Cluster Storage p4000 StoreVirtual - bad Performance

[mcarraro]

Hay,

I have 2 p4300 g2 with 2 nic gb, 8disk raid5, and 1 p4330 g8 with 4nic gb 8disk raid5, all storages pluged on switch 1910 and all ports of storage are aggregate on switch and mtu9000, all p4000 on cluster running software version 12.

The server is DL380p gen8 with 2nic agregate to san with mtu9000 and 2nic agregate to lan, over this server running xenserver 6.5.

The VM running is Oracle Linux 6.6 x64 with filesystem ext4, on my tests using dd...

dd if=/dev/zero of=disco bs=1M count=1000

The result is 60mb/s

dd if=disco of=/dev/null bs=1M count=1000

The results  5gb/s

My question is, why 60mb/s when it could have 200mb/s (2nic  1gb nic by nic = 2gb +- 200mb/s).

[a_o]

There could be different issues in such a setup that can affect performance.

First thing to know is that you would never get wire speed performance, just based on the architecture of StoreVirtual.

Not looking at any hardware or switch configuration issues, the first thing I would look at, is if you can implement MPIO in 

Linux. This will impact read performance.
 

Also, are the P43x0 NICs configured with bonding that match the switch's LACP configuration?

What about the DL380p NICs,  how are they configured WRT aggregation?

Lastly, have you tested your storage performance at the hypervisor level?
VM performance can be affected by hypervisor settings - i.e. networking stack etc.

[mcarraro]

Thnak  @a_o!

All bond nic is configured over LACP, and switch have LACP enable on the group port aggregation.

I tested performance using local disk over hypervisor, the result it good 260mb/s.

I will configure MPIO for that I need to break the "bond" of "nic" and add a different network each "nic".

Example.
STG1 = 192.168.200.1  / 192.168.201.1
STG2 = 192.168.200.2  / 192.168.201.2
STG3 = 192.168.200.3  / 192.168.201.3

CLUSTER STORAGE IP (VIP) = 192.168.200.100 /192.168.201.100

XenServer = 192.168.200.10  / 192.168.201.10

In XenServer I have already enabled the MPIO.

In this way I would have MPIO configured correctly??

[HPstorageTom]

dd is actually a single threaded process and you are limited by the latency each I/O is experiencing and the measured 60 MBps is in the expected range for a single threaded process.

If you are using XenServer, then you should stay with the NIC bonding. MPIO is actually not supported for XenServer.

[mcarraro]

greats man @HPstorageTom

The xenserver version running is 6.5, in CMC i have varius sessions over one IP, this not multipath?

I'll test the vmware on the same server, and compare the results...

[HPstorageTom]

You do have multiple iSCSI sessions from the same initiator. If you do have multiple NICs in your XenServer, then you do use NIC bonding on this XenServer.

If I do use multipath (MPIO) in server connected to a StoreVirtual system, then I do have multiple NICs in this server and each NIC with his own IP address. MPIO is then taking care of the distribution of the I/O's across the NICs. We do support for instance the native MPIO of VMware and Windows. The attached screenshot shows an example of a Windows server with 2 NICs (192.168.58.17 and 192.168.58.18) using only the native Windows MPIO to connect to a StoreVirtual system (192.168.58.29).

[mcarraro]

@HPstorageTom

Thanks your reply.

I can set two IP on my cluster? Bcause having two Target ok?

[HPstorageTom]

No you use a single cluster VIP to connect to.

If you do have OS that is supported for MPIO, then you set a separate IP address for each iSCSI NIC of this server.

[mcarraro]

@HPstorageTom Good!!!!!!

Ok, xenserver don't suport MPIO using p4000. FINISH!!

I can modifi any parameter on p4000 or xenserver to maximize performance? I an using lacp in all bonds nic (Storages and XenServer's).

Because is much dificult my xenserver use all performance network...

Thanks!!!!