From: Shirley Ma on
On Tue, 2010-08-10 at 18:43 -0700, Shirley Ma wrote:
> > Also I found some vhost performance regression on the new
> > kernel with tuning. I used to get 9.4Gb/s, now I couldn't get it.
>
> I forgot to mention the kernel I used 2.6.36 one. And I found the
> native
> host BW is limited to 8.0Gb/s, so the regression might come from the
> device driver not vhost.

Something is very interesting, when binding ixgbe interrupts to cpu1,
and running netperf/netserver on cpu0, the native host to host
performance is still around 8.0Gb/s, however, the macvtap zero copy
result is 9.0Gb/s.

root(a)localhost ~]# netperf -H 192.168.10.74 -c -C -l60 -T0,0 -- -m 64K
TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.10.74 (192.168..
10.74) port 0 AF_INET : cpu bind
Recv Send Send Utilization Service Demand
Socket Socket Message Elapsed Send Recv Send Recv
Size Size Size Time Throughput local remote local remote
bytes bytes bytes secs. 10^6bits/s % S % S us/KB us/KB

87380 16384 65536 60.00 9013.59 53.01 8.21 0.963 0.597

Below is perf top output:

578.00 6.5% copy_user_generic_string
381.00 4.3% vmx_vcpu_run
250.00 2.8% schedule
207.00 2.3% vhost_get_vq_desc
204.00 2.3% _raw_spin_lock_irqsave
197.00 2.2% translate_desc
193.00 2.2% memcpy_fromiovec
162.00 1.8% gup_pte_range

We can compare your results with mine to see any difference.

Thanks
Shirley

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
From: Shirley Ma on
On Tue, 2010-08-10 at 23:01 -0700, Shirley Ma wrote:
> On Tue, 2010-08-10 at 18:43 -0700, Shirley Ma wrote:
> > > Also I found some vhost performance regression on the new
> > > kernel with tuning. I used to get 9.4Gb/s, now I couldn't get it.
> >
> > I forgot to mention the kernel I used 2.6.36 one. And I found the
> > native
> > host BW is limited to 8.0Gb/s, so the regression might come from the
> > device driver not vhost.
>
> Something is very interesting, when binding ixgbe interrupts to cpu1,
> and running netperf/netserver on cpu0, the native host to host
> performance is still around 8.0Gb/s, however, the macvtap zero copy
> result is 9.0Gb/s.
>
> root(a)localhost ~]# netperf -H 192.168.10.74 -c -C -l60 -T0,0 -- -m 64K
> TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.10.74 (192.168..
> 10.74) port 0 AF_INET : cpu bind
> Recv Send Send Utilization Service Demand
> Socket Socket Message Elapsed Send Recv Send Recv
> Size Size Size Time Throughput local remote local remote
> bytes bytes bytes secs. 10^6bits/s % S % S us/KB us/KB
>
> 87380 16384 65536 60.00 9013.59 53.01 8.21 0.963 0.597
>
> Below is perf top output:
>
> 578.00 6.5% copy_user_generic_string
> 381.00 4.3% vmx_vcpu_run
> 250.00 2.8% schedule
> 207.00 2.3% vhost_get_vq_desc
> 204.00 2.3% _raw_spin_lock_irqsave
> 197.00 2.2% translate_desc
> 193.00 2.2% memcpy_fromiovec
> 162.00 1.8% gup_pte_range
>
> We can compare your results with mine to see any difference.

When binding vhost thread to cpu3, qemu I/O thread to cpu2, macvtap zero
copy patch can get 9.4Gb/s.

TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.10.74 (192.168.10.74) port 0 AF_INET : cpu bind
Recv Send Send Utilization Service Demand
Socket Socket Message Elapsed Send Recv Send Recv
Size Size Size Time Throughput local remote local remote
bytes bytes bytes secs. 10^6bits/s % S % S us/KB us/KB

87380 16384 65536 60.00 9408.19 55.69 8.45 0.970 0.589

Shirley

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/