Joel Coltoff wrote:
> A recent thread on Loooooong backups got me to look at my configuration. It
> always seemed to me that it should run a bit faster than it does. I wasn't too
> concerned given that it started at 9:00 PM and was done by 2:00 AM. It never
> got in the way of things. We are moving to a much larger server and I'd like
> to resolve this for the flood of new files we'll have. We are running 2.4.4p2.
>
> I've been trying different numbers in my interface setup and they don't seem
> to have any effect. This is what I have in my amanda.conf file. Most of the
> DLEs are on the host that runs amanda. I don't have a setting for "netusage"
> so I get the default. When the dump is done amstatus reports.
>
> network free kps: 10700
>
> define interface local {
> comment "a local disk"
> use 10000 kbps
> }
>
> define interface ethernet {
> comment "10 Mbps ethernet"
> use 400 kbps
You're limiting yourself quite a bit here. Amanda won't
start another dumper if the current dumpers are using more
than 4% of your bandwidth, so you'll probably never see
two remote dumps happening simultaneously. I would suggest
setting it much higher (like maybe 7 or 8 Mbps), and if
other apps are impacted too much during the dumps, lower it
a little bit each day until the other apps are no longer
too affected.
> }
>
> My disklist looks like this
>
> phoenix.wmi.com /export/cdburn project-files 2 local
> phoenix.wmi.com /export/cac project-files 2 local
> phoenix.wmi.com /export/opt project-files 2 local
> phoenix.wmi.com /export/project project-files 2 local
>
> goliath.wmi.com /users /users {
> user-files
> exclude "./jc517/lightvr/*/*.o"
> exclude append "./jc517/lightvr/*/*.bz2"
> } 1 ethernet
>
> goliath.wmi.com /export /export {
> user-files
> include "./wumpus" "./plover" "./uclibc" "./vendor"
> } 1 ethernet
>
> Finally, here is the tail of amstatus
>
> network free kps: 6700
> holding space : 10119040k ( 99.92%)
> dumper0 busy : 4:20:38 ( 94.20%)
> dumper1 busy : 0:02:16 ( 0.82%)
> taper busy : 3:31:28 ( 76.43%)
> 0 dumpers busy : 0:14:44 ( 5.33%) no-diskspace: 0:14:44 (100.00%)
> 1 dumper busy : 4:19:40 ( 93.85%) no-bandwidth: 2:45:21 ( 63.68%)
> not-idle: 1:34:19 ( 36.32%)
> 2 dumpers busy : 0:02:16 ( 0.82%) no-bandwidth: 0:02:16 (100.00%)
>
>
> If I run amstatus I'll see "no-bandwidth" associated with 1 dumper busy more
> often than not.
> What's a reasonable number to use so that I have more than 1 dumper running
> at a
> time? I guess the real question is should a single dump saturate connections
> to the localhost?
Unless you have other things running on your network while the
backups are running, and they can't tolerate the slowdown (such
as VOIP traffic), you should raise the bandwidth limit fairly high.
A single dump may not be able to stream more than a few Mb/sec,
due to factors such as disk speed, CPU, and network architecture.
As for your 'local' speed (where the client and server are the same
host), that interface has practically infinite bandwidth (not really,
but from a backup perspective it does), your limiting factor is bus,
disk controller, and disk I/O, and you're better off using a
combination of spindle numbers and inparallel to control the local
backups.
Frank
>
> Thanks
> - Joel
>
>
--
Frank Smith fsmith AT hoovers DOT com
Sr. Systems Administrator Voice: 512-374-4673
Hoover's Online Fax: 512-374-4501
|
