Presentation: Extended summary of “Cloudy with a Chance of Short RTTs Analyzing Cloud Connectivity in the Internet"
- 1. Extended summary of "Cloudy with a Chance of Short RTTs Analyzing Cloud Connectivity in the Internet" Università degli Studi di Trieste Dipartimento di Ingegneria e Architettura Corso di Laurea in Ingegneria Elettronica e Informatica Relatore: Prof.Alberto Bartoli Candidata: IsabellaFilippo A.A. 2022/2023
- 2. Introduction: Cloud Computing • Model for delivering computing services over the internet • Flexible and on-demand access ✓Storage ✓Processing ✓Applications • Crucial for the ever-increasing growth of network services on the internet 2 Instead of unlimited computing capacity, applications of today demand lower latencies.
- 3. Introduction: Need for Short RTTs 3 Human Reaction Time (HRT): delay difference between visual stimulus and associated motor response • Typically less than 250 ms end-to-end • Requirement for remote human-in-the- loop application Perceivable Latency (PL): latency when the user starts experience lags in application • Typically within 80-100 ms end-to-end • Strict requirement for interactive application Motion-To-Photon (MTP): delay between user input and its effect reflecting on display • Typically within 10-20 ms end-to-end • Strict requirement for immersive applications
- 4. Introduction: Problem 4 • Focus of the study: o Comparison between Speedchecker measurements and a previous experiment o Analysis of the global cloud access latency o Analysis of the impact of wireless last-mile on overall latency Can current cloud infrastructure support the low latency requirements of such apps?
- 5. Methodology: Vantage Points and End-Points Vantage Points • More than 115,000 Speedchecker (SC) probes • Probes are softwareized application • Cover 95.6% of Internet population • Comparison with RIPE Atlas: 5 9K vs 115K total probes wired vs wireless End-Points • 195 cloud regions operated by 9 major cloud providers of different size • Operators have different WAN deployments
- 6. Global Cloud Access Latency 6 • Achieving Motion-to-photon (MTP) threshold (≤ 20 ms) is difficult for real Internet users globally • Achievable Perceivable latency (PL) threshold (≤ 100 ms) in Europe, Oceania and North America • In South America only probes geographically “close” to data centers enjoy low latencies • Severe under-provisioning of DCs in Africa results in unachievable Human Reaction Time (HRT) threshold (≤ 250 ms)
- 7. Influence of Wireless Last-Mile • Cloud providers o Continue to expand their WAN deployment o Reduce user path lengths to their network through peering o Last-mile remains out of reach • Previous studies show LTE and Wi- Fi to be the major latency bottleneck o Packet drops, invariable delays 7
- 8. Influence of Wireless 8 • Absolute latency due to last-mile similar for home (Wi-Fi + Router) and cellular (only wireless), i.e. ≥ 20 ms • Wired connection shows similar behaviour to RIPE Atlas Wi-Fi Router: Wired part of the connection
- 9. Last-mile wireless connection remains a challenge for cloud access. Conclusions: 1. Impact on latency comes from geographical distance to data centres 2. Choice of measurement platform can significantly influence results 3. Wireless last-mileaccess contributes significantly to overall latency 9 Discussion: Edge computing: o Could help to reach low latency requirements but…
- 10. Thank You For Your Attention Presented by Isabella Filippo The Khang Dang, Nitinder Mohan, Lorenzo Corneo, Aleksandr Zavodovski, Jörg Ott, Jussi Kangasharju. Cloudy with a Chance of Short RTTs Analyzing Cloud Connectivity in the Internet.