How to over­come the top 5 server work­load con­straints

and get more out of your IT bud­get

DEMM Engineering & Manufacturing - - GOOD BUSINESS - *Sur­vey con­ducted by Spice­works in 2016.

Server mem­ory is a com­po­nent that’s ei­ther suf­fi­cient or in­suf­fi­cient. If you have a suf­fi­cient amount of mem­ory for your work­loads, you likely don’t even think about RAM be­cause other prob­lems con­sume your at­ten­tion. How­ever, when you have in­suf­fi­cient mem­ory, your servers and your or­gan­i­sa­tion’s pro­duc­tiv­ity slow to a crawl be­cause DRAM feeds your CPUs. That’s why in a re­cent Spice­works sur­vey of over 350 IT de­ci­sion-mak­ers, 47 per­cent noted that they planned to add more server mem­ory in the com­ing year, even though half of all their servers were al­ready run­ning at the max­i­mum in­stalled mem­ory ca­pac­ity.* These find­ings come as no sur­prise be­cause of how mem­ory helps over­come five of the most press­ing work­load con­straints.

Michael More­land, Cru­cial Server DRAM Prod­uct Mar­ket­ing Man­ager said, “We part­nered with and com­mis­sioned the Spice­works sur­vey to see and hear what chal­lenges IT pros are ex­pe­ri­enc­ing in their data cen­tres. Based on what they’ve said, it’s clear that they be­lieve server mem­ory is crit­i­cal for im­prov­ing sys­tem per­for­mance. Adding more mem­ory to servers im­proves CPU per­for­mance and ef­fi­ciency, which ul­ti­mately helps al­le­vi­ate the top five work­load con­straints they men­tioned: limited bud­get, un­ex­pected or un­pre­dictable work­load de­mands, limited floor space, rapid growth in user base, and power or cool­ing costs.”

Spice­works asked, “What are the top chal­lenges you cur­rently face in over­com­ing server work­load con­straints?” and re­spon­dents were able to se­lect up to three an­swers.

The 353 re­spon­dents se­lected by Spice­works were re­quired to have pur­chase in­flu­ence in their or­gan­i­sa­tion and were re­quired to have at least 30 physical servers and be us­ing vir­tu­al­i­sa­tion soft­ware. Over­all, 23 in­dus­tries were rep­re­sented (rang­ing from tech­nol­ogy to en­ergy to man­u­fac­tur­ing) and 74 per­cent of re­spon­dents were run­ning 100 or more physical servers, with 41 per­cent run­ning over 200 boxes.


More­land added, “When we sur­veyed the 350- plus IT man­agers from around the world, they listed a limited bud­get as their top work­load con­straint. Mak­ing the most of scarce re­sources is a hall­mark of mod­ern IT, and that’s why it’s crit­i­cal to keep the to­tal cost of own­er­ship ( TCO) down. That’s where adding server mem­ory can help – max­ing out a server’s mem­ory pro­vides fuel for the CPU to run op­ti­mally, al­low­ing you to use fewer servers to ac­com­plish more. With each server func­tion­ing more ef­fi­ciently, it lim­its the power, cool­ing, and bur­den­some li­cens­ing costs that come with having more servers in your server room.”

Get­ting the most out of dwin­dling bud­gets of­ten comes down to com­par­ing ac­qui­si­tion

cost to TCO. If you in­crease a server’s ef­fi­ciency, you de­crease its TCO be­cause you’re get­ting more per­for­mance out of it over the same amount of time. Since mem­ory is what feeds pro­cess­ing cores, it’s one of the most ef­fec­tive ways to make your CPUs more ef­fi­cient and pro­duc­tive, al­low­ing you to han­dle grow­ing work­loads with­out having to buy new servers.

Specif­i­cally, more mem­ory gives your sys­tem more of its fastest re­source to get data to the CPU. And the faster data gets to the CPU, the less time it spends idling, con­sum­ing power, and do­ing lit­tle to no work. Since mem­ory re­sides closer to the CPU, it takes less time for data to get from DRAM to the CPU than is does to go from stor­age to the CPU. For ex­am­ple, hard drives typ­i­cally get data to the CPU in mil­lisec­onds, ver­sus en­ter­prise SSDs which get data to the CPU in mi­crosec­onds. This is cer­tainly a vast im­prove­ment, but it’s still a higher la­tency than DRAM, which is able to get data to the CPU in nanosec­onds (with la­tency, lower is of course bet­ter). When you con­sider the mil­lions of in­struc­tions that are fed to the pro­ces­sor each day, feed­ing data to the CPU via mem­ory de­liv­ers a sig­nif­i­cant per­for­mance dif­fer­ence. Time is money, and more mem­ory helps de­liver the best pos­si­ble re­turn on your CPU in­vest­ment.


Vir­tu­alised work­loads are all about main­tain­ing con­sis­tent qual­ity of ser­vice (QoS) and elim­i­nat­ing on-again/off-again vari­ance. Over­all, more RAM helps elim­i­nate ser­vice vari­ance be­cause it pro­vides ex­tra re­sources for vir­tu­alised ap­pli­ca­tions to store and use ac­tive data (which lives in mem­ory). Since un­pre­dictable work­load spikes quickly ex­haust avail­able mem­ory, the sys­tem scram­bles to find avail­able re­sources, per­for­mance drops, and disk thrash is typ­i­cally the re­sult. More mem­ory solves this prob­lem by giv­ing your ap­pli­ca­tions more flex­i­bil­ity to meet ris­ing and fall­ing work­loads that like to spike.


Think of floor space lim­its as a con­struc­tive prob­lem to solve: What’s the min­i­mum amount of servers you need to ac­com­plish your work­load? This kind of think­ing helps lighten your en­ter­prise foot­print be­cause ev­ery server that’s un­der­utilised costs you more. For ex­am­ple, if you used five maxed­out servers to ac­com­plish the work­load of 10 half-full or old servers, you’d save on power, cool­ing, and soft­ware li­cences – the big killer. When floor space is at a pre­mium, there’s re­ally only one thing to do: scale up. Scal­ing up al­most al­ways in­volves in­creas­ing a server’s in­stalled mem­ory ca­pac­ity to get as much out of the box and feed as many VMs as pos­si­ble.

More­land con­tin­ued, “Vir­tu­alised ap­pli­ca­tions are heav­ily de­pen­dent on ac­tive data, and when there’s a spike in work­load ac­tiv­ity, avail­able server mem­ory re­sources are de­pleted and QoS drops. Fill­ing your servers to their max­i­mum RAM ca­pac­i­ties re­duces the strain on the sys­tems when ac­tiv­ity gets in­tense. Limited square footage in your server room places a pre­mium on mak­ing the most of each in­di­vid­ual server. When you don’t have the physical space to scale out, scal­ing up by max­ing out the mem­ory of each server can ul­ti­mately match the per­for­mance of mul­ti­ple half-full ones. Fewer servers and equal per­for­mance equates to less TCO be­cause your en­ter­prise foot­print is smaller and you don’t have to spend big on huge li­cens­ing costs.”


Host­ing more users re­quires more mem­ory to main­tain the same QoS. By giv­ing the sys­tem more RAM, you gain more flex­i­bil­ity and in­crease its abil­ity to han­dle un­pre­dictable work­load de­mands caused by the sud­den growth in your user base.


Al­though fully pop­u­lat­ing the mem­ory in a server raises its to­tal power con­sump­tion, the to­tal con­sumed en­ergy is of­ten less than us­ing mul­ti­ple par­tially full servers to de­liver a com­pa­ra­ble level of per­for­mance. More DRAM helps your servers use power in the most ef­fi­cient man­ner pos­si­ble from a work­load per­spec­tive (feed­ing and run­ning the CPU). Plus, if you’re us­ing fewer physical servers, your to­tal power and cool­ing costs will likely be less.

The bot­tom line is that mem­ory is like fuel for your CPUs – as long as they have enough of it, they’re OK. But there’s a sig­nif­i­cant dif­fer­ence be­tween having enough RAM and truly im­prov­ing work­load ef­fi­ciency. With just enough RAM, you’re cer­tainly able to run ap­pli­ca­tions, but with the max­i­mum in­stalled mem­ory ca­pac­ity, you’re of­ten able to use fewer servers to get more done at a lower to­tal TCO. Don’t starve your CPUs. Know your work­load, and if it’s CPU or mem­ory-de­pen­dent, im­prove ef­fi­ciency for less with more RAM, not more servers.

More­land con­cluded, “Power and cool­ing costs are di­rectly re­lated to the num­ber of servers in your de­ploy­ment. Having more servers con­sumes more power and pro­duces higher tem­per­a­tures than a room with fewer servers – not to men­tion how much it raises your soft­ware li­cence costs. In the case of servers, more isn’t al­ways bet­ter, es­pe­cially when they can be maxed-out with mem­ory and meet or sur­pass the same level of per­for­mance of half-full servers. Fo­cus on qual­ity, not quan­tity, when it comes to your server de­ploy­ment and re­duce your power, cool­ing, and li­cence costs.”

Newspapers in English

Newspapers from New Zealand

© PressReader. All rights reserved.