The freak lightning strike blamed for knocking out European cloud operations at both Microsoft and Amazon earlier this week was apparently not guilty. Data Center Knowledge is now reporting Ireland’s Electricity Supply Board as pinning the outage on equipment failure, not weather; “our initial assessment of lightning as the cause has now been ruled out,” according to an EBS statement. This should lead customers to ask far harder questions of both Amazon and Microsoft than they have so far. An outage caused by freak weather is possibly understandable, but for routine failure of equipment at one utility power provider to bring down two flagship cloud data centers is ridiculous. Where were the surge protectors? Why couldn’t the backup generators cope with a fluctuation in voltage? Where was the secondary source of mains power? And days later, a few of Amazon’s customers are still feeling the effects; the AWS Status site reported that, as of 01:10 Pacific, “Less than 2% of the affected volumes are still in the process of having their recovery snapshot created.” Not good. Not good at all.
The European cloud operations of both Amazon and Microsoft were yesterday affected by a lightning strike on equipment belonging to the Irish utility company responsible for providing electrical power to both Dublin data centers. Rich Miller at Data Center Knowledge quotes Amazon’s status report at the time, providing an explanation as to why backup generators on-site failed to do their job. Storms happen, and electrical equipment will get damaged by lightning, but it must be embarrassing for both Amazon and Microsoft that neither backup generators nor alternative power suppliers were able to prevent a prolonged outage. And who thought it was a good idea for two of Europe’s largest data centers to rely upon electricity flowing through a single location? Services at both data centers are gradually being restored, although it may be a few days before everything is back online. Even my music is affected.
A new report by Jonathan Koomey argues that data centers are using a lot less power than we thought. 1.1%-1.5% of global electricity use (1.7%-2.2% in the U.S.) is still an awful lot, but Koomey’s research suggests that global electricity consumption in data centers “only” increased by 56% in the five years to 2010, instead of doubling as predicted. As John Markoff notes in today’s New York Times, the recession’s downward pressure on demand combines with innovations in power management to explain the lower than predicted figure. GigaOM’s Katie Fehrenbacher has more, and links to a recent GigaOM post from Koomey which sparked some useful comments. Elsewhere, Dean Takahashi reports on a new research project involving AMD, HP, Clarkson University and the New York State Energy Research and Development Authority (NYSERDA). Project partners intend to explore the feasibility of powering containerised data center pods using wind or solar power. If successful, this work overcomes one serious hurdle to deploying data center pods off-grid. Now someone else just needs related projects to cool the box, and give it sufficient bandwidth.
A paper co-authored by researchers from Facebook and multicore chip specialist Tilera reports that servers comprising large numbers of low-power processor cores deliver clear performance and power advantages in certain circumstances. Dean Takahashi at VentureBeat focuses on the processors being “four times more energy efficient,” whilst Rich Miller at Data Center Knowledge reports that they “boost memecached efficiency” over alternatives with fewer cores. Stacey Higginbotham sees the work as validation of the claims being made by companies such as Tilera, with their alternative chip designs. She also points to the need for more generic benchmarks with which to compare the performance of different architectures, especially as the hardware configuration for a single massive database may need to be radically different from that used in supporting other usage requirements.
Intel has reported record quarterly revenues of $13.1 billion, slightly higher than analyst expectations. Customers continue to spend on data center equipment, with revenues for Intel’s data center group up 15%. However, despite a lot of market interest in low power Atom chips, revenue in that area was down 15% on last year. Rik Myslewski at The Register blames “the netbook-cannibalization effect of Apple’s iPads” for the fall. Quoted in the New York Times, Evercore Partners’ Patrick Wang suggested that “I fully believe that it is the data center — the cloud — that is driving Intel.” Apple, Google, Intel and others have all exceeded market expectations in the past few weeks, but it may still be too soon for complacency. Strong numbers mask the reality of large companies working hard to adjust to new customer requirements whilst building processes that will see recent growth sustained.
Low-power computing specialist SeaMicro has shipped a new server featuring 768 cores, a significant increase on the 512 cores of the previous product released just a few months ago. GigaOM’s Stacey Higginbotham notes that the performance boost is accompanied by 25% reduction in power consumption. Over at VentureBeat, Dean Takahashi reports that the 17.5″ (44.5cm) high box “can replace… 60 traditional servers, four rack switches, four terminal servers, and a load balancing server. It uses a quarter of the power and a sixth of the space.” SeaMicro currently uses processors provided by Intel, although Larry Dignan picks up on the company’s intention to be “processor agnostic,” noting that ARM-powered servers should follow next year. SeaMicro’s innovations around increasing processor density whilst reducing power consumption sound impressive, but these servers aren’t yet for everyone. RAM configurations and — especially — the relatively low power of these “wimpy cores” mean that today they’re best-suited to serving high volumes of reasonably quick, reasonably simple tasks such as delivering web pages.
IBM is due to announce two new Japanese data centers this morning, one focused on delivering the company’s SmartCloud in the region, and the other supporting LotusLive customers. Elsewhere in the Asia Pacific market, The Australian reports that Amazon is scouting for data center sites in and around Sydney, with a view to offering additional capacity to supplement the region’s existing Amazon Web Services region in Singapore.
Chris Evans highlights this week’s news from Hitachi Data Systems, which sees the storage company begin to market a “converged infrastructure” solution. The Unified Computing Platform will see HDS enter a space that HP has been striving to make its own; a space in which customers can move up and down through the product range as their requirements change. However, as Chris outlines, the information released by HDS to date prompts far more questions than it answers.
The abundant geothermal energy that Iceland is using to power its new breed of data centers comes at a cost; the sub-surface activity that generates all that hot water also makes the area prone to volcanic eruptions. Grímsvötn erupted last month, bringing darkness to Rekjavik’s normally bright summer skies and causing some limited disruption to European air traffic. However, according to those behind the island’s Thor Data Center, it kept on operating as normal. They’ve even provided pictures of their air filters to demonstrate how little effect the volcano had on air near their facility.
This week, the Open Data Center Alliance released its first deliverables to the world, with the aim to make it easier to compare and contrast commercial solutions and increase technical interoperability between clouds. But to reach new standards, the alliance must overcome significant challenges.