> From: tech-boun...@lists.lopsa.org [mailto:tech-boun...@lists.lopsa.org] > On Behalf Of Michael Ryder
Ok - I don't think it's been clear, what you have connected to where, and what type of storage (or what type of storage operation) you're trying to do... You have some kind of storage - presumably enterprise grade SAN storage array, with hardware raid controller, in raid6 I guess? It is attached via the aforementioned redundant FC network, to some backup server, with redundant HBA's. Right? And it's all single initiator, so there is no competition from other initiators competing for storage performance. Right? The backup server is connected via 1Gb ether, to some other servers. You're pushing backups from those other servers, across the ether, and ultimately onto the SAN storage. Since the FC is supposed to be much faster than the 1GB ether, you would expect the ether to be the bottleneck. 800 Mbit is suboptimal, but kind of close to acceptable... So it's weird to be encountering this limit around 350 Mbit. Weirder still, on the backup server, you toggle the dominance of the HBA's, the performance suddenly jumps to 800Mbit, but only temporarily. Have I got all that right? So... What protocol are you talking over the ethernet? Because some protocols, such as iscsi, would very likely perform blocking writes, waiting for data to flush out to disk - In which case, the SAN configuration and raid configuration could truly make a big difference. You would want to make sure you have some sort of write-back cache enabled. Also, supposing you have raid5 or 6 or similar, if you're doing any sort of random IO (rather than a single sustained sequential stream) then raid5 or 6 or similar would very likely perform poorly, which could hit you if your job is waiting for buffer flushes out to disk. You're definitely wise to approach firmware & driver updates. There's a very good chance that will make a difference. Even if you can't use dd, you should be able to measure the IO performance of the backup server talking directly to the backend storage. Either perfmon, or timing how long a large file copy takes (as long as you have some way of clearing the system cache in between, which could mean rebooting the server, or something like that.) Also, you actually can use dd if you install cygwin. But I understand that might be undesirable. _______________________________________________ Tech mailing list Tech@lists.lopsa.org https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech This list provided by the League of Professional System Administrators http://lopsa.org/
