Steve

Abstract

As computation continues to move into the cloud, the computing platform of interest no longer resembles a pizza box or a refrigerator, but a warehouse full of computers. These new large datacenters are quite different from traditional hosting facilities of earlier times and cannot be viewed simply as a collection of co- located servers. Large portions of the hardware and software resources
in these facilities must work in concert to efficiently deliver good levels of Internet service performance, something that can only be achieved by a holistic approach to their design and deployment. In other words, we must treat the datacenter itself as one massive warehouse-scale computer (WSC). We describe the architecture of WSCs the main factors influencing their design, operation, and cost structure, and the characteristics of their software base. We hope it
will be useful to architects and programmers of today's WSCs, as well as those of future many-core platforms which may one day implement the equivalent of today's WSCs on a single board.

Full text available on-line

http://dx.doi.org/10.2200/S00193ED1V01Y200905CA…

Steve

If I'm incorrect I would love to know, but it would help to understand where such a strong assertion comes from.

Good question and one that there has been much debate about on this site. Small switched mode power units such as those used in servers are actually not too bad, achieving between 92% and 96% efficiency at full load. Remember that they are never run at full load as to provide resillience the PSUs can never get past 50% to work in an N+N pair. Sometimes there are three or four PSUs in a blade chassis and these can get to 60% or 70% loading.

Large scale DC power supplies can achieve very good efficiencies but one could argue about the 2-3% difference as being too insignificant to measure (frankly I agree). The real saving in capital and operational costs is in missing out the UPS stage. The efficency loss of converting from DC off the batteries back to AC again can be as much as 15%. Remember that UPS kit also has to be run in resillient configuration of N+N or N+1 so cannot run at full efficency and full load.

Thanks

Steve

Lets just agree to disagree. None of the big boys use server-scale computers anymore. Pizza box servers don't deliver the degree of efficiency that folks need when they have tens of thousands of computing elements in a data center. Why have tens of thousands of power supplies when you can have tens and whu use server scale fans which are inneffficient when you can have industrial scale ones?

Steve

Pizza boxes aren't going away, although they are likely to change.

Google do the same as BT do in their 21st Century Data Centers. Local rack based DC power supplies. Yes they get AC delivered to make the 12V but that AC is unprotected. Google have ripped out the UPS and battery room stage, delivering low cost data center scale DC power at the rack level. That takes hundreds of millions of dollars out of their data center capital costs and improves both PUE and reliability at the same time.

Google go AC – DC stop.

20th Century Data Centers go AC – DC – AC – DC because they have a server-scale power supply and need protected ac. Just bonkers in my humble opinion – and I have been saying so for years.

Steve

Blade-scale goes one better with some shared components, perhaps larger fans and shared Power supplies. A larger cabinet.

Data Center Scale throws away the single boxes and fans and power supplies and scales them up. Great big efficient fans, huge DC power units. Better airflow. Maybe container sized like Microsoft with their Verari FOREST containers.

That is what is different and the word server confuses our thinking hence my new categories.

Steve