[Xen-devel] problem with LSI-megaraid ?

[Peter_bier <peter_bier@xxxxxx>]

I have been running both xen-86_64 and xen-86 on a Fujitsu-Siemens RX300
I tested the write-performance of a linux guest. The machine is configured
with an LSI Megaraid Controller, Model LD 1 RAID0 139G, Rev 516D. 

Then Version of xen is 3.0.3

When doing "dd if=/dev/zero of=/dev/sdb1 bs=64k count=10000", I got a through-
put of about 3.6 Megabytes per second. Device "sdb1" on the client was 
connected to sdb12 on the backend which was a slice on the disk controlled
by the megaraid device. When doing the corresponding "dd" command on the 
privileged domain I got a througbut of about 25 Megabytes per second. 

I installed Xen on another machine only capable of running 32-bit code. I did
similar tests and got ( partially due to a faster disk ) throughputs of 
35Megabytes per second from the client and about 40 MB/s from the priviledged
domain in respective tests. 

I suspected that the problem was probably related to the 64-bit Version but 
after installing it on the Rx300 I got exactly the same results. In a final
test I connecte a USB-disk to the RX300 and got a throughput of 25 MB/sec onto
that disk. I suspect that Xen is cooperating quite inefficiently with the 
controller. I found out that Writing speed was much faster when using a 
file-based VBD, even when the Systems Memory was significantly smaller than
the amount of transfered data ( I limited domain0's Memory to 384 MB and wrote
3 GB ). But this is not a solution because on the other System the process
of writing permanently activated the oom-killer in domain0 and crashed the
entire system within a few minutes. Preventing it from killing xen-related
processes kept the System up, but it did no longer respond to external requests
of the net. 

Reading throughput on the client domain was about the same as in domain 0




_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel