3

I have a HP DL320e g8 v2 server running CentOS 7.1 (kernel 3.10.0-229.4.2.el7.x86_64). This server comes with a B120i onboard SmartArray controller, and it was configured with two identical 3TB Seagate disks in RAID-0. I've installed CentOS 7.1 minimal system, with XFS and bs=4096.

After system´s installation, I got theses rates to read and write:

dd if=/dev/zero of=test.img bs=4k count=2560000
2560000+0 records in
2560000+0 records out
10485760000 bytes (10 GB) copied, 25.4219 s, 412 MB/s

dd if=test.img of=test2.img bs=4k
2560000+0 records in
2560000+0 records out
10485760000 bytes (10 GB) copied, 413.832 s, 25.3 MB/s

dd if=test.img of=/dev/null bs=4k
2560000+0 records in
2560000+0 records out
10485760000 bytes (10 GB) copied, 400.053 s, 26.2 MB/s

As you can see, the WRITE rate is fast (about 400 MB/s), but the READ rate is very slow (about 25 MB/s). I've done two tests, first writing to another file and after writing to /dev/null... and in both cases the read rate is slow.

I get the same 25 MB/s read rate when coping a file from this server to another via NFS.

Somebody has any ideia about what could be happening?

I've tried using CentOS 6.6 and got the same results. This server will be a backup (bacula) server, and the disk read rate impacts the bacula performance...

Tks

ewwhite
  • 197,159
  • 92
  • 443
  • 809
Elieser Topassi
  • 103
  • 1
  • 1
  • 10
  • 1
    RAID 0 for a backup server? – ewwhite Jun 03 '15 at 13:59
  • 1
    Do you have any cache on this RAID controller? – ewwhite Jun 03 '15 at 14:16
  • RAID0 because it's only for bacula spool... it's just to obtain 6TB. Not for permanent storage. – Elieser Topassi Jun 03 '15 at 15:02
  • Read cache is enabled, but write cache is disabled. I know bacula recomends use of write cache even the batery is not present, but I dont enabled it. – Elieser Topassi Jun 03 '15 at 15:04
  • How much RAM do you have there? Linux can buffer your write memory pages. Did you align partitions correctly? Do you have SATA ports on mainboard? Write speed is too fast for two disks. Something wrong here. Have you tried choose bigger block size when coping data using dd for example `1MiB`? – Michal Sokolowski Jun 03 '15 at 20:00
  • I have 4 GB of RAM. The server has three partitions, and I don't know about alignment (I will google about it now). The motherboard hasn't SATA ports, only the B120i SAS port (which support sata disks - that is my case). I've tried with 256K block size, and I got similar results. – Elieser Topassi Jun 03 '15 at 20:14
  • Okay, how about read test directly from block device? Do the test: `dd if=/dev/sdX of=/dev/null bs=1Mib count=10000`, please replace **X** with correct disk letter. I'd say that I have this controller somewhere in HP gen 8 server and it preforms well. What about the raid firmware, is it updated? – Michal Sokolowski Jun 04 '15 at 13:56
  • Driver is updated, last version from HP site. I'll run te test next sunday... – Elieser Topassi Jun 05 '15 at 17:44
  • dd if=/dev/sdb of=/dev/null bs=1Mib count=10000 10000+0 records in 10000+0 records out 10485760000 bytes (10 GB) copied, 360.212 s, 29.1 MB/s – Elieser Topassi Jun 06 '15 at 14:03

2 Answers2

0

Last sunday, I formatted and reinstalled SO using the B120i in AHCI mode. I've used the Linux MD for RAID, and I've got better results.

Now, the read rate is about 300 MB/s, and the copy rate from the disks to a LTO-4 tape is about 140 MB/s. It's copying 1 TB to tape in 2h30.

It's not the best solution (I would like to use the B120 RAID mode), but for now it's ok.

Thank you everybody.

Elieser Topassi
  • 103
  • 1
  • 1
  • 10
0

I had same problems on HP Microserver G8 with same B120i controller. Seems that there are perfomance problems on newer drivers. Last hpvsa driver without these issues is 1.2.8-160, and the newest supported by it is CentOS is 6.5.

Илья Войсковский
  • 1