How to set up direct SAN mode (FC) ?

[Burnwell]

Hi

Sofar we have been running network based backups, but want to switch to direct SAN mode, and have installed a FC HBA in the backup server.
However, we are a bit worried of screwing things up.

We have a dedicated Windows 2003R2 Veeam 5 backup server with one FC HBA
In the SAN we can choose between "vmware" or "windows" host profile for this backup server. Which one should we go for?
There doesn't seem to be a read-only option..
Is there some windows LUN automount stuff that we should be aware of?

Can any of you point us in the right direction?

Best regards
JJ

[Burnwell]

Is there some windows LUN automount stuff that we should be aware of?

Nevermind this question, found it in the FAQ

[jtplawton]

you can do the following

Click Start > Run.
Type diskpart and press Enter.
Type automount and press Enter.
If Automount is enabled, type automount disable to turn off the feature. This prevents resignaturing of your VMFS LUNs
Close the diskpart command prompt.

on my server just to be sure and I also give the Veeam Backup Server only read access to the SAN Storage

then check out this guide to connect to the Storage via iSCSI

http://jpaul.me/?p=334

[Burnwell]

Thank you for the reply.
We are using fibre channel SAN though for our production.
Got it working now, but for some reason we cannot get full backup speeds over 60MB/s, no matter what we do.
It doesn't seem to matter if we backup to local disks (15K SAS) or to iSCSI target (raid6, 16 disks SATA).

On the FC SAN we have made a small dedicated disk pool with a few LUN's for testing this (raid 5, 7 disks 15K SAS), and when benchmarking these from the physical Veeam backup server we get average read/write speeds of 200MB/s+. Same goes for the iSCSI target - if we from the Veeam server copy files directly from LUN on FC SAN to LUN on iSCSI taget, network usage on the dedicated 1Gbit iSCSI NIC is 99% meaning close to the 125 MB/s maximum. if we copy files directly from LUN on FC SAN to local disks speed is around 200MB/s.

Why can't we make Veeam go faster?
With "low" compression settings, cpu usage is ~40 %
Veeam server is physical running Win2003 with quadcore cpu and 4GB RAM.

Regards
JJ

[foggy]

Basically, performance bottlenecks can be caused by three main reasons described in this thread. It seems you are ok with the first two of them, so maybe the source data retrieval speed is the case.
Also make sure you have your backup server is configured properly to run in Direct SAN mode and does not fail over to a Network mode.

Btw, what target storage is you job optimized for? Is it local target?

[Burnwell]

Tried testing the source data retrieval speed as described in the thread you pointed to.

1 VM, shutdown with one 30GB thick vmdk. Filesize on the datastore is 31.457.280,00K

Job settings:
Direct SAN access (disabled failover to network mode)
Backup mode: incremental
Inline data deduplication enabled
Storage compression level: low
Storage optimize for: local target
CBT disabled
Integrity checks enabled
No vmware tools quiescence, application aware image processing and guest file indexing

First Run: Full
Total VM size: 30,00 GB
Processed size: 30,00 GB
Processing rate: 93 MB/s
Backup mode: SAN without changed block tracking
Start time: 03-09-2011 13:15:42
End time: 03-09-2011 13:21:14
Duration: 0:05:31

Resulting in a 17.679.224KB .vbk file
During the first ~20GB speed was constantly around 60MB/s. After that speed went up (white space) so the Processing rate ended at 93 MB/s.
But still only 60MB/s when there is actual data. CPU usage was 40-45% all the time.

Second Run: Incremental
Total VM size: 30,00 GB
Processed size: 30,00 GB
Processing rate: 310 MB/s
Backup mode: SAN without changed block tracking
Start time: 03-09-2011 13:22:19
End time: 03-09-2011 13:23:58
Duration: 0:01:39

Resulting in a 8.237KB .vib file

According to this, our source data retrieval speed looks quite good, so where is our bottleneck?
I tried the above tests to both local storage and iSCSI target - no difference at all.

Regards JJ

[Vitaliy S.]

If you have the same performance with local target and iSCSI that means target storage shouldn't be a bottleneck. Please be aware that multiple VMs will have different processing rates due to these reasons.