Re: How to improve small packet performance

[minnie wu <minniewkitty@xxxxxxxxxxxx>]

   Here is the experimental setup I used:
   A traffic generator[Smartbits] is sending from one of its ports to the PC on eth0. Linux routes the packet out eth1 back to the Smartbits. /proc/sys/net/ipv4/ip_forward set to 1.
  The PC is a PIII 863MHz with 256M of RAM. And the NIC cards do not share the same irq.
  The unidirectional test:
   e100-v3.0.0dev12-napi    47k packets/sec
   eepro100-napi-020619     89k packets/sec
  
  minnie
jchapman@xxxxxxxxxxx wrote:

I downloaded the eepro100 NAPI patch and tried it in my test setup (a
1GHz P-III configured as a 2-port bridge, 2.4.20-8 RH9 kernel). I see
e100-napi perform much better than eepro100, which is what I was
expecting

e100-v3.0.0dev12-napi 93k packets/sec
eepro100-napi-020619 49k packets/sec

When doing a bidirectional test, the eepro100 driver locks up
completely, although it seems to survive with 128 byte packets (where
the event rate is less).

What's your test setup? How are you generating your test traffic? Do
your NIC cards share the same irq?

-jc

---

Do You Yahoo!?
暑期大片齐聚雅虎通 网络摄像头+雅虎通调频收音机等你来拿