people toward online learning, smart working, and leisure services, generating unprecedented Internet load while most people stayed at home.Italy’s total lockdown was described as keeping 80–90% of people at home.
- 8. Conclusions: Incoming campus traffic drastically decreased, while outgoing traffic more than doubled as PoliTO shifted to online collaboration and e-learning.The conclusion frames these changes as the campus-network impact of lockdown measures.
- 8. Conclusions: More than 600 daily classes served over 16 000 students and generated traffic peaks of 1.5 Gbit/s through PoliTO’s in-house teaching infrastructure.The same infrastructure supported remote teaching at campus scale during the lockdown.