Date: Mon, 10 Nov 2003 11:32:42 -0500 From: "Boring, Andrew" <Andrew.Boring@millerzell.com> To: <freebsd-questions@freebsd.org> Subject: Disk block size tuning - multiple questions Message-ID: <E7913F191C480F458C2B9B40F493FA3EB1D6@WS-EXCHANGE2003.corp.millerzell.com>
next in thread | raw e-mail | index | archive | help
Greetings! First question: I am configuring a server to hold source trees for 4.8/4.9 and maybe = -current for production and experimental upgrades on my several FreeBSD = servers. Currently, I only keep the -release sources for maintaining = patchlevels on the individual boxes themselves. The last time I started = adding the -current source tree on a non-production server, I ran out of = inodes before actually completing the CVS sync. What newfs = blocksize/fragment size and inode density is optimal for 10 to 16 GB = partition containing just source trees? The default stripe size on the = RAID controller is set to 128k for mirrored drives. Hardware: Compaq ProLiant DL360, dual 18GB (mirrored) drives, with = Compaq 5i SmartArray controller. FreeBSD 5.1 Next question: Are there optimal newfs block/fragment/inode options for a quad-disk, = 1+0 RAID for a mail server running Postfix? Specifically, the queue = directory and the home (Maildir) directories are what I'll need to = optimize. I think the queue directory will be heavy read/write, whereas = I imagine that the Maildir directories (kept on a different partition) = will be more write-intensive over-all. The queue directory will reside = on a (approximately) 1 GB /var partition, and the Maildirs will be on a = roughly 27 GB /home partition (no other user files on /home). The = default stripe size on the RAID controller is set to 128k for mirrored = drives. I was looking at the tuning(7) recommendations of something like = block size 8192 and frag size 1024 for the queue and Maildir = directories. SInce Maildir will have to accomodate occasional large = files with attachments, this seems like it might be a happy medium. I = don't have a specific usage pattern developed on the number of/size of = email files and attachments, so I am planning on a vague, nebulous = "average email use". Hardware: SGI 1200, quad 18GB (RAID 1+0) drives, Compaq 4200 SmartArray = controller, FreeBSD 4.9 Another question: I've been doing a lot of reading on filesystem/RAID levels for database = use (I'm not a DB guy). I don't have a server ready for this yet, but i = would like some generic recommendations on filesystem setup. I've read = that mirrored stripes (0+1 or 1+0) is better overall than RAID 5. Any = other tuning tips, such as block size, inode, async/sync/softupdates, = etc for a partition to hold an SQL database (such as MySQL or Postgres, = etc)? The tuning(7) man page seems to indicate a default block size = (16k) but fewer inodes for databases. Hardware: SGI 1200, quad 18GB (RAID 1+0) drives, Compaq 4200 SmartArray = controller, FreeBSD 4.9 Last question: With a good battery-backed caching controller (such as the Compaq = SmartArray cards), is SoftUpdates of any use? My understanding is that = SoftUpdates does in software the same sort of caching that RAID = controllers would do. Am I correct or way off base? Should I mount all = drives synchronous without SoftUpdates, or asynchronous with = SoftUpdates, or what is the recommended choice when using real = server-class SCSI RAID controllers? I am more concerned overall with = higher-availability than higher-performance. Thanks in advance! (please Cc: me on all responses for this thread; I'm not subscribed) -- Andrew Boring Miller Zell Desktop Services <andrew.boring@millerzell.com>=20 "Microsoft DNS service terminates abnormally when it receives a response to a dns query that was never made. Fix information: run your DNS service on a different platform." -- bugtraq http://www.securityfocus.com/archive/1/6212
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?E7913F191C480F458C2B9B40F493FA3EB1D6>