Cumulo-nimble: giving the cloud a competitive edge

Like com­pa­nies all around the world, US fast-food chain Taco Bell respond­ed to the pandemic’s com­mer­cial impact by accel­er­at­ing its shift to the cloud. As cus­tomers’ tra­di­tion­al pat­terns of restau­rant and dri­ve-through con­sump­tion changed rapid­ly – and per­ma­nent­ly – to include kiosk, mobile and web order­ing, often through third-par­ty deliv­ery ser­vices, Taco Bell moved the remain­der of its group IT to cloud ser­vices.

But this 100% cloud-based approach stops at the restau­rant door. Giv­en that many of its 7,000 out­lets don’t have fast and/or reli­able inter­net con­nec­tions, the com­pa­ny has recog­nised the lim­i­ta­tions of the pub­lic cloud mod­el and aug­ment­ed its approach with edge com­put­ing. This set-up enables the com­pa­ny to process data near the phys­i­cal point at which it is cre­at­ed, with only a peri­od­ic require­ment to feed the most valu­able mate­r­i­al back to the cloud and receive updates from it.

Taco Bell is just one of thou­sands of firms seek­ing to exploit the fast-evolv­ing – and much-hyped – dis­trib­uted IT capa­bil­i­ty that edge com­put­ing can offer. 

“Edge com­put­ing is get­ting so much atten­tion now because organ­i­sa­tions have accept­ed that there are things that cloud does poor­ly,” observes Bob Gill, vice-pres­i­dent of research at Gart­ner and the founder of the consultancy’s edge research com­mu­ni­ty.

Issues of laten­cy (time-lag) and lim­it­ed band­width when mov­ing data are key poten­tial weak­ness­es of the cen­tralised cloud mod­el. These dri­ve a clear dis­tinc­tion between the use cas­es for cloud and edge com­put­ing. But the edge is also a focus for many organ­i­sa­tions because they want to add intel­li­gence to much of the equip­ment that sits with­in their oper­a­tions – and to apply AI-pow­ered automa­tion at those end points. 

Ear­ly adopters include man­u­fac­tur­ers imple­ment­ing edge com­put­ing in their plants as part of their Indus­try 4.0 plans; logis­tics groups seek­ing to give some auton­o­my to dis­persed assets; health­care providers with med­ical equip­ment scat­tered across hos­pi­tals; and ener­gy com­pa­nies oper­at­ing wide­ly dis­persed gen­er­a­tion facil­i­ties.

“For such appli­ca­tions to be viable and effi­cient, their data must be processed as close to the point of ori­gin or con­sump­tion as pos­si­ble,” says George Elis­saios, direc­tor of prod­uct man­age­ment at Ama­zon Web Ser­vices. “With edge com­put­ing, these appli­ca­tions can have low­er laten­cy, faster response times and give end cus­tomers a bet­ter expe­ri­ence. Edge com­put­ing can also aid inter­con­nec­tiv­i­ty by reduc­ing the amount of data that needs to be back­hauled to data cen­tres.”

Combining cloud and edge computing

In some ways, the emer­gence of edge com­put­ing rep­re­sents a new topol­o­gy for IT. So says Paul Sav­ill, glob­al prac­tice leader for net­work­ing and edge com­put­ing at Kyn­dryl, the provider of man­aged infra­struc­ture ser­vices that was recent­ly spun out of IBM. 

Com­pa­nies are look­ing at the edge as “a third land­ing spot for their data and appli­ca­tions. It’s a new tier between the pub­lic cloud and the intel­li­gence at an end device – a robot, say,” he explains.

But most organ­i­sa­tions don’t expect their edge and cloud imple­men­ta­tions to exist as dis­tinct enti­ties. Rather, they want to find ways to blend the scal­a­bil­i­ty and flex­i­bil­i­ty they have achieved with the cloud with the respon­sive­ness and auton­o­my of inter­net-of-things (IoT) and satel­lite proces­sors installed at the edge. 

Gill believes that “cloud and edge are pure yin and yang. Each does things the oth­er doesn’t do well. When put togeth­er effec­tive­ly, they are high­ly sym­bi­ot­ic.”

They will need to be, as more and more intel­li­gence is moved to the edge. More than 75 bil­lion smart dig­i­tal devices will be deployed world­wide by 2025, accord­ing to pro­jec­tions by research group IHS Mark­it. And it is nei­ther desir­able nor real­is­tic for these to be inter­act­ing con­tin­u­ous­ly with the cloud.

“When you start to add in mul­ti­ple devices, you see a vast increase in the vol­ume, veloc­i­ty and vari­ety of the data they gen­er­ate,” says Greg Han­son, vice-pres­i­dent of data man­age­ment com­pa­ny Infor­mat­i­ca in EMEA and Latin Amer­i­ca. “You sim­ply can’t keep mov­ing all of that data into a cen­tral point with­out incur­ring a sig­nif­i­cant cost and becom­ing reliant on net­work band­width and infra­struc­ture.”

In such sit­u­a­tions, edge IT per­forms a vital data-thin­ning func­tion. Satel­lite proces­sors sit­ting close to the end points fil­ter out the most valu­able mate­r­i­al, col­late it and dis­patch it to the cloud peri­od­i­cal­ly for heavy­weight analy­sis, the train­ing of machine-learn­ing algo­rithms and longer-term stor­age. Proces­sors at the edge can also apply data secu­ri­ty and pri­va­cy rules local­ly to ensure reg­u­la­to­ry com­pli­ance. 

Gill notes that edge com­put­ing has shift­ed quick­ly “from con­cept and hype to suc­cess­ful imple­men­ta­tions. In many ver­ti­cal indus­tries, it is gen­er­at­ing rev­enue, sav­ing mon­ey, improv­ing safe­ty, enhanc­ing the cus­tomer expe­ri­ence and enabling entire­ly new appli­ca­tions and data mod­els.”

Overcoming challenges

Before achiev­ing such gains, many edge pio­neers are like­ly to have sur­mount­ed numer­ous sig­nif­i­cant chal­lenges. Giv­en that the tech­nol­o­gy is imma­ture, there are few wide­ly accept­ed stan­dards that busi­ness­es can apply to it. This means that they’re often faced with an over­whelm­ing­ly wide range of designs for tech rang­ing from sen­sors and oper­at­ing sys­tems to soft­ware stacks and data man­age­ment meth­ods.

Such com­plex­i­ty is reflect­ed in a wide­spread short­age of spe­cial­ist exper­tise. As Sav­ill notes: “Many com­pa­nies don’t have all the skills they need to roll out edge com­put­ing. They’re short of peo­ple with real com­pe­tence in the orches­tra­tion of these dis­trib­uted appli­ca­tion archi­tec­tures.”

The goal may be to blend cloud and edge seam­less­ly into a uni­fied mod­el, but the start­ing points can be very dif­fer­ent. There are two fun­da­men­tal­ly dif­fer­ent – though not total­ly con­tra­dic­to­ry – schools of thought, accord­ing to Gill. The ‘cloud out’ per­spec­tive, favoured by big cloud ser­vice providers such as Ama­zon, Microsoft and Google, views the edge as an exten­sion of the cloud mod­el that extends the capa­bil­i­ties of their prod­ucts. 

The oth­er approach is known as ‘edge in’. In this case, organ­i­sa­tions devel­op edge-native appli­ca­tions that occa­sion­al­ly reach up to the cloud to, say, pass data on to train a machine-learn­ing algo­rithm.

Adher­ents of either approach are see­ing sig­nif­i­cant returns on their invest­ments – when they get it right. 

“We may be in the ear­ly phase of exploit­ing that com­bi­na­tion of IoT, edge and cloud, but the capa­bil­i­ties enabling these dis­trib­uted archi­tec­tures – the soft­ware con­trol and orches­tra­tion tools and the inte­gra­tion capa­bil­i­ties – have already reached the point where they’re high­ly effec­tive,” Sav­ill reports. “Some com­pa­nies that are fig­ur­ing this out are see­ing oper­a­tional sav­ings of 30% to 40% com­pared with more tra­di­tion­al con­fig­u­ra­tions.”

In doing so, they are also herald­ing a large-scale resur­gence of the edi­fice that cloud helped to tear down: on-premis­es IT – albeit in a dif­fer­ent form.  

“In the next 10 to 20 years, the on-premis­es pro­file for most com­pa­nies will not be servers,” Elis­saios pre­dicts. “It will be con­nect­ed devices – and bil­lions of them.”