headers
Mark Smith | 29 Jan 2011 12:12
Picon

Intel Pro/1000 in a Qube2 observations

Hi there,

Since things appear to be working with this board I've done a few rough tests with some interesting results
(only 100BaseT results at the moment as thats all my current router/hub supports)

1)  By default the hardware checksums are not enabled ... which seems kind of odd you can see the before and
after here :

	Before:
	qube2# ifconfig -v wm0
		wm0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
        	capabilities=2bf80<TSO4,IP4CSUM_Rx,IP4CSUM_Tx,TCP4CSUM_Rx,TCP4CSUM_Tx,UDP4CSUM_Rx,UDP4CSUM_Tx,TCP6CSUM_Tx,UDP6CSUM_Tx>
        	enabled=0
        	address: 00:15:17:0e:b8:e2
        	media: Ethernet autoselect (100baseTX full-duplex,flowcontrol,rxpause,txpause)
        	status: active
      	  	input: 556569 packets, 42363982 bytes, 410 multicasts
        	output: 922440 packets, 1358364969 bytes, 7 multicasts
        	inet 192.168.0.7 netmask 0xffffff00 broadcast 192.168.0.255
       	 	inet6 fe80::215:17ff:fe0e:b8e2%wm0 prefixlen 64 scopeid 0x2

	then run :  ifconfig wm0 ip4csum tcp4csum udp4csum

	After:
	qube2# ifconfig -v wm0
		wm0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
        	capabilities=2bf80<TSO4,IP4CSUM_Rx,IP4CSUM_Tx,TCP4CSUM_Rx,TCP4CSUM_Tx,UDP4CSUM_Rx,UDP4CSUM_Tx,TCP6CSUM_Tx,UDP6CSUM_Tx>
        	enabled=3f00<IP4CSUM_Rx,IP4CSUM_Tx,TCP4CSUM_Rx,TCP4CSUM_Tx,UDP4CSUM_Rx,UDP4CSUM_Tx>
        	address: 00:15:17:0e:b8:e2
        	media: Ethernet autoselect (100baseTX full-duplex,flowcontrol,rxpause,txpause)
(Continue reading)

Permalink | Reply |
headers
Izumi Tsutsui | 29 Jan 2011 13:40
Picon
Gravatar

Re: Intel Pro/1000 in a Qube2 observations

> 2)  If I read a large file from the Qube2 with the checksums disabled
> I can get 4.5Mb/s sustained.  If I enable the checksums I get 5.9Mb/s
> sustained (so worth enabling)

Yes.

> 3)  If I copy a large file with the checksums disabled the NFS
> connection WILL drop off, disconnect and basically break!
> If I enable the checksums it seems to stay up. (so REALLY worth enabling)
 :
> 		resvport locallocks rsize=32768 wsize=32768 intr noatime

Probably 32KB UDP is too large for cobalt to serve?
UDP NFS uses large and fragmented packets, and hwcksum won't help
sums of fragmented packets. TCP 8KB might be better.

> 4)  CLI responsiveness is far more sluggish when the hardware
> checksums are enabled than when they are disabled ...
> which doesn't really make sense either ?!

Probably some UDP errors on heavy traffic?
You can see network statistics by "netstat -s" command.

---
Izumi Tsutsui
Permalink | Reply |

Gmane