Slow write speeds in DomU on LSI SSD RAID, read ok

[Marvin Müller]

Dear all,

 

I have a hypervisor with (fully patched) XenServer 7.6 (E5-2620 v4, 32GB, LSI 9750-4i, 2x Samsung SM883 SSD). Looking at Dom0, I get reasonable read/write speeds:

 

[root@localhost ~]# dd if=/dev/sda of=/dev/zero bs=1M count=1024
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB) copied, 1.0037 s, 1.1 GB/s

[root@localhost ~]# dd if=/dev/zero of=/root/testfile bs=1M count=1024
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB) copied, 1.89972 s, 565 MB/s

[root@localhost ~]# hdparm -tT /dev/sda

/dev/sda:
 Timing cached reads:   21612 MB in  1.97 seconds = 10965.82 MB/sec
 Timing buffered disk reads: 3124 MB in  3.00 seconds = 1041.28 MB/sec

 

The controller does not have a BBU enabled. I did multiple runs, they result in pretty much equal speeds.

 

Now looking at a Ubuntu 16.04 VM:

root@server:/home/toor# dd if=/dev/xvda of=/dev/zero bs=1M count=1024
1024+0 records in
1024+0 records out
1073741824 bytes (1,1 GB, 1,0 GiB) copied, 1,18745 s, 904 MB/s

root@server:~# dd if=/dev/zero of=/root/testfile bs=1M count=1000
1000+0 records in
1000+0 records out
1048576000 bytes (1,0 GB, 1000 MiB) copied, 10,3562 s, 101 MB/s

The guest is HVM, installing Xen Tools did not help much. All Hypervisor settings are default, same on the guest.

 

I read through several threads, but most of them covering slow read and write access. I have a second host with a similar setup with the same controller. Any advice how to speed up writes is highly appreciated. Thank you very much!

[Alan Lantz]

My guess is reads are cached giving you a false sense of true read speed and writes are more of an actual

measure of your true speed. If it makes you feel any better I did this on my test system and I got 

133MB/sec and 8 MB/sec ! I was a little surprised.

 

--Alan--

 

[Tobias Kreidl]

You should run something more realistic like iometer or bonnie++ to take out the cache component. The dd utility is IMO often very misleading in what it reports because of its simplistic nature.

 

-=Tobias

[Marvin Müller]

Thanks very much for the input. I did further runs with bonnie++ with similar results. It seems that the root cause was the array not yet being fully in sync. After now being in good health, performance is much better.

 

Marvin

[Tobias Kreidl]

Ah, good to hear! The utility iostat can be very helpful for identifying things like latency and I/O saturation.

 

-=Tobias