RE: [Bonding-devel][PATCH]Zero Copy Transmit Support (Update)

[Godse, Radheka]

> -
> + 
> +	/* We let the bond device publish all hardware
> +	 * acceleration features possible. This is OK,
> +	 * since if an skb is passed from the bond to
> +	 * a slave that doesn't support one of those
> +	 * features, everything is fixed in the
> +	 * dev_queue_xmit() function (e.g. calculate
> +	 * check sum, linearize the skb, etc.).
> +	 */

This is very inefficient if the bond slaves don't
support these features.

I believe you when you say you saw improvement in the
case where the slaves do support TSO, but if you test
a non-TSO slave case I bet you'll see a marked
decrease in system utilization at least.
...

Note: 2.6.2 patch enables Zero Copy Transmit support by default while
registering the bond device... but, I continue to see better CPU
utilization (~50% improvement) when this patch was tested in various
scenarios(see data below). Test results are from a dual Itanium 2 1.4GHz
system but I saw a similar trend on a PIII Xeon 550MHz x 8 system.
 
bond-patch-2.6.1 and 82546EB Copper NICs with hardware acceleration
support, 
Test results with TSO on and off
ETH1  ETH2  Throughput/Performance  System Utilization
on    off  1870                    35 %
on   on   1870                    35 %
off   off   1871                    34 %

bond-patch-2.6.2 and 82546EB Copper NICs with hardware acceleration
support, 
Test results with TSO on and off
ETH1  ETH2  Throughput/Performance  System Utilization
on    off  1879                    14 %
on   on   1880                    14.08 %
off   off   1879                    13 %
------------------------------------------------------------------

bond-patch-2.6.1 and 82543GC NICs without hardware support for TSO
ETH1  ETH2  Throughput/Performance  System Utilization
off   off   1815                    36.85 %
 
bond-patch-2.6.2 and 82543GC NICs without hardware support for TSO
ETH1  ETH2  Throughput/Performance  System Utilization
off   off   1848                    11.85 %
------------------------------------------------------------------

bond-patch-2.6.1, and one 82546EB NIC with & one 82543GC NIC without
hardware support for TSO 
ETH1  ETH2  Throughput/Performance  System Utilization

on    off   1817                    30.69 %

bond-patch-2.6.2, and one 82546EB NIC with & one 82543GC NIC without
hardware support for TSO 
ETH1  ETH2  Throughput/Performance  System Utilization
on    off   1836                    24.54 %
------------------------------------------------------------------

bond-patch-2.6.1 and 82543GC NICs
Test results with all HW acceleration bits on for 1 NIC and turned off
for the other NIC
ETH1  ETH2  Throughput/Performance  System Utilization
on    off  1851                    35 %

bond-patch-2.6.2 and 82543GC NICs
Test results with all HW acceleration bits on for 1 NIC and turned off
for the other NIC
ETH1  ETH2  Throughput/Performance  System Utilization
on    off  1822                    17 %
------------------------------------------------------------------

bond-patch-2.6.1 and 82543GC NICs
Test results with all HW acceleration bits turned off for both NICs
ETH1  ETH2  Throughput/Performance  System Utilization
off    off  1816                    35 %
 
bond-patch-2.6.2 and 82543GC NICs
Test results with all HW acceleration bits turned off for both NICs
ETH1  ETH2  Throughput/Performance  System Utilization
off    off  1604                    16.23 %
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/