> From: Michael Ryder [mailto:mryder1...@gmail.com] > Sent: Sunday, February 9, 2014 10:48 PM > > Yes, it's enterprise -- a pair of HP EVA arrays. We're talking to them via > 4Gb > fibre-channel. Single-initiator zones. There's a combination of RAID levels > going on on the arrays.... but it all used to perform so *well*. > The connection goes like this > EVA --> > (4Gb FC) --> > Windows Backup Proxy Server --> > (1Gb Eth, TCP port 1500) --> > Windows Backup Server (TSM)
What kind of performance does the Proxy server get, performing IO directly to the EVA, just simply reading/writing disk as fast as possible? It sounds to me, like your new switches are buffering data, and delaying the "data has been flushed" signal. Which would be good for most things, and in fact, which I would call "broken" because the proper behavior would be to correctly deliver the signal when requested. As you said, the "correct" fix for such a problem would be a driver/firmware upgrade. But anyway - Counterintuitively, you might benefit by *disabling* write-back and/or buffering. This is analogous to a surprising finding I had in ZFS: You actually *hurt* zfs performance, if you have a pool of HDD's and SSD's in the backend, with HBA write-back enabled. Performance is improved by disabling the write-back (Unless your pool is purely a bunch of HDD's, in which case, the write-back helps.) _______________________________________________ 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/
