* [Routing] Performance Comparison between 2.6 and 2.4 kernel
@ 2004-09-11 0:52 Song Wang
2004-09-11 2:11 ` jamal
0 siblings, 1 reply; 7+ messages in thread
From: Song Wang @ 2004-09-11 0:52 UTC (permalink / raw)
To: netdev
Hi,
I'm looking for the network packet routing
performance comparison between the latest 2.4
and 2.6 kernel.
I did some initial testing on my mips-based router.
2.6 kernel (2.6.8.1) performed way below 2.4 kernel
(2.4.17). It sucks. Turning on preemption on 2.6
kernel even makes routing performance worse.
Anybody did a comprehensive study on the routing
performance for 2.6 and 2.4 kernel?
Thanks.
__________________________________
Do you Yahoo!?
Y! Messenger - Communicate in real time. Download now.
http://messenger.yahoo.com
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [Routing] Performance Comparison between 2.6 and 2.4 kernel
2004-09-11 0:52 [Routing] Performance Comparison between 2.6 and 2.4 kernel Song Wang
@ 2004-09-11 2:11 ` jamal
2004-09-15 18:06 ` Song Wang
0 siblings, 1 reply; 7+ messages in thread
From: jamal @ 2004-09-11 2:11 UTC (permalink / raw)
To: Song Wang; +Cc: netdev
What mips board is this? NAPI based?
I would say that you should get higher numbers with 2.6.x.
Are you running any sort of apps that need gettimeofday?
I have certainly seen higher numbers on 2.6 vs 2.4 starting at some
point where Andis gettimeofday patch went in - but this was on x86.
cheers,
jamal
On Fri, 2004-09-10 at 20:52, Song Wang wrote:
> Hi,
>
> I'm looking for the network packet routing
> performance comparison between the latest 2.4
> and 2.6 kernel.
>
> I did some initial testing on my mips-based router.
> 2.6 kernel (2.6.8.1) performed way below 2.4 kernel
> (2.4.17). It sucks. Turning on preemption on 2.6
> kernel even makes routing performance worse.
>
> Anybody did a comprehensive study on the routing
> performance for 2.6 and 2.4 kernel?
>
> Thanks.
>
>
>
> __________________________________
> Do you Yahoo!?
> Y! Messenger - Communicate in real time. Download now.
> http://messenger.yahoo.com
>
>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [Routing] Performance Comparison between 2.6 and 2.4 kernel
2004-09-11 2:11 ` jamal
@ 2004-09-15 18:06 ` Song Wang
2004-09-15 18:08 ` David S. Miller
0 siblings, 1 reply; 7+ messages in thread
From: Song Wang @ 2004-09-15 18:06 UTC (permalink / raw)
To: hadi; +Cc: netdev
Hi,
The mips CPU on the board is MIPS32 compatible.
I don't think it's NAPI based.
The gettimeofday patch you mentioned came into
the 2.6 kernel at what version?
Do you think it's related to HZ change from 100
to 1000? Actually I did change HZ back to 100,
but it didn't help.
-Song
--- jamal <hadi@cyberus.ca> wrote:
>
> What mips board is this? NAPI based?
> I would say that you should get higher numbers with
> 2.6.x.
> Are you running any sort of apps that need
> gettimeofday?
> I have certainly seen higher numbers on 2.6 vs 2.4
> starting at some
> point where Andis gettimeofday patch went in - but
> this was on x86.
>
> cheers,
> jamal
>
> On Fri, 2004-09-10 at 20:52, Song Wang wrote:
> > Hi,
> >
> > I'm looking for the network packet routing
> > performance comparison between the latest 2.4
> > and 2.6 kernel.
> >
> > I did some initial testing on my mips-based
> router.
> > 2.6 kernel (2.6.8.1) performed way below 2.4
> kernel
> > (2.4.17). It sucks. Turning on preemption on 2.6
> > kernel even makes routing performance worse.
> >
> > Anybody did a comprehensive study on the routing
> > performance for 2.6 and 2.4 kernel?
> >
> > Thanks.
> >
> >
> >
> > __________________________________
> > Do you Yahoo!?
> > Y! Messenger - Communicate in real time. Download
> now.
> > http://messenger.yahoo.com
> >
> >
>
>
_______________________________
Do you Yahoo!?
Declare Yourself - Register online to vote today!
http://vote.yahoo.com
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [Routing] Performance Comparison between 2.6 and 2.4 kernel
2004-09-15 18:06 ` Song Wang
@ 2004-09-15 18:08 ` David S. Miller
2004-09-16 11:30 ` jamal
0 siblings, 1 reply; 7+ messages in thread
From: David S. Miller @ 2004-09-15 18:08 UTC (permalink / raw)
To: Song Wang; +Cc: hadi, netdev
On Wed, 15 Sep 2004 11:06:03 -0700 (PDT)
Song Wang <wsonguci@yahoo.com> wrote:
> The mips CPU on the board is MIPS32 compatible.
> I don't think it's NAPI based.
NAPI is an attribute of the network adapter, not
the cpu.
Although I would very much anticipate this being
a MIPS or network adapter specific problem since
on x86 routing is pretty fast in 2.6.x
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [Routing] Performance Comparison between 2.6 and 2.4 kernel
2004-09-15 18:08 ` David S. Miller
@ 2004-09-16 11:30 ` jamal
2004-09-23 0:29 ` Song Wang
2004-09-29 19:37 ` Song Wang
0 siblings, 2 replies; 7+ messages in thread
From: jamal @ 2004-09-16 11:30 UTC (permalink / raw)
To: David S. Miller; +Cc: Song Wang, netdev
On Wed, 2004-09-15 at 14:08, David S. Miller wrote:
> On Wed, 15 Sep 2004 11:06:03 -0700 (PDT)
> Song Wang <wsonguci@yahoo.com> wrote:
>
> > The mips CPU on the board is MIPS32 compatible.
> > I don't think it's NAPI based.
>
> NAPI is an attribute of the network adapter, not
> the cpu.
>
> Although I would very much anticipate this being
> a MIPS or network adapter specific problem since
> on x86 routing is pretty fast in 2.6.x
Note, I could hit 4-500Kpps on a bcom sb1250 800Mhz; trying to simulate
Andis gettimeofday changes (which are in 2.6 since i havent seen a 2.6
port for that board ) gives another 100Kpps. Theres still room for
improvement just havent had time.
You are right though it could be a 2.6.x arch specific issue. Bacchus?
cheers,
jamal
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [Routing] Performance Comparison between 2.6 and 2.4 kernel
2004-09-16 11:30 ` jamal
@ 2004-09-23 0:29 ` Song Wang
2004-09-29 19:37 ` Song Wang
1 sibling, 0 replies; 7+ messages in thread
From: Song Wang @ 2004-09-23 0:29 UTC (permalink / raw)
To: hadi, David S. Miller; +Cc: netdev
> >
> > Although I would very much anticipate this being
> > a MIPS or network adapter specific problem since
> > on x86 routing is pretty fast in 2.6.x
>
> Note, I could hit 4-500Kpps on a bcom sb1250 800Mhz;
> trying to simulate
> Andis gettimeofday changes (which are in 2.6 since i
> havent seen a 2.6
> port for that board ) gives another 100Kpps. Theres
> still room for
> improvement just havent had time.
> You are right though it could be a 2.6.x arch
> specific issue. Bacchus?
>
> cheers,
> jamal
>
>
Hi Jamal and Dave,
I continued to do some testing in a minimal
environment, basically the routing between the
two ethernet ports on the mips board. Only
ethernet driver is loaded. The user-space
only has busybox.
The difference between 2.6.8.1 and 2.4.17 kernel
is 3000pps less for routing. However, the bridging
performance is the same between two kernels.
Jamal, where is the Andis gettimeofday changes
you mentioned?
Thanks.
-Song
__________________________________
Do you Yahoo!?
Yahoo! Mail is new and improved - Check it out!
http://promotions.yahoo.com/new_mail
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [Routing] Performance Comparison between 2.6 and 2.4 kernel
2004-09-16 11:30 ` jamal
2004-09-23 0:29 ` Song Wang
@ 2004-09-29 19:37 ` Song Wang
1 sibling, 0 replies; 7+ messages in thread
From: Song Wang @ 2004-09-29 19:37 UTC (permalink / raw)
To: hadi, David S. Miller; +Cc: Song Wang, netdev
Hi, jamal,
Could you please kindly point where the gettimeofday
patch is? Is it in 2.6.8.1 kernel?
BTW, what is the packet size for you performance
number, 64 bytes?
I appreciate it.
-Song
--- jamal <hadi@cyberus.ca> wrote:
> On Wed, 2004-09-15 at 14:08, David S. Miller wrote:
> > On Wed, 15 Sep 2004 11:06:03 -0700 (PDT)
> > Song Wang <wsonguci@yahoo.com> wrote:
> >
> > > The mips CPU on the board is MIPS32 compatible.
> > > I don't think it's NAPI based.
> >
> > NAPI is an attribute of the network adapter, not
> > the cpu.
> >
> > Although I would very much anticipate this being
> > a MIPS or network adapter specific problem since
> > on x86 routing is pretty fast in 2.6.x
>
> Note, I could hit 4-500Kpps on a bcom sb1250 800Mhz;
> trying to simulate
> Andis gettimeofday changes (which are in 2.6 since i
> havent seen a 2.6
> port for that board ) gives another 100Kpps. Theres
> still room for
> improvement just havent had time.
> You are right though it could be a 2.6.x arch
> specific issue. Bacchus?
>
> cheers,
> jamal
>
>
_______________________________
Do you Yahoo!?
Declare Yourself - Register online to vote today!
http://vote.yahoo.com
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2004-09-29 19:37 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2004-09-11 0:52 [Routing] Performance Comparison between 2.6 and 2.4 kernel Song Wang
2004-09-11 2:11 ` jamal
2004-09-15 18:06 ` Song Wang
2004-09-15 18:08 ` David S. Miller
2004-09-16 11:30 ` jamal
2004-09-23 0:29 ` Song Wang
2004-09-29 19:37 ` Song Wang
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).