> From: opensolaris-discuss-boun...@opensolaris.org [mailto:opensolaris- > discuss-boun...@opensolaris.org] On Behalf Of valrh...@gmail.com > > Everything is working pretty well, and over NFS, I can get a solid 80 > MB/sec if I'm copying big files. This is adequate, but I am wondering > > of the dedup table. I ran zilstat.ksh, but it always came up with > zeros, which suggests there's no point in a ZIL log SSD. > > Is there anything left to tune? If so, how do I go about figuring out > how to increase performance? Right now, I'm just copying large files
As you've seen, even if you didn't have optimized disks, usually the network is the bottleneck. You can almost max out a Gb ethernet with a single disk, and you can certainly max out the Ether with 2 or more disks. 80MB/s = 640 Mbit/sec, which is a sizeable portion of the Gb ether. Normally you're doing well to get anywhere from 70-90% efficient use of an Ethernet. I suspect your calculation (640Mbit/sec) was at the user level, right? You did not measure that with wireshark or whatever; you did not account for NFS and TCP overhead, right? In which case, you're actually seeing a solid 700Mbit or higher on the wire. And that's evidence you won't get any higher on that wire. The next thing to do is to start thinking about how to accelerate your net. You might go with 10G ether, but that's expensive. You might think about link aggregation (lacp). My experience with lacp is that it frequently causes packet loss, which hurts connections more than it helps; so if you go with lacp, be really really sure to hammer on it, and watch for packet loss (error count increasing on the interface.) Or the 3rd option is add another interface, with another IP address, and some clients connect using IP #1, and some clients connect using IP #2. Etc. _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
