Performance issues - SMB/CIFS

After much benchmarking to choose the best filesystem and configuration for OMV to be a backup target for vRanger backups (large files), I chose XFS as the filesystem and serving it up with SMB. My existing vRanger server was backing up to a a different device (DataDomain 510) and was providing good performance, so I know the vRanger server is configured acceptably.

Having switched the backups to my OMV implementation, I was very disappointed to see abysmal performance, but I'm not sure why. I see my top process(es) is smbd. Here are a couple of the charts (since a picture is worth a thousand words), hoping someone might be able to offer some suggestions to make the OMV a viable alternative:

With regard to Memory Usage: why does the Diagnostic/Overview page show that only 9% of the memory is used, yet page cache has consumed almost all of it. Will increasing memory significantly increase the performance.

The hardware is SuperMicro X6DHP-8G

So, any suggestions on how to identify the problem here? I have the same hardware using Openfiler. Here's a comparison of the performance between OMV vs. OpenFiler on the same hardware. I'd really like to have OMV compete with OpenFiler, as it's much more polished and better supported.

I had had these options set, except for max protocol = SMB2

I wasn't sure whether these were to be on the Settings page, or the extra options for the Share itself.

Also, after setting those and Saving, wasn't sure if I needed to disable then reenable the settings for them to kick in.

Whatever, some mix of all that, I'm not seeing any noticeable benefit.

Comparing my "PD-DF140" (top) with "OMV" (bottom), this is telling:

[root@[b]pd-df140[/b] ~]# dd if=/dev/zero of=/mnt/vg-sda2/sda2-shares/output.img bs=3k count=256k conv=fsync
262144+0 records in
262144+0 records out
805306368 bytes (805 MB) copied, 10.5415 s, 76.4 MB/s

root@[b]omv[/b]:~# dd if=/dev/zero of=/media/5d27afc6-3886-4889-8f7f-4f831e781efc/output.img  bs=3k count=256k conv=fsync
262144+0 records in
262144+0 records out
805306368 bytes (805 MB) copied, 108.506 s, 7.4 MB/s

But I'm not sure what it's telling me. This is the same hardware, so why such dramatic difference in this performance test?

In a loosely related post here, I questioned why iSCSI would be so much slower than SMB. Little did I know that SMB was unacceptably slow.

Now we're talkin'! I'll give that a go and report back.
(hopeful)

BTW: I keep seeing these cool plugins and additional functionality. Badly needed is some interface to define monit alerts (thresholds). I've disabled the load average alerts because I was getting slammed by those alerts, but I'd like to be able to tweak them. I also would like to change the subject line that it sends. But my attempts to modify these alerts (following this post) haven't been successful.

Hmmm...
When I click on "Install Backports 3.16 kernel" it opens an "Install kernel headers..." with a Start button. ooookkkkkaaaay....so I press Start and it says "Updating...", but doesn't seem to do anything after that. Looking at top I'm not seeing anything evident running for this task. What should I be looking for? How do I know it's actually doing what is supposed to be doing? I've been waiting for quite a while already.

I stumbled through installing the 3.16 kernel manually. But sadly, it did not solve the problem:

root@omv:~# uname -r
3.16.0-0.bpo.4-amd64


root@omv:~# dd if=/dev/zero of=dd if=/media/947281ee-363e-455e-ac7d-fa42313a2962/output.img  bs=3k count=256k conv=fsync
262144+0 records in
262144+0 records out
805306368 bytes (805 MB) copied, 73.6799 s, 10.9 MB/s

But in the OMV-Extras, I see HWRaid repository. Maybe I'll try to find something compatible with my RAID controller.

I'm open for any other suggests though.

hwinfo showing 3ware driver for AMCC controller?

45: SCSI 100.0: 10600 Disk
  [Created at block.243]
  Unique ID: LUEV.+AaKD7U+Dk3
  Parent ID: GA8e.5VJ33KzxQQ8
  SysFS ID: /class/block/sdc
  SysFS BusID: 1:0:0:0
  SysFS Device Link: /devices/pci0000:00/0000:00:02.0/0000:01:00.0/0000:02:01.0/host1/target1:0:0/1:0:0:0
  Hardware Class: disk
  Model: "AMCC 9500S-4LP  DISK"
  Vendor: "AMCC"
  Device: "9500S-4LP  DISK"
  Revision: "2.08"
  Serial ID: "Z1E8XWLXDA772200856F"
  Driver: "3w-9xxx", "sd"
  Driver Modules: "3w_9xxx"
  Device File: /dev/sdc (/dev/sg2)
  Device Files: /dev/sdc, /dev/disk/by-id/scsi-1AMCC_Z1E8XWLXDA772200856F, /dev/disk/by-path/pci-0000:02:01.0-scsi-0:0:0:0
  Device Number: block 8:32-8:47 (char 21:2)
  Geometry (Logical): CHS 729454/255/63
  Size: 11718684672 sectors a 512 bytes
  Config Status: cfg=new, avail=yes, need=no, active=unknown
  Attached to: #31 (RAID bus controller)

Though, comparing to my openfiler install, looks like the same driver (and version) is being used (kernel version=2.6.32-131.17.1.el6-0.11.smp.gcc4.4.x86_64)

I suspect the answer to this problem is in here somewhere, but it's getting far too deep for me.

I hate to give up and "downgrade" to OpenFiler, I desperately want OMV to be my solution, but if we're out of ideas...

ryecoaaron, thank you so much for your attention and support. I'm hoping you still have some options/ideas for me.

Yes, I write to OF all the time. Yes, read speeds are similar. (and interesting...true...that my write speed would be faster than reads. Could just be how the tool is doing it's thing).

But because this (OMV) is to be a backup device, I need performance on the write speeds. What used to take about 20 minutes for a backup job to run, it was taking over 1.5 days, at less than 2Mbs.

Nah.. Since the dd command rules out any network traffic latency... Its a pure local device I/o test

I was able to boot into a 2.16 kernel of a LiveCD just to test and found that the performance was still bad. I checked the RAID controller to see that the write cache was enabled - it WAS NOT. When I tried to Enable it, it through a cryptic error. I then noticed that the BBU of the RAID controller is failed, so probably why I couldn't enable the write cache. Could *THAT* be why performance is so pathetic?!?