From: Chris R. <cro...@gc...> - 2009-03-12 20:08:46
|
Mike Dresser wrote: > Matthias Meyer wrote: > >> Dear all, >> >> How scalable is backuppc? >> Where are the limits or what can produce performance bottlenecks? >> >> I've heard about hardlinks which can be a problem if theire are millions of >> it. Is that true? >> >> > The file system can become... interesting to fix or backup when you get > a few million hard links, especially if you're using XFS. Amen to that... [root@archive-1 trash]# df -i /dev/sdb1 Filesystem Inodes IUsed IFree IUse% Mounted on /dev/sdb1 6442438528 55813203 6386625325 1% /data [root@archive-1 trash]# mount |grep /dev/sdb1 /dev/sdb1 on /data type xfs (rw,noatime,nodiratime,nobarrier,logbufs=8,sunit=512,swidth=6656,logbsize=262144) [root@archive-1 trash]# cat /etc/redhat-release CentOS release 5.2 (Final) [root@archive-1 trash]# uname -srvmp Linux 2.6.18-92.1.18.el5 #1 SMP Wed Nov 12 09:19:49 EST 2008 x86_64 x86_64 > There _appears_ to be some bugs in Debian etch's xfs tools, last time I had to > run an xfs_repair -n on etch it took 6 days.. after upgrading to lenny > it takes about 10 minutes or so. I'm guessing there were some major > improvements in the xfs tools from etch to lenny with regards to memory > usage. > >> Anyone would share his experience about CPU usage in relation to bandwidth >> usage? Particularly by using rsync? >> >> > I see about 2-15m/s on our rsync backups CPU usage is pretty low, it's > mostly disk i/o that holds it up. The newer faster machines back up > faster, so there's a bottleneck on the clients as well. > Same story here. CPU is, to a close approximation, never a problem. For an added twist, most of my clients are on the far end of a satellite link. After the initial backups, even running 16 backups in parallel, I hardly peak above 4Mbit/sec (per iftop). >> I have a 2x1GHz Server with 2GB RAM and 4x500GB SATA Disks in Software >> Raid5. >> >> > server here is a 2x1.8ghz opteron 265, 5GB ram, 8x1TB with 3ware raid5. > Backup window is set from 18:00 to 23:00, it generally finishes all the > backups within that 5 hours. Full's every 8 days, incr every day. > > * Pool is 1852.64GB comprising 8466234 files and 4369 directories > (as of 3/12 02:26), > * Pool hashing gives 9654 repeated files with longest chain 85, > * Nightly cleanup removed 7189 files of size 35.62GB (around 3/12 > 02:26), > * Pool file system was recently at 44% (3/12 12:05), today's max is > 44% (3/12 00:00) and yesterday's max was 44%. > > There are 42 hosts that have been backed up, for a total of: > > * 789 full backups of total size 10819.30GB (prior to pooling and > compression), > * 284 incr backups of total size 913.00GB (prior to pooling and > compression). For comparison, I have a Xeon X3320, 8GB RAM and 16 Seagate ES.2 1 TB drives (ST31000340NS) on a Adaptec 51645 using a RAID6 setup (effectively 13 data drives, 2 parity, 1 hot spare). Fulls every 7 days, incremental every day. I Started off keeping 4 weeks of fulls and two weeks of incrementals, and cleaning the whole pool nightly. All (~125) backups completed within the 20:00-07:00 time frame and BackupPC_nightly finished in under 5 hours (with a concurrence of 8). Then I doubled retention on both fulls and incrementals and have been fighting steadily worsening performance problems since. Dropping the retention back to initial levels is taking a while (BackupPC_trashClean seems to finally be gaining some ground). * Pool is 587.09GB comprising 31187909 files and 4369 directories (as of 2009-03-11 14:46), * Pool hashing gives 2294 repeated files with longest chain 944, * Nightly cleanup removed 1648979 files of size 50.98GB (around 2009-03-11 14:46), * Pool file system was recently at 11% (2009-03-12 11:00), today's max is 11% (2009-03-12 07:00) and yesterday's max was 11%. There are 125 hosts that have been backed up, for a total of: * 495 full backups of total size 2384.27GB (prior to pooling and compression), * 855 incr backups of total size 380.11GB (prior to pooling and compression). Chris |