Date: Mon, 18 Apr 2005 21:53:47 +0200 From: Willem Jan Withagen <wjw@withagen.nl> To: Claus Guttesen <kometen@gmail.com> Cc: Kris Kennaway <kris@obsecurity.org> Subject: Re: NFS defaults for read/write blocksize....(Was: Re: 5.4/amd64 console hang) Message-ID: <4264104B.2030600@withagen.nl> In-Reply-To: <b41c755205041801433fad9c65@mail.gmail.com> References: <6eb82e05041500274172afd3@mail.gmail.com> <20050416122222.GA12385@totem.fix.no> <6eb82e0504160536572e068c@mail.gmail.com> <20050416183755.GB61170@xor.obsecurity.org> <4262CFBF.4090709@withagen.nl> <b41c755205041801433fad9c65@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Claus Guttesen wrote: >>>>By the way, I'm thinking that more frequently hang might related with >>>>large read/write block in mount_nfs -r/-w (I use 8192, original is 1024). >>> >>> >>>That's certainly possible since non-default settings don't get as much >>>testing. It would be good to get a traceback. >> >>Has it even been considered to up these values to something bigger?? > > > Read- and write-size of 32768 seems to work optimal for me: How did you come to this conclusion? What kind of workload? > nfssrv:/nfsmnt /localsrv/nfsmnt nfs > rw,tcp,intr,nfsv3,-w=32768,-r=32768 0 0 > > Nfs-server is an i386 and clients are i386 and amd64. This is in line with what the graphs suggest: Use Laaarrrrrggggeee sizes. But because of the non-significance of bonnie, I'm reluctant to accept my work as 100% proof. --WjW
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4264104B.2030600>