On May 18, 2010, at 9:35 AM, Robert Hartzell wrote:
> On Mon, 2010-05-17 at 15:08 -0400, Paul Mather wrote:
>> I am currently assembling a quote for an LTO-4 tape backup system. So far,
>> I am looking at using a 16-slot Quantum SuperLoader 3 with LTO-4HH drive as
>> the tape unit. Married to this will be a server to act as the backup server
>> that will drive the tape unit using Bacula to manage backups. The server
>> will be a quad core X3440 system with 4 GB of RAM and four 1 TB SATA 7200
>> rpm hard drives in a case that has room for eight hot-swap drives. I plan
>> on using FreeBSD 8 on the system, using ZFS to raidz the drives together to
>> provide spool space for Bacula. I will be using an Areca ARC-1300-4X PCIe
>> SAS card to interface with the tape drive.
>>
>> My main question is this: is the Quantum SuperLoader 3 LTO-4 tape drive
>> supported by Bacula 5 on FreeBSD? In particular, is the autoloader fully
>> supported? The Bacula documentation indicates the SuperLoader works fully
>> under Bacula, though not explicitly whether under FreeBSD.
>>
>> The backup server will serve a GigE network cluster of perhaps a dozen
>> machines with over 6 TB of storage, most of which is on the cluster's NFS
>> server. Does anyone have good advice on sizing the spool/holding/disk pool
>> for a Bacula server? Is it imperative to have enough disk space to hold a
>> full backup (i.e., 6 TB in this case), or is it sufficient to have enough
>> space to maintain streaming to tape? (I don't have much experience of
>> Bacula, having used it only to back up to disk.) In other words, do I need
>> more 1 TB drives in my backup server?
>>
>> Finally, is 4 GB of RAM sufficient for good performance with ZFS? Will ZFS
>> on FreeBSD be able to maintain full streaming speeds to tape, given the
>> various reports of I/O stalls under ZFS reported recently?
>
> ZFS loves ram. More ram = better performance. I'm not at all familiar
> with zfs performance on feebsd but zfs version 13 that's used on freebsd
> 8 is pretty old. ZFS is currently at version 22.
That must be on OpenSolaris. My current Solaris 10 system reports using pool
version 15. FreeBSD 8-STABLE is now up to pool version 14, and I believe
porting is currently underway to jump it up more pool levels.
> I/O stalls? Is that a freebsd issue?
Some folks have reported short, bursty I/O stalls during very intense write
workloads. I've seen it reported on FreeBSD, but also in those threads there
has been mention of it happening on, e.g, OpenSolaris, too. The general
workaround advice currently on FreeBSD is to lower the vfs.zfs.txg.timeout
kernel tuneable from its default of 30 seconds to something lower. IIRC, this
problem may also only affect systems with large ARC sizes.
Thanks for the RAM advice; I will try and bump it up.
Cheers,
Paul.
------------------------------------------------------------------------------
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users
|
