Les Mikesell <lesmikesell@gmail.com> wrote on 09/17/2012 02:44:20 PM:

> On Mon, Sep 17, 2012 at 11:54 AM, Timothy J Massey
> <tmassey@obscorp.com> wrote:
> >
> > However, I have recently inherited a server that is >3TB big, and 97%
> > full, too!  Backups of that system take 3.5 *days* to complete.  I *can't*
> > live with that.  I need better performance.
>

<Quick fixes snipped>

> In any case you have to keep in mind that if you ever
> have to restore, there will be considerable downtime.  For the price
> of a disk these days, it might be worth keeping a copy up to date with
> native rsync that you could just swap into place if you needed it and
> back that up with backuppc if you want to keep a history.


Fortunately, BackupPC is a "backup of the backup" right now, and is not expected to be used for real.  Yet.  That's why I can take the time and try to actually solve the problem, rather than apply band-aids.

But that will likely end in November, if not sooner.

> > It looks like everything is under-utilized.  For example, I'm getting a
> > measly 40-50MB of read performance from my array of four drives,
>
> If every read has to seek many tracks (a likely scenario), that's not
> unreasonable performance.


Which is why I asked if others were getting better performance.  They are.  So it's not just inherent to the task.  There's something different.

>   My physical drive and network
> > lights echo this:  they are *not* busy.  My interrupts are certainly
> > manageable and context switches are very low.  Even my CPU numbers look
> > tremendous:  nearly no time in wait, and about 50% CPU idle!
>
> I think you should be seeing wait time.  Unless perhaps you have some
> huge files that end up contiguous on the disk, I'd expect the CPU to
> be able to decompress and checksum as fast as the disk can deliver -
> and there shouldn't be much other computation involved.


I'm not.  I've got two of these new boxes built.  In both cases, they have 2-4% wait time when doing a backup.  One is a RAID-5 and one is a RAID-6.

Might it have something to do with md?  Could the time that would normally be considered wait time for BackupPC be counted as CPU time for md?  That doesn't seem logical to me, but I can say that there just isn't any wait time on these systems.

> Mine seem to track the target host disk speed more than anything else.
>  The best I see is   208GB with a full time of 148 minutes.  But that
> is with backuppc running as a VM on the East coast backing up a target
> in California and no particular tuning for efficiency.  Compression is
> on and no checksum caching.

That's the same settings I'm using.  But that's about double the performance I'm getting.  247GB in 340 minutes, or about 12MB/s.

I've just turned compression off for a couple of hosts.  We'll see how this affects their performance.  I'll let everyone know.

Timothy J. Massey

 
Out of the Box Solutions, Inc.
Creative IT Solutions Made Simple!

http://www.OutOfTheBoxSolutions.com
tmassey@obscorp.com
      22108 Harper Ave.
St. Clair Shores, MI 48080
Office: (800)750-4OBS (4627)
Cell: (586)945-8796