allanp81 Posted January 20, 2020 Share Posted January 20, 2020 For some reason, since upgrading to 6.8.x it seems that my server becomes somewhat unresponsive if I copy large files (10Gb+) across my network to it. I have tried copying to a share that has cache enabled and also to a share with no cache enabled. I see the same symptoms whereby for about the first 20-30 seconds it copies at ~110Mb/sec, maxing out a gigabit connection but then starts to drop massively, sometimes down to single digit Mb/sec. While this is happening it becomes almost completely unresponsive at times. Dockers become inaccessible, or just take a very long time to respond. I know I need to attach my syslog, but there are literally zero errors stated in it. Quote Link to comment
JonathanM Posted January 20, 2020 Share Posted January 20, 2020 27 minutes ago, allanp81 said: I know I need to attach my syslog, but there are literally zero errors stated in it Actually we need the diagnostics zip file in its entirety, not just the syslog. Quote Link to comment
allanp81 Posted January 20, 2020 Author Share Posted January 20, 2020 OK, see attached. downloadbox-diagnostics-20200120-2205.zip Quote Link to comment
allanp81 Posted January 20, 2020 Author Share Posted January 20, 2020 Doing some more testing, and it seems that while copying to a non-cache enable share still slows down significantly, it doesn't seem to cripple the server like copying to a cache enabled share does. It still suffers exactly the same speed drop from ~110Mb/sec to anywhere between 10 and 50 meg/sec, it fluctuates constantly. Quote Link to comment
allanp81 Posted January 20, 2020 Author Share Posted January 20, 2020 Here is a visual example of what happens: As you can see, it starts of maxed out and then just proceeds to drop and then go all over the place. Quote Link to comment
allanp81 Posted January 20, 2020 Author Share Posted January 20, 2020 Interestingly as well, copying from the same client machine to my win10 VM on my server copies at ~110meg consistently until the copy finishes. The VM lives on a btrfs pool created outside of the array and mounted using the unassigned devices plugin so I know that there is no issue with the network cabling etc. Copying from the VM then to array then exhibits the same issue with the slowing down and then inconsistent speed fluctuations. Quote Link to comment
JorgeB Posted January 21, 2020 Share Posted January 21, 2020 As you can see, it starts of maxed out and then just proceeds to drop and then go all over the place. This suggests the Lan is working correctly since first GBs are cached to RAM but then the device being written to can't keep up. Quote Link to comment
allanp81 Posted January 21, 2020 Author Share Posted January 21, 2020 I can understand that if copying to a share without cache enabled, I'd expect to see 40-50Mb/sec, but would it not be consistent rather than fluctuate so much? Also, when writing direct to cache I see exactly the same symptoms but it's worse in that it seems to basically cripple my server until sometime after the copy finishes! I don't recall seeing these issues prior to 6.8.x. I also see similar when backing up to my other server which is also now running 6.8.x and running on entirely different hardware with no cache drives specified. Quote Link to comment
JorgeB Posted January 21, 2020 Share Posted January 21, 2020 6 minutes ago, allanp81 said: but would it not be consistent rather than fluctuate so much? Not always, what if you write directly to the array with turbo write enable? Quote Link to comment
allanp81 Posted January 21, 2020 Author Share Posted January 21, 2020 (edited) @johnnie.black So by switching from "Auto" to "reconstruct write" seems to solve the issue when copying directly to the array, doesn't fix the issue with cache though obviously. This also seems to solve the issue of slow/intermittent speeds when backing up to my other server. Is there a potentially that my cache drives have issues even though nothing is reported as being wrong? I have 2 drives that aren't the same make but are the same size. Edited January 21, 2020 by allanp81 Quote Link to comment
JorgeB Posted January 21, 2020 Share Posted January 21, 2020 51 minutes ago, allanp81 said: So by switching from "Auto" to "reconstruct write" seems to solve the issue when copying directly to the array, doesn't fix the issue with cache though obviously. This suggests like mentioned that is the device(s) the can't keep up with writes, not all SSDs are super fast, also make sure they are trimmed regularly. Quote Link to comment
allanp81 Posted January 21, 2020 Author Share Posted January 21, 2020 Unless I'm being stupid, surely they can handle more than 50Mb/sec and not fluctuate so much and also not cripple the server at the same time? Quote Link to comment
rl2664 Posted January 21, 2020 Share Posted January 21, 2020 I had a WD green 120Gb as cache drive and I saw the same results. Changed to a Samsung 860 Evo 250 GB and all problems were gone. Perhaps you have the chance to test another ssd. Quote Link to comment
JorgeB Posted January 21, 2020 Share Posted January 21, 2020 9 minutes ago, allanp81 said: surely they can handle more than 50Mb/sec and not fluctuate so much Not what it looks like, try just one of them at a time or a different model, even a disk should be faster Quote Link to comment
allanp81 Posted January 21, 2020 Author Share Posted January 21, 2020 (edited) I have a mix of disks, could that be the problem? One is a Samsung SSD 850 PRO and the other is a SanDisk SD8SB8U512G1001. Running Trim didn't help either, I'll have to look into maybe replacing the drives then. I don't have anything spare though so if replacing doesn't fix I've wasted £100 or so. Edited January 21, 2020 by allanp81 Quote Link to comment
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.