AOE problem on Server 2008

[MarkB]

When I increase the ethernet adapter MTU, windows gives me issues with an AOE disk.

Server is HP DL380 G4, disk is 2Gb GBT partition of a Coraid SR2461 10Gb raid5 disk.

copying/deleting files to the AOE disk is ok, but Disk Manager hangs, as does chkdsk, defrag, io meter etc.
Then server won't shutdown cleanly.


I set a dedicated Broadcom gigabit adapter to MTU of 9000

set [HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters] to
GlobalMaxTcpWindowSize = 0x01400000 (DWORD)
TcpWindowSize = 0x01400000 (DWORD)
Tcp1323Opts = 3 (DWORD)
SackOpts = 1 (DWORD)


changing the adapter MTU back to standard, and the issues go away, but performance isn't great.

Any suggestions ?

[anton (staff)]

AoE has nothing to do with TCP so your TCP related settings change whatever they change but not affect AoE. This is number one. Make sure ALL of your underlying hardware does support for 9K frames (switch, server-side NIC and so on). Try using cross-over cables instead of switches (at least for testing). And update CoRAID firmware to the most recent one. After you'll do it if you'll still have issues we'll dig a bit deeper into your problem. Thanks!

[MarkB]

The connection is a crossover cable, and the Coraid firmware is the latest.


I'll try 2003 R2 and see if that is better.

[anton (staff)]

Good. Now please check do you have latest drivers for your Windows machine. And are they reported as 9K issue-less ones.

The connection is a crossover cable, and the Coraid firmware is the latest.


I'll try 2003 R2 and see if that is better.

[MarkB]

On 2008, using latest HP driver.

Configured the NIC using HP Network Config util.

Set Jumbo Frames on and all other enhancements off.

After the reboot Starport has lost the drive, and the server hung trying to change the NIC settings & once again doesn't shutdown again.

On the NIC, the only bindings enabled are the Starether NDIS & ipv4

Has Windows 2008 been used successfully ?

[anton (staff)]

Yes.

[MarkB]

Reverted to 2003 Enterprise SP2.

It is working without the glitches, but the performance of 15Mb/sec with iometer is not great. Target volume 2Tb of 11 x 1Tb disk in Raid5.
I have set jumbo frames on the connected nic.

update:
tried HDTune instead.
Much better, with 70Mb/sec, which is better than the servers mirrored scsi os drive.

[anton (staff)]

1) With I/O Meter you should be careful setting test parameters - I/O size, number of workers, mixed reads & writes etc.

2) 2008 should work in any case. I'll ask QA to take a look.

[Bohdan (staff)]

Hi.
We have found the reason of this.
Windows Server 2008 has NDIS 6 while StarPort uses NDIS 5.1. Therefore NDIS 6 emulates NDIS 5.1 and this causes the performance degradation.
We are working on this issue.