openMosix and FireWire
<<< Previous		Next >>>

The tests with Firewire

Portofolio

With the firewire link up and running I re-ran the tests as described before but now over eth1394 rather than eth0; the results are below.

Below are the results with 15 processes

[root@dhcp71 portfolio]# time ./portofolio.pl
starting 15 processes
started, pid: 1627
started, pid: 1628
started, pid: 1629
started, pid: 1630
started, pid: 1631
started, pid: 1632
started, pid: 1633
started, pid: 1634
started, pid: 1635
started, pid: 1636
started, pid: 1637
started, pid: 1638
started, pid: 1639
started, pid: 1640
started, pid: 1641
Simulation finished

real    8m39.132s
user    16m1.970s
sys     0m10.900s
[root@dhcp71 portfolio]#

Given the fact that portofolio isn't really an IO intensive program, I wasn't really suprised to see just a mere 9 seconds improvement.

oMFS over IEEE 1394

As oMFS is a more IO intensive task, I'm really hoping to see some important improvement here.

[root@dhcp71 mfstest]# time ./mfscopy.sh
10000+0 records in
10000+0 records out
cp /tmp/testfile.zero from here to node 1
cp /tmp/testfile.zero from mfs-node 1 to local
cp: skipping file `/mfs/1/tmp/testfile.zero', as it was replaced while being copied
nodes equal! do not cp
nodes equal! do not cp
cp /tmp/testfile.zero from here to node 2
cp /tmp/testfile.zero from mfs-node 2 to local
cp: skipping file `/mfs/2/tmp/testfile.zero', as it was replaced while being copied
cp /tmp/testfile.zero from mfs-node 2 to mfs-node 1
cp: skipping file `/mfs/2/tmp/testfile.zero', as it was replaced while being copied
cp /tmp/testfile.zero from mfs-node 1 to mfs-node 2
cp: skipping file `/mfs/1/tmp/testfile.zero', as it was replaced while being copied
cp /tmp/testfile.zero from here to node 1
cp /tmp/testfile.zero from mfs-node 1 to local
cp: skipping file `/mfs/1/tmp/testfile.zero', as it was replaced while being copied
cp /tmp/testfile.zero from mfs-node 1 to mfs-node 2
cp: skipping file `/mfs/1/tmp/testfile.zero', as it was replaced while being copied
cp /tmp/testfile.zero from mfs-node 2 to mfs-node 1
cp: skipping file `/mfs/2/tmp/testfile.zero', as it was replaced while being copied
cp /tmp/testfile.zero from here to node 2
cp /tmp/testfile.zero from mfs-node 2 to local
cp: skipping file `/mfs/2/tmp/testfile.zero', as it was replaced while being copied
nodes equal! do not cp
nodes equal! do not cp
delete /mfs/1/tmp/testfile.zero
delete /mfs/2/tmp/testfile.zero
finished ok

real    0m3.039s
user    0m0.110s
sys     0m1.490s

This is obviously faster: getting from 0m21.945 seconds to 0m3.039 seconds is a good thing. However, I fear that I should look for a test that runs longer than 20 seconds in order to see the real speed improvement. The actual running and forking of a process needs time, and 3 seconds might be the minimum time required to actually do this. We only had a x7 improvement, while one could have expected a x40 improvement.

However that would have reduced the time the application should run to less than 1 second, whis is almost impossible given the overhead of starting and stopping a process.

<<< Previous	Home	Next >>>
Installing FireWire Networking		Conclusion