MSA2324i & iSCSI MPIO problem

[mzob]

Hi, I'm gonna try to be as short as possible, giving all the info I think is relevant. PLS ask for more info if it's needed.

HW:
3 physical servers (names: 001F, 002F, 003F)
3 virtual servers (names: 001V, 002V, 003V)
1 MSA2324i storage

LAYOUT:
WinServer 2008 R2, Failover cluster
Storage has 2 controllers, each with 2 iSCSI ports. IP addresses are:
A1 - 10.0.0.83
A2 - 10.0.1.83
B1 - 10.0.0.84
B2 - 10.0.1.84
There are 2 switches, one with A1 & B1, the other with A2 & B2.

001F, 002F, 003F servers have 2 NIC with 2 ports each (4 LAN ports per server). Each is configured in the same way:
NIC1,P1 to storage (name: MSA1, 10.0.0.x)
NIC1,P2 disconnected
NIC2,P1 to storage (name: MSA2, 10.0.1.x)
NIC2,P2 MS Virtual Network Switch (connected to domain, DHCP)
there's one more LAN connection, named "domain - virtual network" that's connected to domain and has physical server's fixed IP address.

Storage has 2 LUNs: one is used entirely as CSV, the other is set up to be available explicitly to one physical server only (used for backup purposes).

So, what happens is that I have the MPIO (multipath) set, and I know it by looking at network traffic (via task manager) when I set a large file copying from Clusterstorage\Volume1 to a local disk. But, the utilization on the both MSA1 & MSA2 (iSCSI LAN) is always low. Very low.
So on the server 002F (physical) where the both LUNs are (explicit one and CSV one), when I set copying from explicit volume to desktop, it goes up above 100 MB/sec. But when I set copying of the SAME file from CSV to desktop, it is around 10-15 MB/sec. And I can't get it to work. It uses only one MSA connection (monitored via task manager) and it uses it barely (utilisation never goes more than 15%).

Why? What's wrong here? In iSCI initiator, when I look under Target (1 listed), then click "devices..." I see ("disk -1", "disk 1", "disk 2", "disk 3") three times (repeats). After selecting "disk 1" or "disk 2" or "disk 3" and clicking on MPIO button, in the "paths" I see 4 paths for every disk on the list, all paths marked as "connected" and "active" and load balancing is set to RRwithSubset. If I try to change load balancing policy I get an error (to whatever policy I set it to).
If I look on disk manager, select disk, then MPIO, I see 2 paths as "active/optimized" and two as "active/unpotimized", and load balancing set to RRwithsubset. It's identical for both LUN (the CSV and the explicit one), yet the filecopy is incredibly slow for CSV.

[mzob]

To add more info:
test is copying a 31GB file. Results:

001F (from csv to local) --> 17 MB/sec
003F (from csv to local) --> 60+ MB/sec
002F (from csv to local) --> 17 MB/sec
002F (from explicit to local) --> 100+ MB/sec

When copying csv-2-local, on slow connections, only MSA2 shows traffic, and it's around 12%. Never more, never going up.

[mzob]

Another observation: the CURRENT OWNER of the CSV has the speed. Other two nodes don't. When 003F was the fastest, it was also the owner. I moved the ownership to 002F, which improoved it's speed significantly (over 100 MB/sec) but the speed on the ex-owner dropped to 17 MB/sec.

I don't think it's normal...

[Martin Smoral]

Sounds like everything is configured correctly. I am at a loss too. I would be interested in what the resolution is.