Hi! I am doing som lab/testing.

I'm running an OES2 Linux server on as a XEN domU (paravirtualized) on
top of SLES10SP1. The virtual/domU server is configured to be an
eDirectory server with DHCP, DNS and NSS.

Everything works fine until I try to transfer some amounts of data
to/from the NSS volume (RAID5 created from iManager, using physical
partitions mapped through XEN). I had no evident I/O performance
issues during installation or during daily use.

But when I try to copy some large files, I/O is horrible. Like now I
tested transferring a 580 MB iso file from XP to my NSS share, using
Novell Client 4.91.4, and it took 20 minutes. I seem to get a
throughput of less than 5-6 Mbps, and the network between my XEN box
and the XP box is pure 1000Base-TX.

This test/lab server, is not a new machine, though. It has a Athlon
XP2000+ CPU and two PATA HDD's and two SATA drives (using a Promise
PCI SATA 150/300 controller) and with 2 GB RAM.

I know that particular SATA controller on PCI is not optimal for
performance, but it doesn't explain why the I/O over the net is _so_
low, because local I/O during installation and such wasn't that bad.

I am wondering about the CPU, though. The CPU load in the virtual
machine is really high, like 95 % and such. Example:

top - 16:36:05 up 1:45, 2 users, load average: 6.61, 7.18, 5.76
Tasks: 196 total, 2 running, 194 sleeping, 0 stopped, 0 zombie
Cpu(s): 95.7%us, 1.0%sy, 0.0%ni, 0.0%id, 0.0%wa, 0.0%hi, 0.0%si, 3.3%st
Mem: 899264k total, 887432k used, 11832k free, 6180k buffers
Swap: 787144k total, 52k used, 787092k free, 487252k cached

2978 novlxsrv 16 0 343m 3360 2448 S 41.2 0.4 6:32.85 novell-xsrvd
2975 novlxsrv 16 0 343m 3360 2448 S 27.6 0.4 3:00.35 novell-xsrvd
3005 novlxsrv 16 0 343m 3360 2448 S 27.2 0.4 9:56.15 novell-xsrvd

But the strain on CPU seems to be triggered by accessing the NSS share
or of course if I access iManager, iMonitor and stuff like that, who
require tomcat and stuff ... Five minutes after the NSS transfer ended,
the OES2 machine is back to 0.4%-26.0% CPU load ...

If I recall correctly, my CPU should meet the requirements from
Novell, but maybe I really need more power? (I am planning on running
another eDirectory instance virtualized as well, but maybe not on this
CPU ...)

Or maybe there's a XEN or NSS bug?