Date: Sun, 9 Dec 2001 20:33:28 -0500 (EST) From: Garrett Wollman <wollman@khavrinen.lcs.mit.edu> To: nik@FreeBSD.ORG Cc: arch@FreeBSD.ORG Subject: Re: Getting rid of /usr file system (was: Using a larger block size on large filesystems) Message-ID: <200112100133.fBA1XSx27617@khavrinen.lcs.mit.edu> In-Reply-To: <20011209104437.A69671@clan.nothing-going-on.org> References: <20011208102658.B11428@dragon.nuxi.com> <200112082050.fB8Ko1T01347@mass.dis.org> <20011209164606.C83634@monorchid.lemis.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Quoting Nik Clayton: >On Sun, Dec 09, 2001 at 04:46:06PM +1030, Greg Lehey wrote: >> Well, since we're all being personal, here's my opinion. [Nik's layout deleted] If we're all going to give our layout plans.... I belong to the old school, but I'll gladly admit that this is very much a comfort issue and hard to empirically justify. Our servers are configured in approximately the following manner: Filesystem Size Used Avail Capacity Mounted on /dev/da0s1a 126M 48M 68M 42% / ; 128M allocated /dev/da0s1e 504M 75M 388M 16% /var ; 512M allocated /dev/da0s1f 2.0G 969M 886M 52% /usr /dev/da0s1g 2.0G 17M 1.8G 1% /var/spool /dev/da0s1h 6.9G 1.2G 5.1G 19% /home /dev/da0s1d 1.8G 1.3G 375M 78% /home/ncvs The `f' and `d' partitions (which should have been `e' and `h' instead) are assigned to purposes appropriate to the machine; we keep the (mail and print) spools on a separate filesystem because they only contain data which should never be backed up. (Some of the contents of /var is backed up, although not the log files.) /usr would need to be bigger on a workstation, if the user expected to use one of the more piggish graphical environments. (Oh, and swap is 1536 kbytes, or 1.5 times physical memory. Most of these machines never swap except when being attacked.) da0 is a 15-Gbyte volume on our Fibre Channel RAID system. We used to have a quarter-gig for /var, but that ran out during an attack on the previous generation of this setup, so we increased it for this round. We generally copy the partition sizes on all of our servers, because of the way we distribute software to the machines (they all end up with an identical complement). Just for comparison, xyz (cvsup3/ftp5) has the following configuration: Filesystem Size Used Avail Capacity Mounted on /dev/da0s1a 124M 47M 67M 41% / /dev/da0s1e 504M 99M 365M 21% /var /dev/da0s1f 2.0G 1000M 854M 54% /usr /dev/da0s1g 11G 357M 9.5G 4% /home /dev/da1s1d 3.9G 2.4G 1.2G 67% /u /dev/da1s1e 81G 52G 23G 69% /y /dev/ccd0c 93G 45G 40G 53% /y/ftp/pub/FreeBSD In this example, you can see that we gave most of the space on the RAID 5 to /home, which doesn't get used much anyway. da1 is a RAID-0 volume on the Fibre Channel array, which we split up into /u (for fast access to the cvsup tree) and /y (for slow access to less-popular ftp mirrors). ccd0 is composed of two internal ST150176LCs, striped; the filesystem parameters on that disk are configured to spread the workload over both drives without unnecessarily breaking up large requests (or putting all of the superblocks on a single drive). My workstation is not part of the shared configuration system (it can't be, since it runs -stale^H^H^H^H^Hcurrent). It looks like this: Filesystem Size Used Avail Capacity Mounted on /dev/da0s1a 124M 73M 41M 64% / /dev/da0s1e 124M 24M 90M 21% /var /dev/da0s1f 1.9G 1.2G 598M 67% /usr /dev/da0s1g 4.9G 800M 3.7G 17% /usr/ports /dev/da0s1h 719M 368M 294M 56% /usr/obj /dev/da1s1d 1008M 562M 365M 61% /usr/src /dev/da1s1e 7.4G 4.7G 2.0G 70% /homes da1 is an old Barracuda 9LP with a narrow interface that I really should replace (since it's slowing down my bus). My / and /usr have seven years' worth of accumulated crap and are perhaps not the best examples. -GAWollman To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-arch" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200112100133.fBA1XSx27617>