On Sun, Sep 06, 2009 at 10:56:16AM -0400, Edward Ned Harvey spake thusly: > Do you have any iozone, or some other benchmark to support that? I have > always had the understanding that even a single local disk can nearly > sustain 1Gbit/sec, with bursts that are higher thanks to cache, while over > the network, due to network overhead, the most you can get is about 800Mbit > from a 1Gbit connection... Especially when that 1Gbit connection is shared > by a bunch of machines...
I have done such tests but I don't have them at hand to share with you at the moment. Expensive single disks such as 15k SAS can do 1Gb/s. But your typical 7.2k RPM SATA does 50-70 it seems. I will see if I can dig some up. The nice thing about AoE is that there is little to no network overhead. It is not TCP nor IP. It runs purely at layer 2. > Here is one typical 500Mb 7200rpm sata disk, which is able to sustain over > 1Gbit/sec throughput: Hmm..Hitachi marketing material...I'll believe it when I have a physical disk in my hands and can benchmark it. However, note that the "Media transfer rate (MBits/sec max)" is 1138. This is surely taken at the outer edge of the platter. What would the transfer rate at the inner edge be? Average them and you might have a more reasonable number with which to compare. > PS. 2G of ram in a server doesn't sound like much to me. I never buy any > laptop with less than 4G anymore. The 2G I was refering to is in the SAN head. Not the server running the virual machines. The 2G in each SAN head is there just to run the kernel, vblade, and provide caching. RAM is so cheap that future machines may have 4 or 8G depending on how this affects the price/performance. RAM is so cheap now that I expect we will just go 8G. > Yes, all the compute nodes will be working in parallel, that's the whole > point. They will all be writing to local scratch space, and consuming a lot > of cpu and ram at the same time. That is a rather different use case than mine in which case you probably have no choice but to shell out very big bucks for fiberchannel or better. Fortunately I have a much more even workload where no one particular machine has terribly great performance demands and it all overages out such that I don't have a bunch of machines hitting the SAN all at the same time. I also stagger backups, updatedb, and other similar things wherever possible to ensure maximum performance. -- Tracy Reed http://tracyreed.org
pgpxKfhnV1CjZ.pgp
Description: PGP signature
_______________________________________________ Tech mailing list [email protected] http://lopsa.org/cgi-bin/mailman/listinfo/tech This list provided by the League of Professional System Administrators http://lopsa.org/
