Extreme bad reading performance with MPIO enabled

[PeterH]

Hi All,

i'm just making my first steps with StarWind and i'm trying to solve a problem with my HA configuration. As described in the numerous articles i built a new HA SAN following all the recommendations. I used the following hard- and software configuration:

Starwind HA:
- INTEL S5000 based Serversystems with single XEON 5335 CPUs, 1. system with 12 GB RAM, 2. system x 8 GB RAM, two onboard nics
- Adaptec RAID 51645 16-port SAS/SATA controller
- 2 mirrored system disks (500 GB SATA Western Digital RE3)
- 10 data disks as RAID6 combined (1 TB Western Digital RE3)
- additional INTEL Pro/1000 PT dual port nic
- Windows Server 2008 R2 64bit with latest updates
- StarWind iSCSI SAN 5.5 (Build 20101201)

MS Hyper-V Cluster
- INTEL S5000 based Serversystems with double XEON 5404 CPUs, 32 GB RAM, two onboard nics
- additional INTEL Pro/1000 PT dual port nic
- Windows Server 2008 R2 64bit with latest updates

I set up 3 different networks.
1. - 192.168.220.0 / 24 as LAN
2. - 192.168.230.0 / 24 as iSCSI-Net 1
3. - 192.168.231.0 / 24 as ISCSI-Net 2 (not in use for iSCSI-transfers right now)

As switches for these three networks i used relatively simple 24-port LevelOne Switches GWS-2476. All switches are configured to allow / ues jumbo frames.
Additionally both StarWind servers are connected with a crosslink cable for the heartbeat.

And here is my problem:
When i build a new target on one of my StarWind servers without HA config and without mpio settings i receive write- and read-performance around 100 mb per second. The transfer speed is nearly identical in both directions. Imho absolutely ok for a single gb-connection.

If i build a HA target and configure mpio i can see both connections (1. SW-server 192.168.230.1 and 2. SW-server: 192.168.231.2) tranferring data to both StarWind servers. My writing performance is as good as i expected it but when i try to read data from the san i receive only 4 - 5 mb per second. With my non HA config i see much more speed.


I tried several changes in my configuration but nothing seems to help:
- latest firmware for all INTEL nics
- jumbo frames on / off
- all recommended TCP/IP setting
- experimenting with serveral TCP-offload settings
... and so on...


Right now i have no more ideas what is responsible for this behaviour and i really need some hints what i can do

Best greeting from germany

Peter

[Max (staff)]

Hello Peter,
I think the problem is hiding inside your MPIO configuration.
Can you confirm that the iSCSI initiator is recognizing both sessions active when showing the device properties window

Mit Freundliche Gruesse aus der Ukraine

[PeterH]

Hello Max,

thank you for your quick response

Yes, i can confirm that both network connections (to SW-Server 1 and to SW-Server 2) are in use when i'm writing or reading data from the storage.

Yes, i can confirm that i can see both devices in my mpio configuration using RoundRobin. So far i thought that would be ok.

But sure i can have made some configuration mistakes.
What can i / can we do? Is it possible to establish a remote control session to look at this together?

Peter

[Max (staff)]

Yes, this is absolutely possible.
Please drop us an email with a "from forum" mark in the topic
support@starwindsoftware.com
I will send you an invitation in reply.

Mit freundliche Gruesse aus der Ukraine