[OpenAFS-devel] Re: Backups using commercial products

Mitch Collinsworth mitch@ccmr.cornell.edu
Wed, 3 Jan 2001 13:19:24 -0500 (EST)


On Wed, 3 Jan 2001, Lyle Seaman wrote:

> 1. be sure to coordinate the volume access with the vol server so
> that it doesn't get moved or deleted out from under you.  This is
> trickier than it sounds.
 
Just out of curiousity what happens now when a backup is in
progress (backup dump or vos dump of a BK volume) and a vos
move or vos remove is issued?  Both of these result in the BK
volume being deleted.  Are they held pending completion of
the backup operation first?


> 2. vos dump is slow because
>    i. it does a zillion seeks on the disk, one at a time.
>    ii. it uses RX at the same time.
> 
> If you want to make it faster, you either have to use a different
> local filesystem which keeps data and inodes closer together,
> or you have to queue up all the disk I/O all at once, so the
> disk scheduler can work more efficiently.  But then you may be
> hurting fileserver latency.

OK, but how does this explain the results people were reporting
back in August that vos dumps always run at ~3 MB/s indepent of
what hardware is used?  Is that because the disk seeks are the
primary bottleneck and seeks take about the same amount of time
regardless of disk type?  I guess I'd been assuming it was one
of those big mysteries of RX.

-Mitch