Array drastically slower than drive itself

[oliverdunk]

Hi,

 

I'm running on a HP Microserver, with the following processor: AMD Turion™ II Neo N54L Dual-Core @ 2200 MHz. I understand that this is fairly lower power, however I am surprised at just how poorly Unraid is running.

 

Right now, I have just a single 1.5TB hard drive in the array. Encryption is enabled using the xfs file system.

 

If I write to /mnt/disk1/, disk speeds are as expected (~84.0MB/s). However, if I write to /mnt/user/, I'm lucky to get a few MB/s. I'm not sure what the problem could be. I believe data written to both mounts is encrypted, so that can't be the cause. Additionally, since I only have one drive in the array, I would expect that there are no parity calculations for Unraid to do.

 

Any thoughts on how to troubleshoot this?

[JorgeB]

User shares always add some overhead, that CPU doesn't support hardware AES so really not a good option to use with encryption, try without and speed should be noticeably faster.

[oliverdunk]

@johnnie.black Appreciate the response! I'm reluctant to try that if possible, since I'd rather not format the drive.

 

Do you agree with what I'm saying about why I don't expect encryption to be the culprit? I can write to a drive encrypted using LUKS, but not managed by Unraid, at high speeds. It seems that the user share overhead is what's making the big difference. And the overhead is much bigger than seems reasonable.

[JorgeB]

3 hours ago, oliverdunk said:

Do you agree with what I'm saying about why I don't expect encryption to be the culprit?

No, I expect it to be considerably faster without encryption.

[oliverdunk]

1 minute ago, johnnie.black said:

No, I expect it to be considerably faster without encryption.

Could you elaborate? It would be nice to understand the thought behind the change before I make it. As I mentioned before, it seems like there's no extra work for Unraid to do. LUKS is handling the encryption, and I can write to that encrypted directory at high speed.

[JorgeB]

User shares will always add some overhead, sometimes just a little, other times a lot, not exactly clear why some users see a much larger impact than others, add to that encryption on a CPU without AES support and it compounds the problem, example:

 

You say copy to disk share with encryption is done at 84MB/s, without encryption you'll likely get close to gigabit line speed, 100MB/s+ (unless the disk is very old and slow), copying to the use share without encryption should also see a noticeable bump in performance, could be a little, could be a lot, like mentioned user share performance can vary a lot from user to user.

 

 

[oliverdunk]

37 minutes ago, johnnie.black said:

User shares will always add some overhead, sometimes just a little, other times a lot, not exactly clear why some users see a much larger impact than others, add to that encryption on a CPU without AES support and it compounds the problem, example:

 

You say copy to disk share with encryption is done at 84MB/s, without encryption you'll likely get close to gigabit line speed, 100MB/s+ (unless the disk is very old and slow), copying to the use share without encryption should also see a noticeable bump in performance, could be a little, could be a lot, like mentioned user share performance can vary a lot from user to user.

 

 

@johnnie.black At your request, I changed the array to have a single XFS formatted drive, without encryption. I restarted the array and made sure I went through the format process. I then ran `dd if=/dev/zero of=test.png bs=1G count=1 oflag=dsync status=progress`, writing to /mnt/user. Unfortunately, I see the same behaviour - my write speed is only about 1MB/s. Would love your suggestions as to what to try next.

 

Thanks for the explanation, by the way! I hope you appreciate that I wasn't trying to be rude. I was just trying to avoid a lot of work if it was a solution that there wasn't a sensible reason for trying. It's unfair to compare you to ISP support, as I trust you know your stuff, but it almost felt a little bit like the "Reset your router again!" advice you can sometimes get.

 

Finally, I have attached diagnostics. I know that can be helpful, and thought I would save you a message.

tower-diagnostics-20200608-1811.zip

[JorgeB]

dd is not a good way to test this, can you try doing a regular SMB transfer from your desktop (assuming using gigabit or better)?

[oliverdunk]

On 6/8/2020 at 6:19 PM, johnnie.black said:

dd is not a good way to test this, can you try doing a regular SMB transfer from your desktop (assuming using gigabit or better)?

Hi,

 

Thanks for the advice. I just wanted to follow up and say that having run some different tests, it seems like my disk speeds were actually ok. It does appear that dd was a bad way to test.

 

It seems that the speed issues I was having may have been related to Docker networking. I found uploading to my Nextcloud instance much faster when in a VM compared to a Docker container. However, I have since moved away from unraid, so will leave troubleshooting that for now.