Date: Mon, 21 Apr 1997 19:26:28 +1000 From: Bruce Evans <bde@zeta.org.au> To: current@freebsd.org, imp@village.org Subject: Re: Speed deamons: How to build a build box? Message-ID: <199704210926.TAA28976@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
> 5) Try to have the disk geometry match the physical disks' > (although I didn't measure anything relating to doing this > or not doing this). This is bad advice. It usually makes no difference, but actually matching "the" geometry is likely to be a small pessimization, and failed attempts to match "the" geometry is likely to be a large pessimization. Most modern disks have multiple geometries, one per zone. To match them, you need a either a different slice for each geometry so that you can put the geometries in the labels and use newfs parameters -t0 -u0, or a different partition for each geometry and newfs parameters -t<actual tracks/cylinder for this partition> -u<actual sectors/track for this partition>. You also need to get some of the disk latencies exactly right (if the latencies depend on the zone, then you need a slice per geometry, since the latencies aren't newfs parameters). >As you can see, the biggest pop by far came from the use of async and >noatime on both /usr/src and /usr/obj. I suspect that 5400 or 7200 I'm surprised that this makes so much difference (a factor of 2). Compilation is not very disk intensive. I didn't notice much change from using -async here (any improvement was eaten by source tree bloat :-]). Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199704210926.TAA28976>