0

I've just built 2 Atom based servers for home purpose (home backup of my websites and for entertainment)

CPUs are Dual core Atom @1.80 GHz
MainBoards are Gigabyte GA-D525TUD

First server: 2*2TB HD in raid 1 software:

[root@host674 ~]# hdparm -tT /dev/md0
/dev/md0:
Timing cached reads: 2432 MB in 2.00 seconds = 1215.88 MB/sec
Timing buffered disk reads: 10 MB in 3.19 seconds = 3.14 MB/sec

Uptime (not during hdparm):

top - 19:42:55 up 57 min, 1 user, load average: 3.28, 3.22, 3.10
Tasks: 109 total, 4 running, 105 sleeping, 0 stopped, 0 zombie
Cpu(s): 0.1%us, 2.0%sy, 0.0%ni, 90.1%id, 0.0%wa, 0.1%hi, 7.7%si, 0.0%st
Mem: 2049144k total, 238116k used, 1811028k free, 1788k buffers
Swap: 4072456k total, 0k used, 4072456k free, 129428k cached

Second server: 4*2TB HD in raid 6 software:

[root@host583 ~]# hdparm -tT /dev/md0
/dev/md0:
Timing cached reads: 2432 MB in 2.00 seconds = 1216.48 MB/sec
Timing buffered disk reads: 12 MB in 3.13 seconds = 3.83 MB/sec

Uptime (not during hdparm):

top - 15:27:13 up 58 min, 1 user, load average: 2.21, 2.29, 2.31
Tasks: 109 total, 2 running, 107 sleeping, 0 stopped, 0 zombie
Cpu(s): 0.1%us, 7.2%sy, 0.0%ni, 85.5%id, 0.0%wa, 0.2%hi, 7.1%si, 0.0%st
Mem: 2049144k total, 242588k used, 1806556k free, 2564k buffers
Swap: 3976044k total, 0k used, 3976044k free, 131912k cached

On both servers, hard disk access is something like this:

hdparm -t /dev/hda
/dev/hda:
Timing buffered disk reads: 2 MB in 3.96 seconds = 517.08 kB/sec

--

Why is load average greater than 3 if cpus are 90% idle? It's fault of the hard disk, right? But why are hard disk such slow if the should be supposing to be doing anything?
Is the atom system the bottleneck?
Is the bottleneck something else?
Should i buy a PCI raid hardware card? (i don't have PCI-X, only PCI on the gigabyte motherboard)
If so, any suggestion?

My goal would be to reach 100MB/s on read.

Thank you a lot!

Improve this question
4
  • This question is better suited to server fault
    – Robb
    Dec 6, 2010 at 13:37
  • Can someone move it than? Thank you!
    – cedivad
    Dec 6, 2010 at 13:45
  • cat /proc/mdstat, please....
    – mattdm
    Dec 6, 2010 at 19:06
  • i realized a few minuts before your answer that the raid array was building. I felt really stupid.
    – cedivad
    Dec 11, 2010 at 18:28

3 Answers 3

Reset to default
0

Since you are using software RAID, you have to take into account the fact that your mainboard's i/o may not be as powerful as a hardware RAID device. First off, if they are SATA, make sure the drives are set for the fastest mode possible; also, setting the drives to AHCI may provide a speed boost. I'd go through your bios settings and look them up to see if they're optimal for data throughput.

One note on load average; many people look at it and consider it analogous with cpu usage. This isn't the case. Load average measures the average number of processes in the run state over three different time periods. This number can be absurdly high on some systems with little impact to performance.

Improve this answer
0

It isn't the software RAID: the stats for hda are horrible too.

The load average is the number of processes in the run queue waiting for CPU, disk, or network resources. In your case it most likely the disk.

The fact that the device is hda and not sda may mean that you have you BIOS configured to emulate PATA IDE devices which may cause a lack of performance. You should change that to be SATA devices. The BIOS may call it "ACHI".

Note, however, you may not be able to use the Linux raid driver with more than 2 drives: http://www.zotacusa.com/forum/topic/2682-nm10-dtx-sata-raid-configuration-issues/

Improve this answer
0

How long ago did you build these? Are you sure the devices are finished building?

mdadm --detail /dev/md0

Maybe it's resyncing in the background?

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .