1

I have a new Dell PowerEdge T320 server with 1 SSD non-RAID and 3 HDDs in RAID 5 all on one Perc H310 controller. RAM is 48 GB and I have Windows Server 2012 R2 Data Center installed on the SSD.

I initially was attempting to create a VHDX on the RAID. The VHDX to create was of fixed size (over a 100 GB) and attempts to do that through the GUI and PowerShell both failed with I/O errors and crashes.

I then, in an attempt to find the problem, tried a simple copy and paste of a 50GB VHDX from the SSD to the RAID. That also caused a crash. I tried further to copy a large file (120 GB) created using fsutil from the SSD to RAID, again fail.

I read somewhere that there was a problem with large files and caching when using the default copy command, so I got a suggested tool "terracopy" which has the option (default) to have system write cache disabled. That went on copying for some time before I decided to stop it since it seemed to be working.

My main need is to create a large VHDX, does anyone have a clue what is going on?

Edit: Additional details:

  • Ran Diagnostics, did not yeild anything useful.
  • Dell replaced the RAID controller, no change.
  • Installed Windows 8 to test the copy and paste, it went on copying for a relatively long time before I stopped it since it looked like there would be no crash.
  • Booted from Linux Puppy USB, did a copy paste of a 120GB file from the SSD to the RAID, no issues.
3m3sd1
  • 11
  • 1
  • 4
  • 1
    sounds like a hardware problem to me. It should be no problem copying file >100GB with normal OS copy commands. It seems the SSD is always involved, try to a copy operation without it. – Peter Hahndorf Feb 04 '15 at 08:01
  • @PeterHahndorf I tried coping from an external drive, still failed. – 3m3sd1 Feb 04 '15 at 08:11
  • 1
    Yeah. Change from "I have a server" to "I have a broken server". Smells like some hardware is off. ECC ram? If not - run a memory test. – TomTom Feb 04 '15 at 09:29
  • is the drive encrypted i.e. with bitlocker ? if so the destination needs to be using the same bitlocker encryption. – JJJJNR Feb 04 '15 at 10:41
  • @JJJJNR No encryption whatsoever, fresh windows installation. – 3m3sd1 Feb 04 '15 at 11:19
  • Run the dset report on the hardware, send to dell. – JJJJNR Feb 04 '15 at 11:48
  • @JJJJNR already sent the report and they found nothing, also I ran preboot diagnostics. The error was about the crash registered in event log or somewhere in Windows I guess. – 3m3sd1 Feb 04 '15 at 12:07
  • Dell also sent a technician that replaced the Perc H310 controller, and that did not solve the problem. – 3m3sd1 Feb 04 '15 at 12:08
  • Where are you starting the copying from? Have you tried using XCopy or RoboCopy from the terminal session of the Hyper-V server, or are you copying from a VM that resides on the host? Separate Hardware desktop/device? – MagnaVis Feb 13 '15 at 16:33

3 Answers3

1

This looks like the old Server 2008 issue they eventually fixed with release of the Dynamic Cache Service addon: http://www.microsoft.com/en-us/download/details.aspx?id=9258

In a nutshell reads and writes are cached to memory which eventually fills it and makes the server unresponsive.

I couldn't find anything regarding server 2012 and this, but the behaviour looks the same. In the past I've seen Robocopy cause the same issue.

The XCOPY /J unbuffered copy bypasses this issue.

Paul
  • 11
  • 1
0

Dell replace the controller without anything to say that it was the issue? Weird. Tell them the problem still exists and to replace everything. Proc/board/ram until fixed.

JJJJNR
  • 870
  • 6
  • 20
  • 32
  • Could it be a Windows problem? I successfully managed to copy the file using TeraCopy. Also, another version of Windows (Windows 8) did not seem to cause an issue. – 3m3sd1 Feb 04 '15 at 23:53
  • Can't see it being a windows issue, I copy large files a lot and have never come across this issue, you could try running process mon at the time of the copy and see where the process fails. – JJJJNR Feb 05 '15 at 00:04
0

Had a similar issue with a Dell controller. When the RAID5 virtual disk on the controller is set to write-through, the copy performance issue exists. When the virtual disk is switched to write-back, the problem disappears.

You could also try using XCOPY with the /J parameter to use unbuffered IO. This will avoid the file system cache overhead and prevent the file system cache from being effectively flushed by the large file data.

Vasco V.
  • 44
  • 3