W2K12R2 with iSCSI-Targets on HPE MSA2040 (VBR 9.5) ...

[Didi7]

Hello,

we are currently trying to speed up things a bit more by introducing SAN-transport instead of using our traditional VBR VMware proxy VMs. Speed with proxies was already very good (around 300-400Mb/s at max on 1GBit/s vNICS).

Our VMware cluster (currently consisting of 2 ESXi hosts) uses 10GBit/s iSCSI direct connections into the storage (works like a charm).

The HPE MSA2040 is equipped with 4x 10GBit/s iSCSI-connectors per controller (we have 2 controllers, resulting in 8x 10GBit/s iSCSI-connectors), Jumbo Frames are activated for all 8 iSCSI Host Interfaces.

Now, we decided to connect our VBR 9.5 server directly to the storage via iSCSI with one 10GBit/s interface. The 10GBit/s interface under W2K12R2 is configured to use Jumbo Frames (Packet size 9014).

We have explicit mappings in our HPE MSA2040 for 2 of 10 LUNs (read-only access) which are explicitly bound to the 10GBit/s iSCSI Host interface, where our W2K12R2 server is connected to.

We can enter the corresponding IP-address clicking Discovery Portal under the tab 'Discovery', used the Advanced tab to select the Local adaptor 'Microsoft iSCSI initiator', selected the corresponding Initiator IP and got our iSCSI targets under the tab 'Targets'.

So far so good. As soon, as we click to connect the target, it takes a while (around 4-5 minutes) till the iSCSI-targets get connected. We can see them in Computer Management under Disk Management and we can use SAN-transport in VBR 9.5 to backup VMs.

We have Firewall rules enabled and we also disabled the Firewall completely under W2K12R2 but connecting the iSCSI targets always takes a very long time. We can see more than 2 LUNs under Devices in the tab 'Targets', there is a 'Disk -1' on 'LUN 0', which represents the HPE MSA2040 enclosure.

We would like to know, if there are any other HPE MSA2040 users, connecting there iSCSI initiators directly to the iSCSI Host interfaces (without an Ethernet Switch) of the HPE MSA2040 from a Windows Server and who experience the same behaviour and if not, what kind of Firmware they use in their HPE MSA2040.

We are also interested to know, what kind of speed you get using SAN-transport under VBR (provided 10GBit/s iSCSI with Jumbo Frames is used).

We currently get around 280Mb/s (Proxy allows one concurrent task) and around 450Mb/s (Proxy allows two concurrent tasks).

Thank you very much for any insights in advance.

Best regards,
Didi7

[Didi7]

So far so good. As soon, as we click to connect the target, it takes a while (around 4-5 minutes) till the iSCSI-targets get connected

Correction:
It takes approximately 10 minutes to connect the iSCSI Targets.

[DaveWatkins]

I've no direct experience but if you're only connecting via one controller it may be that controller is not the active controller for those LUN's and so some level of redirection is going on. If your B&R server has a second 10Gb port I'd try connecting one to each controller and enabling MPIO and see how it goes

[Didi7]

Hi DaveWatkins,

first of all, thanks for replying. Well, we still have 3 iSCSI Host interfaces free and that would be an option. Unfortunately we are not in direct proximity of the storage, so this would take longer to work out.

The HPE MSA 2040 is Active-Active though. The physical VBR-server is connected via the first 10GBit/s Ethernet Adapter and is directly connected to port A4 (the last 10GBit/s Host interface) on the first iSCSI storage controller.

The HPE MSA 2040 has one controller with iSCSI Host interfaces A1, A2, A3, A4 and B1, B2, B3, B4 on the second controller. A1 and B1 is used from ESXi A and A2 and B2 ist used from ESXi B. A4 is connected to my physical VBR-server (no MPIO is used atm).

I have another idea, which might help. Configuring the Microsoft iSCSI initiator, I saw one picture browsing the internet, which suggested to enter the iSCSI target IP as well. That's what I intend to check first.

Will post the results here...

Regards,
Didi7

[Didi7]

Ok, problem solved. Connection to iSCSI Targets works instantaneously now, reboot and re-connection to iSCSI Targets works flawlessly as well.

What was the problem? Well, it all depends on how the Microsoft iSCSI initiator gets configured ...

1. Lets assume, everything on the HPE MSA2040 side is completely configured and the corresponding Host Interface has an IP-address
2. First we configured the iSCSI Target IP by clicking 'Discover Portal' under the tab 'Discovery' using the default Port 3260
3. In the 'Discover Target Portal' dialog, we clicked the 'Advanced' button
4. Then we selected 'Microsoft iSCSI Initiator' as 'Local adapter' and explicitly selected the iSCSI Target IP Address for the 'Initiator IP'
5. The HPE MSA2040 iSCSI Target IQN is instantly visible under the tab 'Targets' in the Microsoft iSCSI Initiator configuration with the Status 'Inactive'
6. Selecting the IQN and clicking 'Connect' in the 'Target' tab of the Microsoft iSCSI Initiator configuration opens the dialog 'Connect to Target'
7. In the 'Connect to Target' dialog the IQN is displayed, here you can decide to reconnect the iSCSI Target after reboot and to 'Enable multi-path'
8. And there is an 'Advanced' button again, not choosing 'Advanced' and clicking OK results in 10 minutes connection time for the iSCSI Target
9. Choosing the 'Advanced' tab, you have the additional option to select the corresponding 'Local adapter', the 'IP Initiator' and the 'Target Portal IP'

And point (9) really makes the difference. Selecting 'Microsoft iSCSI Initiator' as 'Local adapter', selecting the IP-address of your local iSCSI Ethernet Adapter for 'IP Initiator' and selecting the IP-address of the HPE MSA2040 Host Interface, where the local VBR Backup-Server is connected to as 'Target Portal IP' results in an immediate connection to your iSCSI Targets of the HPE MSA2040 without any noticable delay.

SAN-transport speed with a physical VBR Backup-Server as the Proxy (4 Cores and 64GB of RAM) on a direct 10GBit/s iSCSI connection to the HPE MSA2040 storage, results in 420Mb/s transfer speed using 2 concurrent tasks. Using 1 concurrent task results in around 250-300MB/s transfer speed at max.

Not bad for a 3 year old storage system fully equipped with 600GB SAS-drives with 2 RAID-10 and 2 RAID-50 Vdisks.

Though we must say, the two virtual HOTADD-proxies in our ESXi-hosts we used before, weren't bad as well. They pumped between 150-400MB/s of data (using deduplication) through their 1GBit/s vNICs. They will mostly be used for restores in future, as all our HPE MSA2040 LUNs are read-only presented to the physical VBR Backup-Server for safety reason.

Hope this might be helpful for some guys.

Best regards,
Didi7

[Didi7]

We are really more than satisfied. With SAN transport (instead of using HOTADD with 2 VM Proxies) in combination with Veeam B&R 9.5 Update 2, we could speed up backup further and lower backup window by more than 50%. I now get over 600MB/s transfer speed allowing 4 parallel tasks in one backup job using one 10GBit/s iSCSI-connection into our 3 year old HPE MSA2040.

Best regards,
Didi7

[nitramd]

Interesting information. Thanks for taking the time to post.

R,
-nitramd