Transcription
Dogs Never Get Tired:Power and Edge ComputingUsenix Annual Technical Conference (ATC) 2021Session Chairs: Marcelo Martins, AppleDilma da Silva, Texas A&M UniversityPreview by Ketan Bhardwaj, Georgia Tech
Once there were a few big dogs
Had to haul more, Were a bit slow IoT
The big ones already knew it Akamai CDNNetflix Open ConnectGoogle Global CachePushed content cloud closer to users Reduce backhaul demand Several 10s to few 100s ms O(few 1000s) global locations
Challenge #2: How many Edges?How fast do they really needed to be THE ADVENT OF EDGE COMPUTINGMany edges, per use case and workload541115G (target)20WiFi (average)1-2020-80millisecondsRTT UK N-S (IP)(564 miles)Perception of a screenchange after clicking abutton100An airbag toreleaseSpeed of lightper 200 kmmilliseconds3020Augmented7Reality backCar at 140 km/h,3.9 cm1voiceTactileInternet15Two people in a room,latencyadded by speedTell that aof soundsound is a human100TeleRoboticsAuditory reactiontime4G/LTE(average)00RTT USA W-E (IP)(2514 miles)Visual reactiontime803G (average)Screen to brainpropagationHuman blinkmilliseconds 150millisecondsmsHuman FeaturesmsUse CaseRequirements1013015080-150Network &Spatial7Autonomous8DrivingCar at 90 km/h,20 cm 1750ms150Crisp UI with acomputer80AugmentedReality(non gaming)WearableAssistance10TV max interpicture10Virtual RealityTelefónica Alpha / Edge Computing / July 2017Source: Pablo Rodriguez, Telefonica Alpha, SEC’17Telefónica Alpha / Edge Computing /26
So, there were more but smaller ones Verizon partners with AWS to bring more power to its 5G edge FierceWirelessAT&T and Microsoft announce early result of their 5G-cloud partnership - CNN1/31/20, 8)35 ADeutsche Telekom Embarks on Edge Computing for IoT - SDxCentralAT&T and Microsoft announce earlyresult of their 5G-cloud partnershipSponsored HubsWIRELESSTECH5GCiscoDellTechnologiesNext GenWirelessSummitVersaNetworksVideoWirelineGet SDxCentral'sResourcesWebinarsDaily NewsletterEIOTLIVE TVBy Clare Duffy, CNN Business5GUpdated5:47 PM ET, Tue November 26, 2019Learn MoreVerizon partners with AWS to bring more power to its PEN SOURCEIedgeNewsNewYork (CNN Business) — AT&T and Microsoft have announced the next phase of a partnership that's aimed atadvancing developmentin two of the buzziest areas of tech: 5G and cloud computing.by Sue Marek Dec 3, 2019 4:33pmDeutsche Telekom Embarkson Edge Computing for IoTRelated5GResourcesThe companies inked an agreement in July to work together on innovations in 5G and cloud.is next generationnetwork technology that's expected to change the way people live and work. And the cloud is distributed dataAssuring Intelligence for thestorage and computing power that doesn't rely on traditional on-site data centers. AT&T (T) (CNN's parentWANEdge:company) also agreed to migrate its data and workflows to the Microsoft Azure cloud, inwhatwasWebinarreportedly aShare this article:MattKapko Editormore than 2billiondealfor Microsoft (MSFT), according to Reuters.Webinar: Assuring IntelligenVendor"networkSD-WAN EdgeOn Tuesday, Microsoft and AT&T's business unit announced the pilot launch of their so-callededgeDecember 3, 2019 12:22 PMcompute" technology, an early outcome of the partnership. The advancement is expected to help both companiesAssuring Intelligence for theenterprise customers take advantage of the benefits of 5G, as AT&T works to roll out 5G across the US by midWAN Edge2020.Sponsored by Infovista"We've said all year, developers and businesses will be the early 5G adopters, and this puts both at the forefront ofthis revolution," AT&T Business Chief Marketing OWcer Mo Katibeh said in a statement Tuesday.AT&T will implement the Microsoft technology in some of the major metro areas where it's installed 5G — AT&Tsays it will be available for select customers in Dallas early 2020, with Los Angeles and Atlanta slated forimplementation later in the year.Microsoft and AT&T are excited about the technology's potential, but Gartner analyst Bill Menezes says it's not yetclear just how useful this technology will be and when it will be usable.Verizon CEO Hans Vestberg made an appearance at AWS re:Invent conference in Las Vegas to talk about how Verizon plans to incorporate"Right now, it's creatinglot of andnoisethatplaysonedge.a lot(AWS)of current buzzwords in tech," Menezes said. "What itAWS cloudacomputestorageat thenetworkactually produces remains to be seen."
Where are these small ones Where are the edges?Distributed cloud, edge compute, AI/ML, IoT, 5G, VNFs/NFV, FMCEDGEEnterpriseEDGEMEC server,AI/ML, IoT, 5G& IoTEDGE97% ofoperatorsplan VNFexecution inSmart COIoT G/WEDGEVNFs, vEPC,MEC, distributedRAN, vRAN,BBU hotel, FMC,vCPE, AI/ML, IoTgo herePARTIAL EDGENOT EDGE85% of operatorsplan VNF executionin DC Near CO70% of operators planVNF execution inDC Not Near COIoT G/WEDGE82% ofoperatorsplan VNFexecutionon uCPEatcustomersitesuCPESmart COwith MiniDC/CDN [Edge DC]uCPEData Center NearCO[Regional DC]Data CenterNot Near CO[Central DC]20msecs6Source: IHS Markit. NFV Strategies: Global Service Provider Survey, June 2017; Respondents control 61% of global telecom capex
How do the small ones help Reduce latency to few ms Save backhaul bandwidthWhat help do small ones need Handle multiple applications Power/energy constraintsThis session
Prediction-Based Power Oversubscription inCloud Platforms Power is important Power oversubscription islimited Prior work assumes hardto get info Reactive, expensivemechanisms Prediction-based approach Predict VM criticality &utilization Protect critical VMs byintelligent placement 2x oversubscription
Proactive Energy-Aware Adaptive VideoStreaming on Mobile Devices ABR widely used video 360streaming – App adaptation Prior work, mostly reactive causesadaptation oscillations Difficult to predict energy usage Energy use is cummulative Integrated, Pro-active ABRadaptations for video Reactive vs. Proactive Function wise power prediction Practical control algorithm
Video Analytics with Zero-streaming Cameras A lot of cameras (surveillance) Huge amounts cold video(nothing interesting) Especially for ad-hoc queries:retrieve, tag, count Why stream even to edge Cloud processing only needed ifmore accuracy is needed Zero-streaming Leverage spatial skew inlandmarks Cloud provides increasinglycomplex / more accurateoperators based on policy Multi-pass, multi-operators 100x more video in realtime
ASAP: Fast Mobile Application Switch viaAdaptive Prepaging Everyone uses more than one app Switching b/w them can be faster(useful) Cause – slow demand paging Requires careful low level designto avoid overheads Predict likely to be accessedpages / pre-page Ensure reasonable overheads 20-30% faster switch Additional 30% CPU & 25% diskbandwidth usage
Enjoy the session J
Telefónica Alpha / Edge Computing / 26 Challenge #2: How many Edges? 1-20 20-80 milliseconds milliseconds 1 Visual-tactile feedback m sHuman Features m s Network & Spatial m s Use Case Requirements 1 1 Tactile Internet 5G (target) 1 Car at 140 km/h, 3.9 cm Wearable Assistance 4 Tell that a sound is a human voice 7
