thr3ads.net - Xen users - [Xen-users] Network and processing performances [Jun 2008]

If this information is useful, please help other people find it:
Share via:

Andreï V. Fomitchev

2008-Jun-03 09:27 UTC

[Xen-users] Network and processing performances

Hi the XEN-Users list.

I have to evaluate the XEN solution for an industrial project and I am unable to
answer/understand some questions about network and processing performances.

NB: the tested hardware platform is not optimized for XEN. The processor is an
Intel P4 3.2 GHz HT without VMX support. The Dom0 kernel is&nbsp;
2.6.18-6-xen-vserver-686 #1 SMP (from debian etch) and runs on all available
VCPUS; DomX (debian etch 2.6.18-4-xen-vserver-686 #1 SMP) uses 1 VCPU.


1) Network performances:
I tried the bridge and route modes but I did not note real difference between
these modes (the firewall is "down" in Dom0). So, the performance
tests (ssh transfer of 128 MBytes file) have been ran in default bridge mode.
The performances from DomU to Dom0 or to external server and from external
server to Dom0 or to DomU are acceptable (ssh transfers at 10 MBytes/s-16
MBytes/s).
The transfer from Dom0 to DomU (5 MBytes/s) and between DomU (1.5 MBytes/s) are
slow. In the list archive, I found two possible improvements:
&nbsp;- increase the sizes of TCP window and of TXQueue
&nbsp;- disable the CRC checking of TX in DomX ($ ethtool -K
&lt;device&gt; tx off)
The first solution does not impact the results however the second increased the
transfer between DomX to 12 MBytes/s.
Does it exist another improvement?
What are the QoS impacts when the CRC is not checked? I think nothing because
the IP packets do not leave a sure interface and they can not be corrupted.
Please, someone can confirm my hypothesis?

2) global DomX performances
Using lmbench and phoronix benchmark suites, I noted that sometimes the DomU
performances are better that Dom0. For example,
Lame compression: 146s for DomU and 151s for Dom0
GZip compression: 134.50s for DomU and 350.29s for Dom0(!!!), i.e. DomU is 160%
faster that Dom0
LMBench pipe bandwidth communication: 1186 MBytes/s for DomU and 415 MBytes/s
for Dom0.
I does not understand the observed behaviours. Please, someone can explain me
the differences?

And my last question: the CPU is HT (whose two logical processors) but
''xm vcpu-list'' command exhibit that the CPU#1 is not really
used by Dom0
$xm vcpu-list
Name&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
ID VCPUs&nbsp;&nbsp; CPU State&nbsp;&nbsp; Time(s) CPU Affinity
Domain-0&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
0&nbsp;&nbsp;&nbsp;&nbsp;
0&nbsp;&nbsp;&nbsp;&nbsp; 0&nbsp;&nbsp;
r--&nbsp;&nbsp;&nbsp;&nbsp; 241.9 any cpu
Domain-0&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
0&nbsp;&nbsp;&nbsp;&nbsp;
1&nbsp;&nbsp;&nbsp;&nbsp; 1&nbsp;&nbsp;
-b-&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 6.3 any cpu
xen-etch&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
1&nbsp;&nbsp;&nbsp;&nbsp;
0&nbsp;&nbsp;&nbsp;&nbsp; 1&nbsp;&nbsp;
-b-&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 53.4 any cpu
xen-etch1&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
2&nbsp;&nbsp;&nbsp;&nbsp;
0&nbsp;&nbsp;&nbsp;&nbsp; 1&nbsp;&nbsp;
-b-&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 41.2 any cpu
Is it XEN+HT-CPU limitation? How works "xm vcpu-pin"? The
documentation is not explicit.

Thank you for yours responses.

Best regards,

-- 
Andreï V. FOMITCHEV
"Ada is PL/I trying to be SmallTalk" -- Codoso diBlini

&nbsp;

__________________________________________________
Do You Yahoo!?
En finir avec le spam? Yahoo! Mail vous offre la meilleure protection possible
contre les messages non sollicités
http://mail.yahoo.fr Yahoo! Mail 

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

Andreï V. Fomitchev

2008-Jun-03 12:36 UTC

head link

[Xen-users] Network and processing performances

Hi the XEN-Users list.

I have to evaluate the XEN solution for
an industrial project and I am unable to answer/understand some
observed questions about network and processing performances. 

NB: the
tested hardware platform is not optimized for XEN. The processor is an
Intel P4 3.2 GHz HT without VMX support. The Dom0 kernel is&nbsp;
2.6.18-6-xen-vserver-686 #1 SMP (from debian etch) and runs on all
available VCPUS; DomX (debian etch 2.6.18-4-xen-vserver-686 #1 SMP)
uses 1 VCPU.


1) Network performances:
I tried the bridge
and route modes but I did not note real difference between these modes
(the firewall is "down" in Dom0). So, the performance tests (ssh
transfer of 128 MBytes file) have been ran in default bridge mode.
The performances from DomU to Dom0 or to external server and
from external server to Dom0 or to DomU are acceptable (ssh transfers
at 10 MBytes/s-16 MBytes/s).
The transfer from Dom0 to DomU (5
MBytes/s) and between DomU (1.5 MBytes/s) are slow. In the list
archive, I found two possible improvements: 
&nbsp;- increase the sizes of TCP window and of TXQueue
&nbsp;- disable the CRC checking of TX in DomX ($ ethtool -K
&lt;device&gt; tx off)
The first solution does not impact the results however the second increased the
transfer between DomX to 12 MBytes/s.
Does it exist another improvement?
What
are the QoS impacts when the CRC is not checked? I think nothing
because the IP packets do not leave a sure interface and they can not
be corrupted. Please, someone can confirm my hypothesis?

2) global DomX performances
Using lmbench and phoronix benchmark suites, I noted that sometimes the DomU
performances are better that
 Dom0. For example, 
Lame compression: 146s for DomU and 151s for Dom0
GZip compression: 134.50s for DomU and 350.29s for Dom0(!!!), i.e. DomU is 160%
faster that Dom0
LMBench pipe bandwidth communication: 1186 MBytes/s for DomU and 415 MBytes/s
for Dom0.
I does not understand the observed behaviours. Please, someone can explain me
the differences?

And
my last question: the CPU is HT (whose two logical processors) but ''xm
vcpu-list'' command exhibit that the CPU#1 is not really used by Dom0
$xm vcpu-list
Name&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
ID VCPUs&nbsp;&nbsp; CPU State&nbsp;&nbsp; Time(s) CPU Affinity
Domain-0&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
 0&nbsp;&nbsp;&nbsp;&nbsp;
0&nbsp;&nbsp;&nbsp;&nbsp; 0&nbsp;&nbsp;
r--&nbsp;&nbsp;&nbsp;&nbsp; 241.9 any cpu
Domain-0&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
0&nbsp;&nbsp;&nbsp;&nbsp;
1&nbsp;&nbsp;&nbsp;&nbsp; 1&nbsp;&nbsp;
-b-&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 6.3 any cpu
xen-etch&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
1&nbsp;&nbsp;&nbsp;&nbsp;
0&nbsp;&nbsp;&nbsp;&nbsp; 1&nbsp;&nbsp;
-b-&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 53.4 any cpu
xen-etch1&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
2&nbsp;&nbsp;&nbsp;&nbsp;
0&nbsp;&nbsp;&nbsp;&nbsp; 1&nbsp;&nbsp;
 -b-&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 41.2 any cpu
Is it XEN+HT-CPU limitation? How works "xm vcpu-pin"? The
documentation is not explicit.

Thank you for yours responses.

Best regards,

-- 
Andreï V. FOMITCHEV
"Ada is PL/I trying to be SmallTalk" -- Codoso diBlini

__________________________________________________
Do You Yahoo!?
En finir avec le spam? Yahoo! Mail vous offre la meilleure protection possible
contre les messages non sollicités
http://mail.yahoo.fr Yahoo! Mail 

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

Andreï V. Fomitchev

2008-Jun-04 07:13 UTC

head link

Re: [Xen-users] Network and processing performances

Hi,

Thank you Jim for your response. Yes, I use the VBD.

Best regards,

--- En date de : Mer 4.6.08, jim burns &lt;jim_burn@bellsouth.net&gt; a
écrit :
De: jim burns &lt;jim_burn@bellsouth.net&gt;
Objet: Re: [Xen-users] Network and processing performances
À: fomitchev@yahoo.fr
Date: Mercredi 4 Juin 2008, 4h33

On Tuesday June 03 2008 08:36:28 am Andreï V. Fomitchev wrote:
&gt; 2) global DomX performances
&gt; Using lmbench and phoronix benchmark suites, I noted that sometimes the
&gt; DomU performances are better that Dom0. For example, Lame compression:
146s
&gt; for DomU and 151s for Dom0
&gt; GZip compression: 134.50s for DomU and 350.29s for Dom0(!!!), i.e. DomU
is
&gt; 160% faster that Dom0 LMBench pipe bandwidth communication: 1186
MBytes/s
&gt; for DomU and 415 MBytes/s for Dom0. I does not understand the observed
&gt; behaviours. Please, someone can explain me the differences?

If you are using file backed vbds (using file: in your config''s
diskline),
you may be seeing effects of dom0 caching your image.

__________________________________________________
Do You Yahoo!?
En finir avec le spam? Yahoo! Mail vous offre la meilleure protection possible
contre les messages non sollicités
http://mail.yahoo.fr Yahoo! Mail 

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

Xen users - Jun 2008 - Network and processing performances

[Xen-users] Network and processing performances

[Xen-users] Network and processing performances

Re: [Xen-users] Network and processing performances