[uml-user] memcpy performance on UML

[firnnauriel <rin...@ya...>]

Greetings to all. I would like to ask for your advice
regarding this question:
"In general, is the memcpy's performance on UML (User
Mode Linux) environment better than the host OS? If
that is the case, can you give the reason behind it
(ex. UML's architecture)?"

Here's the evironment:
HOST
OS                : Redhat 9 (kernel 2.4.20-8)
Brand/Model:	: Dell PowerEdge 2600
CPU			: Intel Xeon 2.4 GHz
Memory		: 2 GB
Disk Space		: 73 GB

UML
Kernel version    : 2.40.20-lsm1-8um 
Allocated Mem     : 256MB

I used lmbench-2.0.4 for the bandwith and latency
testing. here's the result:

host
memcpy     : 597.1  MB/s (higher is better)
page fault : 2.00000     (smaller is better)

uml
memcpy     : 668.4
page fault : 48.0

The result in page fault might be because of low
memory allocated in UML. But the question is, why is
the memcpy on UML (668.4 MB/s) performs faster that
the host OS?

I already asked the developer of lmbench, and he
adviced me to ask this question to the linux kernel
mailing list.

I would really appreciate any inputs and advice. Thank
you!


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com