November 16, 201015 yr Hey Folks, When I first put together my unRaid box, I used an old drive that was just laying about as a cache drive. I was finding (especially after the speed increase we got awhile ago) that transfer speeds were pretty much the same with or without it, so I removed it. Fast forward to this week. I purchased some new 2TB EARS drives and figured I should set one up as a cache/warm spare. I assumed I would see faster transfer speeds over my gigabit network because of the newer, faster cache drive. It appears to be functioning properly (data is being copied to the cache drive first), but speeds are pretty much identical to what they are when transferring directly to the array... This can't be right! I'm looking at transfer speeds of around 35 MB/s with or without the cache drive. Cache drive transfers *may* get up to 40 MB/s on occasion, but nowhere near what I was expecting to see (50 MB/s - 80 MB/s). Anyone have any ideas what's going on? Is this performance what I should expect to get? Doesn't seem worth using a cache drive if I'm not going to see significant performance increases... Thanks in advance for any thoughts/observations!
November 16, 201015 yr Since the EARS drives are 5400 rpm not 7200 that could be a possible reason. But I would let the experts way in rather than trust my opinion.
November 16, 201015 yr Your source drives or computer may be the issue. How fast are they? Also, are the EARS jumpered?
November 16, 201015 yr It sounds like one of 2 issues. 1. You have your cache drive connected to an expansion card and not directly to the motherboard. (I used a PCI based sata card in my first build and those were the same speeds I was seeing.) 2. It is the performance of the Ears jumpered on unraid. (having never owned an Ears due to my lack of faith in WD I can not state for sure. Easy test, put unmenu on there and run the user script for disk test, that should tell you the read performance of the disks. If the read is low it's an interface issue as jumpers and advance format should mostly affect write speeds) I use a Samsung 502HJ for my cache drive. It's connected to a PCI-E sas card, and I get +/- 140 MB/s using the unmenu script. My peak Network transfer to my server is 91 MB/s, average is 62 MB/s. I am connected using SMB off an OSX machine.
November 16, 201015 yr Author Since the EARS drives are 5400 rpm not 7200 that could be a possible reason. But I would let the experts way in rather than trust my opinion. That's a good point... I'd think performance would be a bit better though, since when I precleared I was getting speeds from 45 MB/s right up to 105 MB/s.... (Averaging around 70-80) Your source drives or computer may be the issue. How fast are they? Also, are the EARS jumpered? The EARS are jumpered, and the source drives are 7200 RPM Seagates... I'll try a transfer from my SSD to make sure this is not it... It sounds like one of 2 issues. 1. You have your cache drive connected to an expansion card and not directly to the motherboard. (I used a PCI based sata card in my first build and those were the same speeds I was seeing.) 2. It is the performance of the Ears jumpered on unraid. (having never owned an Ears due to my lack of faith in WD I can not state for sure. Easy test, put unmenu on there and run the user script for disk test, that should tell you the read performance of the disks. If the read is low it's an interface issue as jumpers and advance format should mostly affect write speeds) I use a Samsung 502HJ for my cache drive. It's connected to a PCI-E sas card, and I get +/- 140 MB/s using the unmenu script. My peak Network transfer to my server is 91 MB/s, average is 62 MB/s. I am connected using SMB off an OSX machine. It *is* connected to an expansion card. I hadn't considered that. I'm using an SIL3132 2 port PCI-e controller from Monoprice. With 2 drives on there, I figured I should still be getting 125 MB/s (theoretical max of course)... Mind you, the fact that I'm getting pretty much the identical speeds as before leads me to believe maybe this is the issue. I'll test it with the unMenu, then I'll plug it directly into a mobo SATA header tonight and see if that helps. I sure would love to see 60 - 70 MB/s Network transfers! Thanks for all the suggestions! I'll repost with an update.
November 16, 201015 yr Another thing.....after selecting the cache drive and starting the array, I stopped the array and then restarted. After doing that I saw my speeds double and stay around 70MB/s.
November 16, 201015 yr Author Another thing.....after selecting the cache drive and starting the array, I stopped the array and then restarted. After doing that I saw my speeds double and stay around 70MB/s. Interesting! I never did that. I'll try that as well!! Was it working before you did that (as far as copying files to it first)
November 16, 201015 yr Oddly enough, I was using a 3132 card as well. My cache drive didn't work at all till after I reboot the server.
November 16, 201015 yr Another thing.....after selecting the cache drive and starting the array, I stopped the array and then restarted. After doing that I saw my speeds double and stay around 70MB/s. Interesting! I never did that. I'll try that as well!! Was it working before you did that (as far as copying files to it first) It was working but the speeds were no different than if I copied straight to the array.
November 16, 201015 yr Author That sounds exactly like what I'm experiencing!! Ok. Sounds like the plan is: 1. Test speeds in unMenu 2. Stop Array. Start Array 3. Test Speeds. 4. I'll throw in a reboot as well (In case it's a 3132 card thing like Giraffeninja) 5. Re-test. [edit] 6. Test using SSD as source drive If speed does not improve: 1. Stop array and swap cache drive to onboard SATA 2. Test again! Nothing like having a plan!
November 16, 201015 yr If your source drive can't read more then 35 MB/s, then you should stop connecting them using USB...
November 16, 201015 yr Author If your source drive can't read more then 35 MB/s, then you should stop connecting them using USB... There's always a smart-a$$ in the audience, eh?
November 17, 201015 yr Author That sounds exactly like what I'm experiencing!! Ok. Sounds like the plan is: 1. Test speeds in unMenu 2. Stop Array. Start Array 3. Test Speeds. 4. I'll throw in a reboot as well (In case it's a 3132 card thing like Giraffeninja) 5. Re-test. [edit] 6. Test using SSD as source drive If speed does not improve: 1. Stop array and swap cache drive to onboard SATA 2. Test again! Nothing like having a plan! Well. That was disappointing . Tried all the above steps, and nothing changed my transfer speed. Makes me wonder if my switch has issues...
November 17, 201015 yr Try this command (to see if the speed issue is with the drive, or with the network) dd if=/dev/zero of=/mnt/cache/test_file.dat bs=1M count=200 It will write a 200 Meg file to /mnt/cache and print the timing statistics when done. You can remove the file it creates once it is done with rm /mnt/cache/test_file.dat
November 17, 201015 yr Well. That was disappointing . Tried all the above steps, and nothing changed my transfer speed. Makes me wonder if my switch has issues... Have you updated the BIOS? Why you do not post your complete system configuration and syslog in case you may have some gremlins in the system. Make sure you are using AHCI for the SATA drives (all of them - many BIOSes have at least two setting here - one for the first four SATA drives and one additional for the rest) I also have a 790GX based motherboard and I recall having the new 3-platter 2TB jumpered EARS drive starting the preclear at over 130 MB/s
November 17, 201015 yr Sadly I am not sure about the whole AHCI thing. I use a GB H57M-USB3 for one of my servers. With AHCI and a Cache drive I get kernel panics when ever I try to write to the cache. With ACHI disabled everything is fine and I get 62 MB/s average 92MB/s peak transfers to the drive.
November 17, 201015 yr Author Sorry guys, haven't been able to address the new suggestions yet! I'll run that code tonight Joe, and I'll post system config and syslog as well confirm status of AHCI and whether it kernel panics with it or not... At the end of the day, it's not really a huge deal whether it runs at 60MB/s or 35. I think it's more of a "I know it should be able to do this, so why isn't it?"... Any chance it could be the network card on my PC's mobo? Asus p5e-vm hdmi. Also, one thing I noticed is that share to share transfers tend to be a bit faster as well. Files transfer at 40 MB/s (20 MB/s up and 20 MB/s down). Not sure if that data has any relevance, but figured I'd throw it in the pot as well
November 18, 201015 yr Author Try this command (to see if the speed issue is with the drive, or with the network) dd if=/dev/zero of=/mnt/cache/test_file.dat bs=1M count=200 It will write a 200 Meg file to /mnt/cache and print the timing statistics when done. You can remove the file it creates once it is done with rm /mnt/cache/test_file.dat Ok. So I ran it 5 times. Speed look good: Tower login: root Password: Linux 2.6.32.9-unRAID. root@Tower:~# dd if=/dev/zero of=/mnt/cache/test_file.dat bs=1M count=200 200+0 records in 200+0 records out 209715200 bytes (210 MB) copied, 1.72121 s, 122 MB/s root@Tower:~# dd if=/dev/zero of=/mnt/cache/test_file2.dat bs=1M count=200 200+0 records in 200+0 records out 209715200 bytes (210 MB) copied, 1.84673 s, 114 MB/s root@Tower:~# dd if=/dev/zero of=/mnt/cache/test_file3.dat bs=1M count=200 200+0 records in 200+0 records out 209715200 bytes (210 MB) copied, 1.68995 s, 124 MB/s root@Tower:~# dd if=/dev/zero of=/mnt/cache/test_file4.dat bs=1M count=200 200+0 records in 200+0 records out 209715200 bytes (210 MB) copied, 2.38996 s, 87.7 MB/s root@Tower:~# dd if=/dev/zero of=/mnt/cache/test_file5.dat bs=1M count=200 200+0 records in 200+0 records out 209715200 bytes (210 MB) copied, 1.60997 s, 130 MB/s
November 18, 201015 yr Author Alright.... So I'm not sure what changed, but I just tested another file and suddenly my average transfer rate is 50 - 60 MB/s!!! Wish I could explain what changed. All I've done since last test was create those test files from the post above, copy out the SysInfo, looked at the logs, and tested transfer speeds from another PC... Anyways, thanks for all the suggestions and advice! Looks like my cache drive has decided to play nice! Cheers, DB
Archived
This topic is now archived and is closed to further replies.