Issue link: https://cp.revolio.com/i/709701
THE DATA CENTER JOURNAL | 19 www.datacenterjournal.com F or a company operating a single traditional data center, particularly if that facility hosts business-critical operations, availability is everything. at data center as a whole (barring, for instance, any backup facili- ties) represents a single point of failure, so the design phase will generally focus heavily on achieving as high an availabil- ity rating as practical given the budget. But a company that operates a second facility, particularly if it has some failover mechanisms in place, has a backup in case one data center suffers an outage. A third or fourth facility creates even more resiliency. Availability concerns therefore ease as resources are distributed among multiple data centers. Because of growing demand for ser- vices thanks to proliferation of mobile de- vices and the looming Internet of ings, providers are facing a bandwidth shortfall. Moreover, the trend of IT centralization resulting from greater reliance on cloud computing and industry consolidation into fewer large corporations means that for a given locality, the nearest data center providing a particular service may be far away. Latency is thus a concern, especially for interactive applications that require fast responses from the provider. is situ- ation has expanded interest in "edge" data centers, which reside closer to customers than the typical main, central data center. Cost and practicality limit the size of these edge data centers, however—they aren't simply replicas of a company's primary multi-megawatt facility. And they need not be, since they are only intended to serve a fraction of the customer base. Hence the use of so-called micro data cen- ters: relatively small deployments designed to reduce latency, increase bandwidth and better serve customers through proximity. By their very nature, they are decentral- ized, so an important question is to what extent availability is a concern when preparing a micro data center plan. reliaBiliTy: one and Many Ignoring backup infrastructure such as a remote site or another approach, a company operating a single data center has a particular availability rating for that facility—say, 99% for the sake of this dis- cussion. at means on average (assuming the value is accurate), the company's IT re- sources will be unavailable 1% of the time. e data center in this case represents a single point of failure. Consider also the case of multiple micro data centers disbursed throughout the company's service area. Say, for ex- ample, the company has 10 of these small facilities scattered around, and in a pinch, 8 of them can carry the entire workload if 2 suffer an outage. Now, let's say each one of these micro data centers has an avail- ability rating of just 90%, meaning it will on average be offline 10% of the time. e company won't run into trouble unless three of the micro data centers go offline. Doing the math for the probabilities, that's a 10% chance for any one being offline, so multiply the chances for all three: 0.1 x 0.1 x 0.1 = 0.001. at means the chance of 3 out of 10 data centers being down is 0.1%, for an overall availability of 99.9%. at's much better than in the case of a single data center with a 99% availability rating. (A thorough reckoning would need to take into account the chance of four being down, five being down and so on, but these events are relatively unlikely.) Of course, these numbers are just for illustration. A large data center may have a much higher availability rating, and it may have backup or failover capabili- ties in the event of an outage. Moreover, a smaller fraction of micro data center outages may be problematic, depending on their individual availability ratings and ability to handle extra workloads if others go offline. But the main point is that given some flexibility in these areas, a deploy- ment of micro data centers with lesser availability can provide services just as reliably as a single, large, highly available data center. More or less availaBle Naturally, a number of factors will determine the availability of each micro data center. One is how much interac- tion it will have with company personnel. Because many downtime incidents are the result of human error, a highly automated micro data center is desirable. Moreover, when numerous instances are deployed at the customer edge—which may span a large geographical area—maintaining a contingent of employees at each site may be cost prohibitive. To meet demand for these facilities, a number of suppliers offer self-contained micro data centers that provide the power- distribution, cooling and fire-suppression Naturally, a number of factors will determine the availability of each micro data center. One is how much interaction it will have with company personnel. Because many downtime incidents are the result of human error, a highly automated micro data center is desirable. Moreover, when numerous instances are deployed at the customer edge—which may span a large geographical area—maintaining a contingent of employees at each site may be cost prohibitive.