0

I am setting up a Windows 2012 R2 Storage Spaces array. It is setup with 24 drives, 22 of them mirrored, with 2 hot spares, in 11 columns. I am useing a LSI 9207-8e HBA card with the latest IT firmware to connect to a Storage Spaces certified JBOD.

My read rate is scaling nicely, but I cannot get much more then 1 drive worth of write rate out of the system.

With 1 Drive mapped directly to the system I get 200MBs read, and 100MBs write.

With 22 Drives Mirrored with Storage Spaces I get 1300MBs read and 100MBs write.

With 3 Drive striped I get 600MBs read, and 120 MBs write.

It almost feels like I am doing a parity instead of a mirror...

I have --

Applied all the latest Windows Updates

Updated the HBA's firmware and bios

Changed the HBA's boot availability, cycling between all 3 options

Tried it with 2 different servers/HBAs, but exact same hardware

Tried it with Multiple SFF-8088 cables with MPIO, and only 1

Played with drive write caching in windows.

Messed with both Interleave and allocation size.

I have tried configurations in 2,4,8 and 11 columns

I am at my wits end here...

I have another Storage Spaces Mirror in another server with the same HBA with much slower drives, and I have no write rate problem on it at all.

If anyone has any ideas they would be very much appreciated.

Litzner
  • 405
  • 1
  • 5
  • 17
  • I'm in the middle of my own storage spaces testing right now and have definitely seen my share of strange things happening. Two things come to mind, 1. How are you doing your testing? and 2 maybe try changing the columns to 8? In one of the MS presentations the speaker mentioned hes seen some strange behavior when columns exceed 8, maybe this is one of those times. – R.K Aug 28 '14 at 19:19
  • So far I have just been testing directly on the volume on the server, using Crystal Disk Mark. I have tried configurations in 2,4,8 and 11 columns. (added that to the list of things I have tried) – Litzner Aug 28 '14 at 19:22
  • Physical disks are all the same size? and virtual disk is fixed provisioned rather than thin? – R.K Aug 28 '14 at 19:28
  • Yes physical disks are all exactly the same model, and the VD fixed to the max size of the pool. – Litzner Aug 28 '14 at 19:29
  • While the test is running, what do you see in Resource Monitor / Disk / Storage? Is any disk saturated (huge queue length)? Depending on how Storage Spaces is spreading the writes and waiting for confirmation from the mirrors, one disk could be slowing everything down. Not really specific to Storage Spaces (something I used to see with almost-dying disks and ZFS on Solaris). – gtirloni Sep 06 '14 at 22:14
  • Under there I can only get it to show me the volume created of all the physical disks, I can't get it to show me what is happening on each individual disk. – Litzner Sep 08 '14 at 19:17
  • Did you ever solve this? – pauska Mar 09 '15 at 12:21
  • No, but I have found in my testing that it is not necessarily an issue with Storage Spaces, as I have tried other storage products and I am experiencing the same issue. If I had to guess it is an issue with my HBAs or the JBOD unit itself. – Litzner Mar 10 '15 at 12:48
  • In several cases of similar slowness, we narrowed it down to the Shadow Copy service. Disabling Shadow Copy solved the problem or at least improved write speed significantly, especially where large files were involved. However, we never figured out exactly how that works, and also you lose backup capability when shadow copies are off. – Elena of ReclaiMe Sep 29 '15 at 13:34

0 Answers0