Date: Fri, 1 Feb 2002 16:47:28 -0600 From: "Mike Meyer" <mwm-dated-1013035648.cf9fb3@mired.org> To: dochawk@psu.edu Cc: questions@FreeBSD.ORG Subject: Re: spreading system across four fast scsi disks Message-ID: <15451.6912.20845.374468@guru.mired.org> In-Reply-To: <200202011730.g11HUKI42130@fac13.ds.psu.edu> References: <15436.40632.383947.819624@guru.mired.org> <200202011730.g11HUKI42130@fac13.ds.psu.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
dochawk@psu.edu types: > mike mentioned, > > Richard E. Hawkins <dochawk@psu.edu> types: > >If you can split the drives across two controllers, even better. > That I can do. I assume I put two little and one big drive on each > channel? I can't say without knowing more about what you're doing, but that sounds reasonable. > > > What about /var, though? The purpose of the split is so that the heads > > > can be in places likely to be used simultaneously. > > > Don't give swap it's own drive. If you're really thinking about giving > > it 9 gig, give it 2.x gig (x should be > 64K of memory so you can get > > dumps) on all four drives. > The reason I changed from 4 18g to 4x9g + 2x18g was to get the heads in > more positions. What I'm not grasping is why I don't save time with a > dedicated swap drive. > > The second sentence is critical. For swap, the system is smart enough > > to interleave swap across multiple disks. Let it do it's thing. > Ahh, it can hit multiple swap sections at once, then? Yes. That's why having multiple swap devices is faster than having a dedicated swap device. > > For example, the 4x9g cheetas might be 1Gig FS, 2Gig swap, 6Gig /home, > > with vinum used to strip /home across the four devices, possibly with > > a a mirror as well. The four fs's would be root, /usr, /usr/src and > > /usr/obj so that rebuilding the world would scream. > When performance matters, I'll be using all or close to all of memory > in a very large array or two, running a single instance of a (possibly > multi-threaded) fortran program. Each array element is only a few > words of memory, and they end up accessed in an essentially random > pattern as I either do an optimization (rather extreme dynamic > programming) or let the simulated entities interact (genetic > algorithms). The models will generally run for hours or > days. In that scenario, fhe four potential file systems that get touched root, /usr, /var, and /home. Change it to 2G of swap and 7G of file system. I'd put root & /usr on one channel, and /var /home on the other. Either put /usr/src and /usr/obj on the other two drives as you mentioned, or symlink them to /var and /home. <mike -- Mike Meyer <mwm@mired.org> http://www.mired.org/home/mwm/ Independent WWW/Perforce/FreeBSD/Unix consultant, email for more information. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?15451.6912.20845.374468>