Messages in this thread | | | Date | Tue, 23 Nov 2004 22:31:43 +0000 | From | "J.A. Magallon" <> | Subject | Re: Network slowdown from 2.6.7 to 2.6.9 |
| |
On 2004.11.23, Stephen Hemminger wrote: > On Mon, 22 Nov 2004 15:38:16 -0800 > Harry Edmon <harry@atmos.washington.edu> wrote: > > > Tried them all - none of them helped. Use "ntop" I can see that my > > throughput on the Intel gigabit ethernet interface on the system maxes > > out at 15.2 Mbps with 2.6.9. With 2.6.7 it made it to 35 Mbps. > > > > Does anyone have any other suggestions as to what to look for to > > diagnose this problem? > > Well, before the TSO changes, if TSO was enabled then TCP would not obey slow > start or do congestion control properly. Did you increase the TCP send/receive > buffers (sysctl's net.ipv4.tcp_rmem and net.ipv4.tcp_wmem)? You may just > be window limited. Also, 2.6.9 has TCP bugs with TSO that can cause panic's. > These have been fixed in 2.6.10-rc2. >
With two 2.6.9 boxes (really, 2.6.9-mm1 and 2.6.9-bproc), I get this:
nada:~> iperf -s ------------------------------------------------------------ Server listening on TCP port 5001 TCP window size: 85.3 KByte (default) ------------------------------------------------------------ [ 4] local 155.210.155.215 port 5001 connected with 155.210.155.212 port 41695 [ ID] Interval Transfer Bandwidth [ 4] 0.0-10.0 sec 112 MBytes 94.0 Mbits/sec annwn:/proc/sys/net> iperf -c nada ------------------------------------------------------------ Client connecting to nada, TCP port 5001 TCP window size: 16.0 KByte (default) ------------------------------------------------------------ [ 3] local 155.210.155.212 port 41695 connected with 155.210.155.215 port 5001 [ ID] Interval Transfer Bandwidth [ 3] 0.0-10.0 sec 112 MBytes 94.0 Mbits/sec
So 94 Mbits/sec, through 100Mb ether. So parhaps the netsork layer is not the culprit, just the driver.
BTW:
I'm getting an oops with 2.6.9 that ends on a call to skb_clone, using e1000. Is this what you refer to ?
If the answer is yes, do you have any pointer to a patch to fix just this ? I would prefer an individual patch instead of full -bk, because I use 2.6.9+bproc, and probably bproc wont patch against -bk.
TIA.
-- J.A. Magallon <jamagallon()able!es> \ Software is like sex: werewolf!able!es \ It's better when it's free Mandrakelinux release 10.2 (Cooker) for i586 Linux 2.6.10-rc2-jam3 (gcc 3.4.1 (Mandrakelinux 10.1 3.4.1-4mdk)) #1
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |