From: David R. <dr...@gm...> - 2007-03-28 21:54:39
|
On 3/28/07, John T. Yocum <jo...@fl...> wrote: > Here is the iostat output, the server is doing two full backups at the > moment, along with a nightly. Server specs: P4 3.2Ghz, 512MB RAM, 300GB > SATA drive. > > [root@backup2 ~]# iostat > Linux 2.6.9-42.0.10.ELsmp (backup2.fluidhosting.com) 03/28/2007 > > avg-cpu: %user %nice %sys %iowait %idle > 2.98 0.00 0.76 50.02 46.24 > > Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn > sda 69.48 798.71 428.68 1556126098 835203048 > sda1 0.00 0.00 0.00 1556 200 > sda2 0.17 2.68 0.57 5219723 1109536 > sda3 0.00 0.00 0.00 1073 320 > sda4 0.00 0.00 0.00 2 0 > sda5 0.23 3.04 1.14 5914146 2215408 > sda6 111.33 792.87 426.97 1544752254 831876168 > > On that server, we're seeing backup speeds as low as .11MB/s which are > backups of several thousand files. Some backups are over 1MB/s, but > those are servers with only 100ish files. Was this a raid 5 or raid 1 system with those SATA drives? I am concerned about the fact that it's reporting 50% of it's time in system instead of near 100% in wait which indicates some other sort of bottleneck. If you run iostat with a 5 second interval over a few intervals does the avg-cpu look similar? Also do you have the dir_index option on? (check by running `tune2fs -l /dev/sda6`) Not sure how much of a difference that makes with backuppc loads, but that does help significantly with metadata operations on large directories. It also looks like you are doing tar backups, have you tried rsync backups? -Dave |