Performance testing new build with 4.5b3

[MartinQ]

I've been out of the forums for a while but now I'm back and have committed to a Pro license. Here are some numbers I'm getting with this rig. Comments? Suggestions?

 

 

Copy from XP to UnRAID over 1GbT network:

===============================

~6%  utilization (Windows Task Manager, Networking)

5 - 8  Load Average (writes to Disk1 + Parity)

7.41, 6.28, 4.23  Load Averages after a 4.3GB transfer

 

 

Copy from UnRAID to XP:

===================

~15%  utilization (Windows Task Manager, Networking)

3+    Load Average

3.37, 2.46, 1.74  Load Averages after a 4.3GB transfer

 

 

Move files within a share on UnRAID using Midnight Commander / PuTTY:

=============================================

6-7 MB/s

5+ (up to 7)  Load Average

 

 

Parity Check:

========

~28,000 KB/sec

1.2 - 1.5  Load Average

 

 

System Specs:

=========

Asus A7N8X-E

  AMD AthlonXP 1900+ (1470MHz, FS Bus 266MHz, PCI Bus 66MHz)

  512MB RAM

 

Ethernet (onboard)

  Marvell 88E801

  Speed: 1000Mb/s

  Duplex: Full

 

Drives

  4 x 500GB SATA (WD5000AAKS)

  Temperatures 34 - 40C

 

Controllers

  Sil3112 (onboard) - Parity

  Sil3114 (PCI1) - Disk1

  Sil3114 (PCI5) - Disk2, Disk3

 

 

 

[RobJ]

There *may* be a problem with a drive, but we need to see your syslog, see my sig for the Troubleshooting link.  What version of unRAID?

 

Would you mind defining 'Load Average', and how you derived those numbers?  I'm not sure what they mean.

 

A good page that may be helpful:  Improving unRAID Performance

[MartinQ]

Hi Rob.

 

The version I'm using is in the thread title.

 

The load averages were obtained by using the 'uptime' command via telnet. I would update them several times while doing a transfer to get an idea of what was going on. You can also monitor them using the 'top' command and they're also accessible via UnMenu > System Info.

 

I have read over the 'Improving unRAID Performance' page a few times and have done everything there.

 

[RobJ]

Thanks, I feel dumb!  Next time I'll do a little homework first.

[MartinQ]

Some more observations using the 'top' command:

 

These are the top processes when accessing a Disk and the approx CPU usage:

Reading:

- smbd 8 - 12%

- unraidd 0.7 - 2%

 

Writing:

- smbd 12 - 40%

- unraidd 5 - 10%

 

 

... but when accessing a Share:

Reading:

- shfs 30 - 53%

- smbd 7 - 14%

- unraidd 1%

 

Writing:

- shfs 30 - 70%

- smbd 13 - 30%

- unraidd 4 - 9%

 

From this is looks like I am maxing out the CPU while doing writes to a share. I am assuming that these numbers are normal. Any perspective on the high CPU usage of shfs?

 

Using 'bwm-ng' to monitor network traffic I see about 10MB/s while writing to a share. I also see frequent dips or momentary stops in traffic.

 

[MartinQ]

Tuesday Update Part II

 

I changed the FSB from 133 (266MHz) to 166 (333MHz effective) so I am no longer underclocking the CPU. The AthlonXP 1900 is now a 2500 and here is what I've noticed.

 

- Parity check has bumped up a MB or two. It now hovers around 30 - 31MB/s.

 

- Writing data to a Share has gone up maybe 2MB/s. It seems to be around 11 - 14MB/s.

 

I also did some PCI hopping of the second 3114 controller and found that using PCI slots 4 & 5 showed significant drops in READ throughput across the network. I also noticed a slight improvement setting the disk read-ahead cache to 2048 but it was very slight.