Hi,

I am running a small server with 8 gigz ram and a cute little Intel Atom and a Samsung 120 Gigz SSD.

Linux niftypika 3.2.0-4-amd64 #1 SMP Debian 3.2.73-2+deb7u2 x86_64 GNU/Linux

Life was good, everything was okay. After 1 week of running this little guy my i/o load got up.

top - 01:49:49 up 9 days, 13:51, 1 user, load average: 8.00, 7.71, 7.45 Tasks: 133 total, 2 running, 131 sleeping, 0 stopped, 0 zombie %Cpu(s): 5.4 us, 3.5 sy, 0.9 ni, 39.0 id, 50.7 wa, 0.0 hi, 0.5 si, 0.0 st KiB Mem: 8186596 total, 6800764 used, 1385832 free, 16528 buffers KiB Swap: 14648316 total, 462764 used, 14185552 free, 3545444 cached

I download a lot of stuff, (normal tcp/ip, no p2p encrypted traffic etc) and things started normal. The server had no issue at all to copy the 5-800 Mbit traffic to disk. Now, every day things get complicated. Every day the i/o load gets worse.

IOTOP showed me what the problem was:

2361 be/4 root 0.00 B/s 0.00 B/s 0.00 % 99.47 % [flush-8:0] 1412 be/3 root 0.00 B/s 107.03 M/s 0.00 % 99.55 % [jbd2/sda4-8]

but it also showed me the system "only" wrote around 70 megz /sec and this is a SSD. (I checked it twice

Total DISK READ: 14.27 K/s | Total DISK WRITE: 65.75 M/s

I googled it and I really got no good answer for this behavior. Perhaps anyone has a clue what is wrong and most importently how I can fix it?

Thanks for reading

niftyhamster

Have you done fstrim on all the partitions?

1 members found this post helpful.

1. Make sure that you use a filesystem that is suitable for SSDs, like ext4, btrfs, xfs or jfs. These filesystems have TRIM support, which is necessary to keep good write performance.
2. Either mount the filesystem(s) with the discard option (which in some cases can decrease performance, especially when deleting large trees) or, IMHO the better option, set up a cron-job or systemd timer to run fstrim once in a while. This will make sure that unused blocks on the SSD are internally marked as unused, so that the SSD controller can keep up its work with good performance.

1 members found this post helpful.

YEAH! Thats it guys! Why does Debian not have any auto trim feature/cronjob/whatever?

But thank you very much for the help <3

I would think that since SSD's are pretty new, and automatic recognition of them has probably not been worked on extensively. This will probably get better with time. You might also want to set the "noatime" option in the fstab for each partition on the SSD.


Mine (only for the SSD) look like this:

<code>
/dev/disk/by-uuid/be5255b7-c6fa-4429-8c3d-fec5b2391266 / ext4 acl,noatime,user_xattr 1 1
/dev/disk/by-uuid/7e14d174-22ef-4633-8883-0823c6a55f0c /boot ext4 acl,noatime,user_xattr 1 2
/dev/disk/by-uuid/6a2fb00a-5d74-4f29-95d8-f09ef967733d /usr ext4 acl,noatime,user_xattr 1 2
/dev/disk/by-uuid/463d9ec2-8481-4120-9dcb-28bd78bef6cc /var/lib/libvirt/svt ext4 acl,noatime,user_xattr 0 0

</code>

1 members found this post helpful.

In the consumer end market around 6-7 years? As a old guy I really know what fragmentation is but I never even thought of it because Windows + Mac OS X are taking care of it without user interaction. One reason I got hit hard on this could be many crypto currency demons running in the background. The amount of fragmented databases must have been enormous ...

Thanks again

Glad you have it solved. Yes, it seems that trim is not fully automatic on many OS's still.