I think most of the focus on cloud computing is based around raw processing power. They forget that all that data needs to be stored somewhere at some point. I think the challenge for drizzle workloads will be finding the correct balance of cpu, memory, i/o latency and power consumption. Most of these cloud computing discussions forget all about i/o latency. We can't.
On Mon, Mar 2, 2009 at 11:59 AM, Martin Scholl <[email protected]> wrote: > Hello all, > > > I like to underline what James Hamilton describes. > We have done some tests with Intel Atom boards recently and our results > are quite promising. > With a $60 board (Atom dualcore 1.6Ghz) you get about 1/3 the "raw" CPU > performance of an state-of-the-art Core2Duo (2.66ghz) (speaking of the > performance of a single core) -- but the Atom uses 8W to achieve this > instead of ~120W of the Core2Duo (!). > There are already "mini-clusters" of said Intel Atom CPUs on the way. > > They will change the way we do computing IMHO. > > > Just my 2 cents, > Martin > > Jim Starkey wrote: >> James Hamilton, once of Microsoft/Live, now Amazon, has a paper well >> worth reading: >> http://mvdirona.com/jrh/TalksAndPapers/JamesHamilton_CEMS.pdf >> >> He argues (among many other things) that cores per system is increasing >> much faster than memory bandwidth, resulting in cores stalled on memory >> access, effectively canceling some or all of the potential gain of >> additional cores. Following this line of reasoning, he argues that more >> cheaper, "balanced" servers have more bang per capital and energy buck >> than more expensive servers. This is a pretty good counter to Brian's >> argument on very high number of cores. The logical extension of >> Hamilton's argument is a server sled with a single power supply and six >> expendable, low cost, low power servers in a 1U package. >> >> The software architecture ramifications are pretty serious. If Hamilton >> is right, scale up is all but dead. Cycles per energy buck favors >> slower processors that balance with memory bandwidth and where >> reliability comes from software, not hardware. >> >> Anyone want to take on the implications for drizzle? >> >> >> > > > _______________________________________________ > Mailing list: https://launchpad.net/~drizzle-discuss > Post to : [email protected] > Unsubscribe : https://launchpad.net/~drizzle-discuss > More help : https://help.launchpad.net/ListHelp > -- Eric Bergen [email protected] http://www.provenscaling.com _______________________________________________ Mailing list: https://launchpad.net/~drizzle-discuss Post to : [email protected] Unsubscribe : https://launchpad.net/~drizzle-discuss More help : https://help.launchpad.net/ListHelp
