The next wave of ca­bling

Network Middle East - - CONTENTS -

Networking ca­bling in­fra­struc­tures need to evolve in this age of dig­i­tal trans­for­ma­tion. Net­work Mid­dle East ex­am­ines what the key con­sid­er­a­tions are to en­sure least dis­rup­tion to busi­ness be­fore up­grad­ing data cen­tre net­works

Networking is chang­ing at a fast pace in the data cen­tre. The net­work ar­chi­tec­tures that make cloud pos­si­ble are fun­da­men­tally dif­fer­ent from the highly over-sub­scribed, hi­er­ar­chi­cal and costly legacy so­lu­tions of the past. In 2020, ad­vanced tech­nolo­gies like 5G are ex­pected to be ap­plied in the data cen­tre, of­fer­ing data cen­tre providers greater op­por­tu­nity to grow and en­hance their busi­ness. While the ben­e­fits of these tech­nolo­gies may take at least a few years to come to fruition, data cen­tre com­pa­nies that in­cor­po­rate them into their busi­ness strat­egy now will be the ones best po­si­tioned to reap the ben­e­fits down the road.

The de­ploy­ment of ma­chine learn­ing, deep learn­ing and other ar­ti­fi­cial in­tel­li­gence tech­nolo­gies is now main­stream, and these tech­nolo­gies power many of the cloud ser­vices we use ev­ery day, says Alas­tair Waite, data cen­tre market de­vel­op­ment, Comm­scope. “Plan for change from day one, build with max­i­mum flex­i­bil­ity and choose a good in­stal­la­tion part­ner to sup­port you, ” he adds.

Networking ca­bling in­fra­struc­tures in the DX age

In­creased adop­tion of high-per­for­mance servers and ap­pli­ca­tions re­quir­ing higher band­width is driv­ing adop­tion of 10 and 25 Gi­ga­bit Eth­er­net switch­ing in com­bi­na­tion with 40 and 100 Gi­ga­bit Eth­er­net. As a re­sult, it has be­come in­creas­ingly nec­es­sary to in­clude net­work fab­rics, soft­ware-de­fined networking, hy­per-con­ver­gence, and soft­ware-de­fined stor­age tech­nolo­gies in the data cen­tre. From a ca­bling stand­point, all this added com­plex­ity and mod­erni­sa­tion of the data cen­tre pro­motes the need for sim­plic­ity, re­li­a­bil­ity and high-den­sity.

To plan an up­grade or a new tech­nol­ogy leap within the cus­tomer net­work, the tech­nol­ogy of choice should have a proven sta­bil­ity record, be suit­able for the new ap­pli­ca­tions that keep com­ing out, and should be fu­ture-ready. Also, it should be clear how the tech­nol­ogy can be im­ple­mented or up­graded. In gen­eral, cus­tomers in the Mid­dle East are aware of the im­por­tance of dig­i­tal trans­for­ma­tion and or­gan­i­sa­tions are in­creas­ingly tak­ing steps to be pre­pared.

Cus­tomers need a clear vi­sion of what the fi­nal re­sults should be and how to struc­ture the up­grade process, to en­sure all steps are un­der con­trol and dis­rup­tions are min­imised, ex­plains Arafat Yousef, manag­ing direc­tor MEA, Nex­ans Ca­bling So­lu­tions.

“It makes sense to choose so­lu­tions that can be up­graded on a mod­u­lar ba­sis, al­low­ing the cus­tomer to add or change fea­tures in line with re­quire­ments. Things be­come even bet­ter when a so­lu­tion al­lows changes to be made at com­po­nents level, while the key frameof-work re­mains fixed. This means bet­ter prepa­ra­tion for fu­ture ex­pan­sions with min­i­mal dis­rup­tions and ac­cept­able cost,” adds Yousef. “We al­ways find big crowds at­tend­ing sem­i­nars and road­shows by all ven­dors, across var­i­ous ar­eas of in­ter­est. Here, peo­ple look for­ward to get­ting the lat­est up­dates about stan­dards and in­dus­try trends, to plan up­grades to their cur­rent net­works and the di­rec­tion of new projects ac­cord­ingly.”

In­ter­nal data cen­tre traf­fic is ex­pected to grow by 80% over the next three years. Be­cause of this, there is a real risk of net­works be­com­ing band­width bot­tle­necks. Re­ports in­di­cate that 65% of sys­tem out­ages are re­lated to ca­bling and that patch­ing mis­takes are the rea­son for 28% of down­time in data cen­tres. How­ever, when plan­ning the in­stal­la­tion or up­grade of net­work ca­bling, the ac­tual cost of ca­bling typ­i­cally only ac­counts for just 4-5% of the to­tal ex­pense of the data cen­tre.

As a stan­dard prac­tice, or­gan­i­sa­tions must move away from tra­di­tional low­den­sity ca­bling to high-den­sity mod­u­lar struc­tured cable so­lu­tions, says Na­bil Khalil, ex­ec­u­tive vice pres­i­dent of R&M META. “By do­ing so, they can im­ple­ment phys­i­cal net­work in­fra­struc­ture in a far more man­age­able and flex­i­ble man­ner. Fur­ther­more, these sys­tems en­able data cen­tres to eas­ily mi­grate to 25, 100 and 200 Gb/s net­works and solve some of the most crit­i­cal net­work chal­lenges.”

Fi­bre: To be or not to be

With the new Cat­e­gory 8 stan­dards ca­pa­ble of sup­port­ing up to 40 GBE, many are left un­de­cided if there is still scope for cop­per in the data cen­tre or whether fi­bre is the only way for­ward.

Fi­bre op­tic ca­bling of­fers sev­eral ad­van­tages over cop­per such as higher through­put, space sav­ings, bet­ter se­cu­rity and fu­ture-proof­ing while cop­per ca­bling can only be de­ployed when it can re­li­ably meet the re­quire­ments of a spe­cific ap­pli­ca­tion. How­ever, one rea­son why or­gan­i­sa­tions still favour cop­per is that they tend to be a cheaper al­ter­na­tive to fi­bre.

Waite ex­plains that there is still scope and de­mand for struc­tured cop­per ca­bling in the data cen­tre. “Cus­tomers still rely on cop­per to pro­vide net­works for out-of-band man­age­ment of servers and switches, wire­less ac­cess points, se­cu­rity ac­cess sys­tems, and the many IOT sen­sors that are lo­cated within the cabi­net, cage and all across the data hall to en­sure the smooth and ef­fi­cient run­ning of the white space. New ad­vances in POE tech­nol­ogy have also en­sured that cop­per re­mains a rel­e­vant trans­port me­dia all over the data cen­tre cam­pus.”

Non-hy­per­scale data cen­tres still use a great deal of cop­per ca­bling to con­nect servers. Eco­nom­i­cally, this makes per­fect sense. Cop­per Twisted pair runs at 1Gbps or 10Gbps to­day. 25Gbase-t and 40Gbase-t are de­fined as the next steps in Eth­er­net stan­dards but are not yet de­ployed to the market, as the band­width de­mand in non-hy­per­scale data cen­tres can still be sat­is­fied with 10G per port.

When it comes to fu­ture-readi­ness, fi­bre is cur­rently hy­per­scale data cen­tres’ first choice, adds Yousef of Nex­ans. “How­ever, when the need for higher speeds ar­rives, new ca­bling will clearly be re­quired: Cat6a will no longer suf­fice.”

Nex­ans rec­om­mends us­ing Cat7a ca­bling as soon as twisted pair is ca­pa­ble of run­ning be­yond 10G. Cat7a sup­ports 25GBASE-T. Cat8 ca­bling is re­quired for 40GBASE-T, but the eco­nomics of sys­tem de­sign have al­ready made this speed ob­so­lete. “We ex­pect only 25Gbase-t to hit the market in the fu­ture. Cat8 would be us­able, but Cat7a is al­ready suf­fi­cient and more af­ford­able.

R&M VP Khalil ex­plains that cop­per and fi­bre both hold their ground de­pend­ing on spe­cific use cases. “Cop­per is still widely used, par­tic­u­larly in the 10G ports shipped glob­ally, how­ever, when the net­work team wants to pro­vide a back­bone in 40G or 100G, they will need to start think­ing about fi­bre ca­bling, es­pe­cially when the dis­tances in­volved ex­ceed 30 me­ters. This is be­cause of the new Cat. 8 stan­dards can pro­vide 40G over 30 me­ters. So, it re­mains a cheaper and vi­able so­lu­tion for small dis­tances.”

Fi­bre-only data cen­tres may not be the most eco­nom­i­cal so­lu­tion, but are usu­ally cho­sen based on other con­sid­er­a­tions. Con­se­quently, the de­ci­sion whether en­ter­prise data cen­tres are bet­ter off with fi­bre or cop­per de­pends very much on the size, lay­out and speed re­quire­ments and should be stud­ied care­fully case by case.

While fi­bre op­tic ca­bling will re­main the medium of choice in the data cen­tre back­bone, Cat­e­gory 8 cop­per ca­bling is cer­tainly a valu­able con­sid­er­a­tion for the data cen­tre edge where up­link speeds in switch-to-server con­nec­tions are mov­ing from 10 to 25 and 40 Gi­ga­bit Eth­er­net speeds, ex­plains Prem Ro­drigues, direc­tor for the MEA & In­dia/ SAAR, Siemon.

Au­to­mated In­fra­struc­ture Man­age­ment – a must for fi­bre sys­tems in data cen­tres?

In the data cen­tre world, grad­ual mi­gra­tion to 100, 200 and 400 Gbit/s is un­der­way. Con­se­quences in­clude higher den­sity and more ca­bling in racks. To­day’s data cen­tres may con­tain hun­dreds of thou­sands of ports and patch cords. The com­plex­ity and dy­namism of the in­fra­struc­tures is reach­ing a scale at which hu­mans can no longer man­age them with­out re­mote-con­trolled hard­ware and soft­ware sup­port. What’s more, there are sig­nif­i­cantly more servers than can be man­aged on-site by qual­i­fied staff.

We are also see­ing data cen­tres move to­wards low loss fi­bre con­nec­tiv­ity, adds Ro­drigues of Siemon. “A key driver for this de­vel­op­ment is the mi­gra­tion of trans­mis­sion speeds from the cur­rent 40 and 100 Gb/s to the next gen­er­a­tion 200 Gb/s and 400 Gb/s fi­bre ap­pli­ca­tions and from 8 Gb/s, to 16 Gb/s and to 32 Gb/s for fi­bre chan­nel-based SANS and as band­width and speeds in­crease, in­ser­tion loss re­quire­ments be­come much more strin­gent.”

Au­to­mated In­fra­struc­ture Man­age­ment (AIM) is an es­sen­tial tool that can re­alise ben­e­fits for the owner/op­er­a­tor at var­i­ous lo­ca­tions across their data cen­tre op­er­a­tion. AIM can reduce timein­ten­sive man­ual pro­cesses, min­imise hu­man er­ror and net­work down­time. It can also of­fer vis­i­bil­ity into end-to- end cir­cuits. And, in the event of a net­work fail­ure, a root cause analysis can be quickly es­tab­lished to help the net­work to get back on­line fast.

As higher fi­bre den­si­ties and net­work vir­tu­al­i­sa­tion be­come preva­lent at the core, the chal­lenge of manag­ing those as­sets also in­creases. Waite adds that sin­gle op­ti­cal fi­bre can no longer be as­so­ci­ated with a sin­gle busi­ness process, with vir­tu­alised servers the data pack­ets run­ning across a chan­nel may be linked to many dis­parate el­e­ments of a busi­ness or even dif­fer­ent cus­tomers in a cloud type de­ploy­ment.

Edge data cen­tres can be sit­u­ated in re­mote lo­ca­tions, tak­ing a de­cen­tralised ap­proach, or placed in ur­ban ar­eas. These edge data cen­tres would be smaller in size, but their num­bers would be higher. The edge data cen­tres need to be read­ily avail­able and ac­ces­si­ble for op­er­a­tion and main­te­nance pur­poses. Fur­ther, 5G will re­quire the sup­port of many edge data cen­tre lo­ca­tions to sup­port low la­tency ap­pli­ca­tions.

Manag­ing and main­tain­ing such a large num­ber of data cen­tres poses a sig­nif­i­cant chal­lenge, ac­cord­ing to Yousef. “This re­quires the im­ple­men­ta­tion of new prac­tices on work or­ders for moves, adds, changes and fault de­tec­tion. Faster MTTR would be a crit­i­cal KPI for edge data cen­tres.”

At the edge, AIM can be used to re­motely track the util­i­sa­tion of pan­els, ca­bling and switch ports, and pro­vide a real-time view on how phys­i­cal-layer as­sets are be­ing used. Such a sys­tem also of­fers numer­ous ex­trin­sic ben­e­fits, thanks to its abil­ity to be com­bined with other man­age­ment tools, thus en­hanc­ing the se­cu­rity of edge data cen­tres.

In ad­di­tion to fa­cil­i­tat­ing the man­age­ment of in­creas­ingly large and com­plex in­fra­struc­tures, there are other ben­e­fits to the use of AIM so­lu­tions, ex­plains Khalil. “Us­ing a sin­gle, cur­rent, con­sis­tent data­base elim­i­nates stranded ca­pac­ity and fa­cil­i­tates end-to-end analysis, ag­ile in­fra­struc­ture man­age­ment, pre­dic­tive analysis, ca­pac­ity util­i­sa­tion and ef­fi­ciency of op­er­a­tion and ad­min­is­tra­tion, and can bring 30 – 50% re­duc­tion in down­time. Sys­tem data can be used for bud­get­ing and IT in­fra­struc­ture in­ven­tory.”

Fu­ture steps in AIM in­clude the use of ar­ti­fi­cial in­tel­li­gence to en­sure in­fra­struc­tures re­main man­age­able. “AI can in­de­pen­dently man­age con­nec­tiv­ity from the data cen­tre to, for ex­am­ple, a smart city, mak­ing pre­dic­tions based on mon­i­tor­ing and ma­chine learn­ing,” adds Khalil.

New fi­bre net­work topolo­gies

The num­ber of fi­bres used for trans­port is on the in­crease. 40 and 100 Gbit/s re­quire eight fi­bres in par­al­lel pairs. Fur­ther­more, the tra­di­tional hi­er­ar­chi­cal net­work topol­ogy with core, ag­gre­ga­tion and ac­cess level can no longer cope with to­day’s needs, re­sult­ing in con­ges­tion along traf­fic routes. To en­sure that data and ap­pli­ca­tions are avail­able in real-time at all times la­ten­cies have to be con­sid­er­ably re­duced, call­ing for sin­gle­mode or multi-mode fi­bre and new kinds of net­work ar­chi­tec­ture.

Spine and leaf, sup­ported by a net­work fab­ric is the next-gen­er­a­tion topol­ogy be­ing de­ployed to re­place core, ag­gre­ga­tion and ac­cess level switch­ing. A spine-leaf ar­chi­tec­ture re­duces la­tency and can be adapted to con­tin­u­ously chang­ing needs.

Waite adds that this is the most ef­fec­tive way to en­able east-west data flow in the data cen­tre. “East-west flow of traf­fic de­scribes the lat­eral flow of data across the networking lay­ers within a data cen­tre, al­low­ing servers to quickly and ef­fi­ciently com­mu­ni­cate with one an­other in sup­port of AI, ma­chine learn­ing, vir­tual re­al­ity and a plethora of other high band­width low la­tency ap­pli­ca­tions.”

A net­work mesh with criss­cross­ing ca­bling guar­an­tees that switches at ac­cess level are no longer more than a hop away from each other. All de­vices are the same num­ber of seg­ments. Un­like the clas­sic three-tier ar­chi­tec­ture, this new topol­ogy has just two lay­ers. The leaf layer is built up of ac­cess switches that con­nect to servers, edge routers, load bal­ancers, fire­walls and other de­vices. The net­work back­bone is pro­vided by the spine layer, which con­sists of rout­ing switches.

“Meshed leaf-spine ar­chi­tec­tures, in­creas­ing fi­bre den­sity, com­plex point-to­mul­ti­point con­nec­tions, and the fact that de­vices can be added or re­al­lo­cated at any time make know­ing the ex­act state of all con­nec­tiv­ity el­e­ments at all times ex­tremely chal­leng­ing. Chang­ing net­work topolo­gies, there­fore, also war­rants the use of AIM so­lu­tions,” says Khalil.

The con­se­quence of de­ploy­ing this topol­ogy is that ev­ery ca­bling chan­nel within the fab­ric, con­nect­ing the spine and leaf switches, must be able to sup­port pre­cisely the same through­put of data. There­fore, each ca­bling chan­nel must be de­signed to sup­port the max­i­mum ex­pected data rate; noth­ing less is ac­cept­able. Com­monly the fab­ric data rate used to be rated at 40GBE amongst many data cen­tre own­ers. To­day this rate no longer ap­plies with many cus­tomers only con­sid­er­ing 100Gb/s and above as the min­i­mum stan­dard for their fab­ric ca­bling go­ing for­ward.

Arafat Yousef, manag­ing direc­tor MEA, Nex­ans Ca­bling So­lu­tions.

Na­bil Khalil, ex­ec­u­tive vice pres­i­dent of R& M META.

Prem Ro­drigues, direc­tor for the MEA & In­dia/ SAAR, Siemon.

Alas­tair Waite, data cen­tre market de­vel­op­ment, Comm­scope.

Newspapers in English

Newspapers from UAE

© PressReader. All rights reserved.