Xen devel - Dec 2006 - Puzzling ttcp behaviour on Xen

We are running 2 ettcp (test TCP) tests on xen as follows:
Dom0 (source) to DomU (sink)
And
DomU (source) to Dom0 (sink)

Dom0 has 2 vcpus and 256MB of memory. 
DomU has 1 vcpu and 256MB of memory.

Results (in seconds) show that communications from DomU to Dom0 are
consistently
faster (we show first 3 tests):

(We are passing a 4KB block each time)
Source->Sink	# Transactions	tcp/ip I	tcp/ip II	tcp/ip
III		
Dom0 -> DomU	100000	      4.77		4.76		4.79
dom0: -t
			1000000		47.86		47.65
47.43		domU: -r
			10000000		481.23	468.51	469.31

DomU -> Dom0	100000		1.81		1.89		1.79
domU: -t
			1000000		17.19		17.23
16.98		dom0: -r
			10000000		171.83	171.95	171.71




We observe the same behaviour when blocks are 8KB (default) or 640B
as well as when we use our inter domain communication channel 
(based on Xen rings, shared memory etc.)

Can anybody explain why?


Thanks so much,
:)
Micha.

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

Moffie, Micha <micha.moffie@hp.com> wrote:
> We are running 2 ettcp (test TCP) tests on xen as follows:
> Dom0 (source) to DomU (sink)
> And
> DomU (source) to Dom0 (sink)
> 
> Dom0 has 2 vcpus and 256MB of memory. 
> DomU has 1 vcpu and 256MB of memory.
> 
> Results (in seconds) show that communications from DomU to Dom0 are
> consistently
> faster (we show first 3 tests):
This is understandable as domU => dom0 uses direct memory mapping
while dom0 => domU involves (at least) one copy.

Cheers,
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

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