[p4] Perforce and Raid 0+1 - Smith, Jeremy R (CACI) - com.perforce.perforce-user

3 messages in com.perforce.perforce-user[p4] Perforce and Raid 0+1

From	Sent On	Attachments
DiCesare, Gene	20 Sep 2004 06:06
Jan Mikkelsen	21 Sep 2004 16:41
Smith, Jeremy R (CACI)	21 Sep 2004 16:56

Subject:	[p4] Perforce and Raid 0+1
From:	Smith, Jeremy R (CACI) (Jere...@med.va.gov)
Date:	09/21/2004 04:56:15 PM
List:	com.perforce.perforce-user

You could always use a three-way mirror too if you've got some spare cash to avoid the single-disk failure problem.

Jer

-----Original Message----- From: perf...@perforce.com [mailto:perforce-user-admin at perforce.com] On Behalf Of Jan Mikkelsen Sent: Tuesday, September 21, 2004 5:42 PM To: 'DiCesare, Gene'; perforce-user at perforce.com Subject: RE: [p4] Perforce and Raid 0+1

Gene wrote:

We are planning to upgrade our Perforce server and are considering going with a Raid 0+1 configuration for our storage. This should allow very brief "unavailable" time for checkpoints and backups.

Reviewing past mail list submissions, seems to indicate this is currently being done at some sites. However, when Perforce Customer support was contacted, they indicated they knew of no such implementations. Giving us a sense it was not supportable.

Are people using Raid 0+1 for their Perforce server?

Probably everyone (unless they use RAID 5 or a Netapp).

Is the process used for checkpoints/backups as simple as: - Break mirror - Reset journal on "live" mirror - Checkpoint and backup "backup" mirror - Re-sync "backup" mirror to "live" mirror Does a checkpoint make changes internal to Perforce beyond creation of a Checkpoint file and clearing the Journal file? Does this process raise issues with Perforce? Have people had luck or issues with such an implementation?

The issue isn't with Perforce, it's with your filesystem. If you break a mirror, it is generally like an unclean shutdown. Did all the data blocks get written? What is the state of the metadata? Etc.

If you are using a filesystem that support snapshots, then you can get a point in time image of the filesystem without those issues (and without breaking a mirror). In the past others have posted about doing this with a Netapp; FreeBSD 5 snapshots are an option. Other operating systems will have different methods.

The other issue with breaking a mirror is that you lose redundancy while the mirror is broken. A total single disk failure during a checkpoing would be bad.

Regards,

Jan Mikkelsen janm at transactionware.com