Looks like you could use some kernel tuning (;-)Here's the link to the DTrace toolkit:http://www.brendangregg.com/dtrace.htmlAlso, what you can do is to run your typical NFS workload, and atthe same time, run:iostat cxnz 1on the NFS server. Disregard the very first result set you get, that's thecumulative total and it's not relevant for these measurements.Then post 3-4 result sets (with "1", you should be getting a perf snapshotonce every second). Date: Fri, 29 Jan 2010 17:02:21 +0100 Subject: Re: [ug-chosug] [NFS] Zfs sharenfs bad performance From: [email protected] To: tripivceta at hotmail.com CC: ug-chosug at opensolaris.org
Actually as i didn't have any other *nix client i local mounted the nfs share on the server and performance a just a little bit better than from OSX client... So my guess is that the problem doesn't come from the OSX client as i thought first. DO you have a link to that DTrace toolkit? All i checked with iostat and nfsstat doesn't show anything relevant Nothing strange except that: Jan 29 09:19:24 local at dismiss /usr/lib/nfs/nfsd[879]: [ID 739773 daemon.debug] Current SO_SNDBUF value is 49152 Jan 29 09:19:24 local at dismiss /usr/lib/nfs/nfsd[879]: [ID 978932 daemon.debug] Set SO_SNDBUF option to 1048576 Jan 29 09:19:24 local at dismiss /usr/lib/nfs/nfsd[879]: [ID 866344 daemon.debug] Current SO_RCVBUF value is 128000 Jan 29 09:19:24 local at dismiss /usr/lib/nfs/nfsd[879]: [ID 115109 daemon.debug] Set SO_RCVBUF option to 1048576 Jan 29 09:19:24 local at dismiss /usr/lib/nfs/nfsd[879]: [ID 739773 daemon.debug] Current SO_SNDBUF value is 49152 Jan 29 09:19:24 local at dismiss /usr/lib/nfs/nfsd[879]: [ID 978932 daemon.debug] Set SO_SNDBUF option to 1048576 Jan 29 09:19:24 local at dismiss /usr/lib/nfs/nfsd[879]: [ID 866344 daemon.debug] Current SO_RCVBUF value is 128000 Jan 29 09:19:24 local at dismiss /usr/lib/nfs/nfsd[879]: [ID 115109 daemon.debug] Set SO_RCVBUF option to 1048576 Jan 29 09:19:33 local at dismiss nfs4cbd[893]: [ID 739773 daemon.debug] Current SO_SNDBUF value is 49152 Jan 29 09:19:33 local at dismiss nfs4cbd[893]: [ID 978932 daemon.debug] Set SO_SNDBUF option to 1048576 Jan 29 09:19:33 local at dismiss nfs4cbd[893]: [ID 866344 daemon.debug] Current SO_RCVBUF value is 128000 Jan 29 09:19:33 local at dismiss nfs4cbd[893]: [ID 115109 daemon.debug] Set SO_RCVBUF option to 1048576 Jan 29 09:19:33 local at dismiss nfs4cbd[893]: [ID 739773 daemon.debug] Current SO_SNDBUF value is 49152 Jan 29 09:19:33 local at dismiss nfs4cbd[893]: [ID 978932 daemon.debug] Set SO_SNDBUF option to 1048576 Jan 29 09:19:33 local at dismiss nfs4cbd[893]: [ID 866344 daemon.debug] Current SO_RCVBUF value is 128000 Jan 29 09:19:33 local at dismiss nfs4cbd[893]: [ID 115109 daemon.debug] Set SO_RCVBUF option to 1048576 Jan 29 09:19:43 local at dismiss /usr/lib/nfs/nfsmapid[901]: [ID 876094 daemon.debug] nfsmapid domain = internal Jan 29 09:19:58 local at dismiss /usr/lib/nfs/lockd[906]: [ID 739773 daemon.debug] Current SO_SNDBUF value is 49152 Jan 29 09:19:58 local at dismiss /usr/lib/nfs/lockd[906]: [ID 978932 daemon.debug] Set SO_SNDBUF option to 1048576 Jan 29 09:19:58 local at dismiss /usr/lib/nfs/lockd[906]: [ID 866344 daemon.debug] Current SO_RCVBUF value is 128000 Jan 29 09:19:58 local at dismiss /usr/lib/nfs/lockd[906]: [ID 115109 daemon.debug] Set SO_RCVBUF option to 1048576 Jan 29 09:19:58 local at dismiss /usr/lib/nfs/lockd[906]: [ID 739773 daemon.debug] Current SO_SNDBUF value is 49152 Jan 29 09:19:58 local at dismiss /usr/lib/nfs/lockd[906]: [ID 978932 daemon.debug] Set SO_SNDBUF option to 1048576 Jan 29 09:19:58 local at dismiss /usr/lib/nfs/lockd[906]: [ID 866344 daemon.debug] Current SO_RCVBUF value is 128000 Jan 29 09:19:58 local at dismiss /usr/lib/nfs/lockd[906]: [ID 115109 daemon.debug] Set SO_RCVBUF option to 1048576 Not really sure if i have to hack this right now or investigate further before changing them. Regards, Olivier On Fri, Jan 29, 2010 at 4:41 PM, a b <tripivceta at hotmail.com> wrote: No way am I going to tell you that, 2MB/s is busted, no ifs, buts, or maybes. Do you have any other clients which connect to the NFS server? What does theirperformance look like?Can you pinpoint the performance problem to OS X NFS client? Is it possible to connect the client and the server directly via a CAT5 crossover cable,to rule the network equipment out? Have you tried DTracing the NFS server? I'd take Brendan Gregg's DTrace toolkit and go to town on the NFS server, unless you can actually prove that there is a problem withthe OS X NFS client <---> SunOS NFS server. Is there anything in /var/adm/messages on the NFS server, NFS related, RPC related, possibly NIC related? Date: Fri, 29 Jan 2010 16:31:21 +0100 Subject: Re: [ug-chosug] [NFS] Zfs sharenfs bad performance From: [email protected] To: tripivceta at hotmail.com CC: ug-chosug at opensolaris.org God i should stop sending mail from the smartphone... Sorry bad replies. So to reframe the questions, how many MB/s do i expect: Around 35MB/s or more ... This is the least i have on similar setups.. How many do i have now: Around 2MB/s .... Please don't tell me this is normal "not lying" performance. Regards, Olivier On Fri, Jan 29, 2010 at 4:25 PM, Mauras Olivier <oliver.mauras at gmail.com> wrote: On Fri, Jan 29, 2010 at 4:10 PM, a b <tripivceta at hotmail.com> wrote: You are reporting 48MB/s over SMB, and that's with "lying", and I'm seeing12MB/s I/Os making it to the disk on my ZFS NFS server, and that's with amuch older system and "no lying", so it seems to me that both observations correspond to expected behavior/performance. I guess you could reframe the question: how many MB/s do you expect fromZFS + NFS? 48, like for CIFS/SMB? How many MB/s are you getting now? Date: Fri, 29 Jan 2010 16:06:42 +0100 Subject: Re: [ug-chosug] [NFS] Zfs sharenfs bad performance From: [email protected] To: tripivceta at hotmail.com CC: ug-chosug at opensolaris.org Sorry bad reply... So how could waiting I/Os to commit on a drive array capable of 180MB write on gigabit network could drop performances that bad??? That's just wrong... i never had that kind of bad performances even with the worst pata drive over 100mb network.... Hotmail: Free, trusted and rich email service. Get it now. _______________________________________________ Switzerland OpenSolaris Mailing List ug-chosug at opensolaris.org http://mail.opensolaris.org/mailman/listinfo/ug-chosug wiki: http://wikis.sun.com/display/chosug LinkedIn: http://www.linkedin.com/groups?gid=161188&trk=hb_side_g ISO images: http://mirror.switch.ch/ftp/mirror/opensolaris.org/release_isos/ IPS: http://ips.osug.ch:10000/release/ Pictures: http://www.flickr.com/photos/31224908 at N08 Hotmail: Free, trusted and rich email service. Get it now. _______________________________________________ Switzerland OpenSolaris Mailing List ug-chosug at opensolaris.org http://mail.opensolaris.org/mailman/listinfo/ug-chosug wiki: http://wikis.sun.com/display/chosug LinkedIn: http://www.linkedin.com/groups?gid=161188&trk=hb_side_g ISO images: http://mirror.switch.ch/ftp/mirror/opensolaris.org/release_isos/ IPS: http://ips.osug.ch:10000/release/ Pictures: http://www.flickr.com/photos/31224908 at N08 _________________________________________________________________ Hotmail: Trusted email with powerful SPAM protection. https://signup.live.com/signup.aspx?id=60969 -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://mail.opensolaris.org/pipermail/ug-chosug/attachments/20100129/c15a58f6/attachment.html>
