Need help with msa p2000 g3 mpio iscsi

[hyvokar]

Hi,

We have an oldish HP MSA p2000 g3 iscsi as a backup target.
Yesterday I enabled MPIO to get more troughput since our backups are gettin quite big.

I ran in to problems. First, my archive backups vere corrupted. After that I formatted the volume and send backups there again. During the night server has bsod'ed and in morning the whole volume was not readable and windows is asking to forma it.  I think I might have an fundamental error in my mpio config. The archive volume has windows deduplication enabled.

The setup is following:
I have 24x900gb disks in two raid6 vdisks.

Management interfaces are assigned to 192.168.100.0 network.
A1 and B1 port has been assigned to 192.168.100.0 network
A2 and B2 port has been assigned to 10.100.11.0 network

Backup server is running win2012 r2
Nic1 has been assigned to 192.168.100.0 network
Nic2 has been assigned to 10.100.11.0 network

MPIO is enabled in windows server.

connecting to storage:
From backup server, start iscsi initiator
go to discovery tab, click discover portal
enter A1 port ip address, OK

go to targets tab, Storage now appears in "discovered targets".
Click connect and Properties.
Click add session, select enable multi-path, click advanced. Select "Microsoft iSCSI Initiator" as local adapter, Nic1 IP as initiator IP, B1 ip as target IP.
Add two more sessions like above, Nic2 -> A1, Nic2 -> B2.
At properties, click devices -> MPIO -> select Least Queue Depth as load balance policy.

Close everything and start computer management.
Under disk drives, I now have two HP P2000 G3 iSCSI Multi-Path Disk Devices.
Right click -> properties, uncheck "Enable write caching on the device", go to MPIO tab, select Least Queue Depth for MPIO Policy. Repeat for another disk.

Some observations, that puzzles me.
When I now copy large files over iSCSI, the copy is extremely fast. However, it seems that the file is actually being cached in the memory and after the copy "finishes", it is actually still being transferred for tens of seconds (observed from task manager -> performance).

For some reason in isci initiator -> target -> properties, in the sessions tab, in the identifiers, some of the identifiers has Data Digest automatically set to "CRC32", some has "None Specified".  Also, since this morning I now have five identifiers. Nic1 -> B2 seems to be double.

Some questions:
Is it ok to have multiple storage ports in the same subnet or should each port be in its own subnet? Is it ok to have iSCSI ports in the same subnet than management ports?

In iscsi initiator, in discovery tab, should I add each IP address to target portals or only A1, or A1 and B1?
In isci initiator target -> properties -> Should I add so many sessions, or just have Nic1 -> A1 and Nic2 -> B2 or something else?
In the device manager, is it OK to disable caching, for maximum stability?

Thanks in advance, please ask for more detail if needed.

E:typos

[Shawn_K]

Hello,

It is not recommended to have the management ports and the host ports on the same network. As a best practice the host ports should be in a seperate vlan.

You need to be careful on your MPIO configuration. NIC1 has two ports while NIC2 has two ports. Be careful that the backup application is correctly configured for two NICs since they are on seperate networks.

When you see the two P2000 devices in Disk Manager are they the two vdisks on the P2000 or have you replicated the P2000? If the MPIO and the iSCSI initiator is not configured correctly, this can happen. Especailly since you say "in the morning I now have five identifiers"

The other concern is how you have the backup configured. You should not be caching on the host side like you describe. However, that is something that is occurring on the software layer or within your host configuration and is not due to the array.

I woiuld suggest going through and checking your host side configuration and then review the software configuration. I personally do not like to cache streaming data unless I have enough CPU and memory to cache and destage that data quickly. Otherwise you can create a bottleneck that causes other problems.

Cheers,

Shawn

[hyvokar]

HI Shawn, thank you for your reply.

Do you mean, that each of iSCSI port (from A1 to B4) should be in it's own vlan/subnet?
A1 10.10.1.1. A2 10.10.2.1, A3 10.10.3.1, A4 10.10.4.1, B1 10.10.5.1, B2 10.10.6.1, B3 10.10.7.1, B4 10.10.8.1

I'm using Veeam Backup&replication to do the backups. As far as I know it does not care about underlying network, as long as OS can connect them.

I have two vdisks in p2000, and they are the ones that show in disk manager. I have no replication configured in P2000, it's a standalone device.

I have not noticed caching before I installed MPIO, so I kinda assumed it's somehow related to that.  I have dual xeon cpus and 48gigs mem on backup server, and it has no additional roles, so resources should be adequate.

With two nics on backup server, two vdisks on the P2000 ("active" backup and backup archive), what would be the best practice to configure nics? I'm backing up a vmware cluster which is in the 192.168.100.0 network.

[Shawn_K]

Hello,

The issues you are having are host based issues / configuration problems.

From the host side there are considerations.

If you have two NICs configured on two different subnets how are you handling failover?

Using VEAM are there two backups or one? Since you mentioned active and archive you have to consider how the archive is working? Is that using a VEAM snapshot of the active backup?

On the host side, MPIO failover should be configured so that you can perform path failover. With two different subnets you will have to consider carefully how that path failover is working. You will need to confirm and test that your failover pathing policy is actually working correctly. You also need to check and confirm the queue depth and multipathing policies are set correctly for your server and NICs. For example MRU or RR? This will depend on hardware and OS versions.

It sounds like the server can handle the load once configured correctly for VEAM and ESXi. However, be aware that with newer versions of ESXi ATS/VAAI may not be compatable with the P2000 firwmare. You should review SPOCK for the latest firmware compatability with the P2000 and ESXi to ocnfirm.

Sorry I cannot give a more firm answer but there are a lot of configuration unknowns here which makes answering your questions fully very difficult.

Cheers,

Shawn

[hyvokar]

Ok, let’s forget esxi, veeam and other stuff.


What is the best practice to configure iscsi with mpio to msa p2000 g3, from physical Windows 2012r2 server with two nics? P2000 has two vdisks.