Slow Mailbox (Migration) Throughput and HP NIC Drivers

Hewlett-Packard_logo-web[1]A small post on an issue I recently encountered when doing preparations for a mailbox migration. The system was an HP Proliant DL370 G6 system, prepared and configured with the OS and Exchange by the customer’s IT department.

Things looked OK and we were going to perform some test migrations to get an throughput estimation for this configuration  to help us organize migration batches. To our dismay, speeds were way lower than we were used to see with similar configurations; mailboxes were migrated with an average speed of 7-8 MB/min, where we used to see something in the 50-120 range.

A quick look on the performance monitor didn’t show anything out of the ordinary, except for very low downstream network throughput. With the servers (the original Exchange server as well as the new one) being on the same subnet and physically next to each other, networking components were also not deemed suspect (other servers on same switch were not experiencing this issue).

I then tried something simple: copying near 100 MB of files from the source Exchange server to the new one. It went at an ridiculous slow speed of 60-80 KB/s. Copying those same files from the new server to the source server was instantly. I verified this against a vacant server on the same switch; copying from and to the source Exchange server on that server was instant, both up- and downstream.

So, if SMB was having trouble getting packets across, that could explain the slow mailbox migration speeds. Attention shifted to the networking configuration on the new Exchange server, which was equipped with a HP NC375i Integrated Quad Port Multifunction Gigabit Server Adapter. I checked the driver version of one of the NC375i’s instances through Network Connections > Properties (of instance) > Configure > Driver (tab). It reported QLogic Corp. driver 4.7.17.926 (qlxgnd64.sys) was used.

After some searching on HP’s support site I discovered an advisory which could apply to my situation as it applies to the same qlxgnd64.sys driver version 4.7.17.926: c03734205, “Advisory: HP NC Network Adapters – Certain HP NC-Series Network Adapters May Experience Very Slow Bandwidth During Large File Transfers on Windows Server 2008 and Windows Server 2008 R2”.

The advisory gives the option to either keep the driver and disabling Large Receive Offload (LRO) or to upgrade to driver version 4.7.18.131. We choose the latter:

image

After upgrading the driver, we moved a mailbox and and throughput speeds were within the expected range again as we found out when producing a quick stats report using the cmdlet (Exchange 2010):

Get-MoveRequest | Where { $_.Status -eq "Completed" } | Get-MoveRequestStatistics | Select DisplayName,TotalMailboxSize,TotalMailboxItemCount,@{n="Speed MB/min"; e={ [int]($_.BytesTransferred.ToMB() / $_.TotalInProgressDuration.TotalMinutes) }}

image

In my opinion, it’s another fine example of the value of testing and validating your configuration and any amendments you make before putting them in production and be cautious with what I call “blindly updating” of system components such as drivers or driver packs (e.g. HP’s SPP or Service Pack for ProLiant).

If you don’t have the luxury of a test- and acceptance environment, just as with Service Packs, Rollups and Cumulative Updates, have a waiting period and check the vendor’s support site for any reported issues before implementing updates yourself; according to this discussion on the HP support forum, the issue with the 4.7.17.926 QLogic driver existed for quite some time.

5 thoughts on “Slow Mailbox (Migration) Throughput and HP NIC Drivers

  1. Hi Michael,
    I think the right syntax should be:

    Get-MoveRequest | Where { $_.Status -eq “Completed” } | Get-MoveRequestStatistics | Select DisplayName,TotalMailboxSize,TotalMailboxItemCount,BytesTransferred,TotalInProgressDuration,@{n=”Speed MB/min”; e={ [int]($_.BytesTransferred.ToMB() / $_.TotalInProgressDuration.TotalMinutes) }}

    Roberto

  2. Besides disabling LRO, its always good to check some networksettings. Like Jumbo Frames, Flow Control (both if SAN’s are involved) and Disabling TCP Nagle and DelayedAck could be giving an improvement.

    René (Cujo ;-) ).

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s