I finally managed to get everything working on my array. I don't have a cache drive installed yet, however I am experiencing some slower write speeds then I was expecting to see based on the post "To Cache drive or not to Cache drive?"

 

In that thread, there was an average rate of 20-30 MB/s write, the peak being reported at 40 MB/s* for individuals who did not have a cache drive.

 

I am currently only getting 14 average MB/s write on my array shares through Windows 7 / SMB. I am fairly confident that it must be a bottleneck on the array itself, as the media that I am transferring off of my HTPC can transfer much quicker between my external HD and the native HD that I currently have hooked up to the HTPC.

 

I must be missing something and was hoping the community could help me out. Or is this normal? Hardware specs listed below:

 

5 3TB Western Digital Green (WD30EZRX)

3x Supermicro AOC-SAS2LP-MV8

Intel i3-2120 (3.3GHz)

Supermicro MBD-X9SCM-F-O

8GB Kingston (DDR3 1333)

Corsair AX850

OS - unRAID Server Release 5.0-rc6-r8168-test

 

Thanks.

I finally managed to get everything working on my array. I don't have a cache drive installed yet, however I am experiencing some slower write speeds then I was expecting to see based on the post "To Cache drive or not to Cache drive?"

 

In that thread, there was an average rate of 20-30 MB/s write, the peak being reported at 40 MB/s* for individuals who did not have a cache drive.

 

I am currently only getting 14 average MB/s write on my array shares through Windows 7 / SMB. I am fairly confident that it must be a bottleneck on the array itself, as the media that I am transferring off of my HTPC can transfer much quicker between my external HD and the native HD that I currently have hooked up to the HTPC.

 

I must be missing something and was hoping the community could help me out. Or is this normal? Hardware specs listed below:

 

5 3TB Western Digital Green (WD30EZRX)

3x Supermicro AOC-SAS2LP-MV8

Intel i3-2120 (3.3GHz)

Supermicro MBD-X9SCM-F-O

8GB Kingston (DDR3 1333)

Corsair AX850

OS - unRAID Server Release 5.0-rc6-r8168-test

 

Thanks.

 

You don't state what you are using to transfer the files.  I.E., Windows Explorer or Teracopy, etc.  You also don't say what kind of files you are transferring.

 

The reason I am pointing this out is that I have observed that when back up the user data files from my Windows 7 computers using Teracopy, the speeds drop down to the range that you mention.  When I am generating either a Bluray of Standard DVD ISO using ImgBurn, the speed steady-state speeds are in the 40MBps range.  I suspect that file size and the number of files has a major impact on transfer speeds---  A transfer with a large number of small files and numerous directories which have to be created have a lot of system overhead which will impact the transfer speed. 

I appreciate the reply,

 

I am transferring large ISO / VOB / and MKV files through Windows Explorer. I simply added my user shares on my array as a network resource through my Windows 7 HTPC. Some of the MKV's are around 20 GB. As I was testing this, I did one file at a time.

 

At one point I did begin another transfer as one was already occurring, when I did this, it simply split the 14 average MB/s that I am receiving between the two file transfers.

It may be that there is a bottleneck within your server but, before you go too far down that avenue, I would suggest that you check out the health of your network connection between unRAID and your Win7 machine by using iperf (available as a package under unMENU).

Great point, I will give that a shot first.

 

I originally had one of my IP phones in between my array and the network, I forgot that the phone was rated for only 10/100. So that was my original bottleneck, but once I removed the phone, the array negotiated 1000 and I was seeing an increase from 10 MB/s (when I was negotiated at 100) to the 14 that I have now. So although I am sure that I have gigabit connectivity now, I suppose there might be driver issues on the NIC's, or God help me no cabling issues.

 

I will update the post after iPERF. Its nice that unRAID has it as a package.

 

Thanks.

Alright, so iPERF is reporting back a decent 78 MB/s average, so its not the network path. Could it be the WD30EZRX's?

Enter "hdparm -tT /dev/[hs]d?" to test disk speeds.

/dev/sda:

Timing cached reads:  12630 MB in  2.00 seconds = 6324.36 MB/sec

Timing buffered disk reads:  70 MB in  3.07 seconds =  22.83 MB/sec

 

/dev/sdb:

Timing cached reads:  13058 MB in  2.00 seconds = 6539.14 MB/sec

Timing buffered disk reads: 384 MB in  3.00 seconds = 127.91 MB/sec

 

/dev/sdc:

Timing cached reads:  12996 MB in  2.00 seconds = 6507.87 MB/sec

Timing buffered disk reads: 356 MB in  3.01 seconds = 118.08 MB/sec

 

/dev/sdd:

Timing cached reads:  12800 MB in  2.00 seconds = 6410.12 MB/sec

Timing buffered disk reads: 356 MB in  3.01 seconds = 118.36 MB/sec

 

/dev/sde:

Timing cached reads:  12798 MB in  2.00 seconds = 6408.80 MB/sec

Timing buffered disk reads: 350 MB in  3.01 seconds = 116.13 MB/sec

 

That looks fine to me, however does that command only result in HD reads?

Whenever trouble shooting, I always like to rule out the obvious, silly stuff, so, just to be clear, you *are* transferring from drive on PC directly to share, right?

 

I ask this because, transferring *between* shares from your PC would halve your transfer speeds, as the data would travel from share to pc to share...

 

I know, I know, but it never hurts to confirm these things before getting into the more complex tests etc...

 

 

 

Absolutely yes, I am transferring from my drive on my actual HTPC, to the share. Not in between.

sda is slow

sda is my flash drive.

 

I am still digging around, any suggestions would be appreciated.

sda is my flash drive.

You stated you have 5 hard disk drives, but you only posted benchmarks for 4 of them.

Correct, that was my mistake, I currently have 4 in the array, not 5. One needed to be sent back RMA.

does this mean you are operating in simulated disk mode? that would surely have an impact on performance i bet.

Hmmm, I am not quite sure what Simulated Disk mode is, could you educate me? Pretty new to storage.

Simulated disks are what allows unraid to rebuild a single failed drive, and you can continue to use the array (not advisable) while the drive is failed or missing. Post a screen shot of your drive assignment page to clear up any confusion.

Absolutely, I appreciate all of the help.

 

See attached.

did you load a NTFS driver for some reason?

No, I am not sure how that was loaded or what it is being used for exactly. That seemed to have occurred at 5:17, at that time I probably started transferring more files. It certainly was not manual.

 

Think its causing a problem?

can you post your syslog for dissection please

I appreciate it, hopefully it will show something, as typically when I am transferring nothing really shows up in the syslog.

 

In the meantime, I am going to mess around with iPERF some more just in case I missed something on my initial test.

syslog-2012-08-16.txt

I am starting to think that this is more of a matter of SMB not utilizing the available network bandwidth that I have. Wireshark shows that the SMB 2.0 protocol is running which I know offers some improvements over 1.0. I just feel like bottleneck is due to the Windows file sharing.  This would explain when using Iperf, it actually used my available bandwidth.

 

How have others handled this limitation? I would think that SMB 2.0 that Vista and Windows 7 offer can utilize more then what I am actually receiving at this point. Since I had a Cisco VPN Client installed, that automatically sets my MTU to 1300. I have since changed it back to 1500 for these transfers and have since received an average of 16 MB/s instead of the 14 MB/s that I was receiving before. But I should still see much better performance than this.

 

I know in Windows XP you can turn on Large System Cache to get much faster file transfers, but I am not sure with Windows 7.

I strongly suspect that the slow writing speed is due to some tiny glitch between the Unraid version you chose and the controllers you use (sas2lp).

 

Put the disks to the motherboard SATA ports and try briefly to see the speed....

 

Suggest to use Teracopy or md5 or you can end with some errors along the way

Teracopy will provide you with the actual speed

Well you were 100 % correct, that was the problem. That pretty much sucks for me, since the unRAID version that I am running on seems to be the only version that will even work with SAS2LP controllers.