How to reduce iSCSI MPIO failover time?

[cdj]

I am running Microsoft iSCSI Initiator 2.0 with MPIO configured. This
Windows 2003 server with one NIC accesses an iSCSI target LUN through two
Cisco MDS 9509/IPS paths. I was able to force a failover by disabling the
network interface (NIC) on one of the MDS 9509/IPS Directors . My issue on
the failover is the failover time. I was getting the following results (with
each step about a few minutes apart):

1. Disabled MDS(A) NIC -> I/O stopped for 56 seconds -> Enabled MDS(A) NIC
2. Disabled MDS(B) NIC -> I/O stopped for 35 seconds -> Enabled MDS(B) NIC
3. Disabled MDS(B) NIC -> I/O stopped for 35 seconds -> Enabled MDS(B) NIC
4. Disabled MDS(A) NIC -> I/O stopped for 54 seconds -> Enabled MDS(A) NIC
5. Disabled MDS(A) NIC -> I/O stopped for 35 seconds -> Enabled MDS(A) NIC
6. Disabled MDS(A) NIC -> I/O stopped for 35 seconds -> Enabled MDS(A) NIC

So I am getting a failover time of 56, 54, 35, and 35 seconds for the MDS(A)
path
and
a failover time of 35 and 35 for the MDS(B) path.

What parameter(s) determine this failover timing and how can I reduce it to
15 seconds.

Thank you,

Hon

[Kenny Speer]

There are a number of ways to reduce this time but all of them depend on
your storage provider. You should contact your storage provider and ask
them what their recommended settings are.

BTW, what target are you using?

~kenny

"cdj" <c...@discussions.microsoft.com> wrote in message
news:13C472B8-F5A3-4BB4...@microsoft.com...

[cdj]

I am using Cisco MDS 9509 Directors with IPS module (iSCSI to Fibre Channel
bridge/router) and an EMC Symmetrix F/C storage array. I have the same setup
on a Red Hat Linux server and was able to reduce the failover time to 10
seconds by changing the iSCSI Initiator parameter ConnFailTimeout to 10 in
/etc/iscsi.conf file. This parameter specifies the session timeout value
before failing the SCSI commands. I was hoping there is a similar timeout
Registry entry for Microsoft's iSCSI Initiator.

Thank you,

Hon

[cdj]

I found this article

http://technet.microsoft.com/en-us/library/aa997069.aspx

I will try around with this registry value

HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\Disk\TimeOutValue

Thanks,

[Kenny Speer]

The Disk class timeout value is the most misunderstood timer in windows
storage. If you change it, there is a very good chance you will lose
support from your storage vendor. If you make this value arbitrarily low,
you will cause more problems than you are trying to solve.

Again, without specific knowledge of what these timers do, you will most
likely create a scenario where your host will no longer survive faults.

To point you in the right direction, you need this key:
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4D36E97B-E325-11CE-BFC1-08002BE10318}\0001\Parameters

Although, the 0001 towards the end may be different on your system. This is
the SCSI Bus controller class and if you have more than 1 bus, this number
changes.

The only parameter that you might have luck with is: LinkDownTime, change
that lower for your cable pull test.

"cdj" <c...@discussions.microsoft.com> wrote in message

news:A48920FE-B829-4BEC...@microsoft.com...

[cdj]

Thank you Kenny,

I will try the LinkDownTime entry, it is currently set at f (or 15) seconds.

Thanks again,

Hon

[Kenny Speer]

Right, when you install the initiator, it creates that entire hive. It's a
new SCSI BUS so it creates all of those parameters.

That page is just listing what changes were made to the registry (i.e. a
whole bunch of keys were added).

Leaving them all alone is recommended unless your storage vendor suggests
you change them. But if you want to experiment then you can safely lower
that timeout. I recommend 5 seconds (long enough for someone who
accidentally pulled the cable to put it back).

~kenny

"cdj" <c...@discussions.microsoft.com> wrote in message

news:93A20525-929E-406B...@microsoft.com...
> Looks like Microsoft iSCSI Initiator (iSCSI-2.0-x86fre.exe) add this
> parameter to the registry:
>
> ADD LinkDownTime=dword:0000000f
>
>
> *** Info from McAfee Site Advisor:
>
> http://www.siteadvisor.com/sites/easystreet.com/downloads/3051368/
>
> I will leave it alone.
>
> Thank you,

[Ramin]

I am trying to reduce the MPIO failover time. I was able to bring it down to
21 sec from about 35 sec by changing the LinkDownTime value from 15 sec to 1
sec.
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4D36E97B-E325-11CE-BFC1-08002BE10318}\0000\Parameters

My data is streaming video from lots of cameras and 21 sec is still too
long, therefore I lose data. Is there anything else that could be done on
the initiator side(Windows 2003 Standard SP2 R2) to reduce this downtime?

[endeval2]

I think you could be suffering from what I found. Nagle's Algorithm for
iSCSI connections.

look at sustainablesharepoint.com iSCSI best practice, for something you can
try to improve speeds.