Hi, I would like to hear your comments to these findings. It seems that in bsd, and other operating sustems, you can change certain values to improve the performace of the ethernet cards when used in a heavy network situation. For example I found this site http://www.psc.edu/networking/perf_tune.html that discusses in detail how to do that. My question is this. Does changing the size of the buffer in /proc/sys/net/core/rmem_default - default receive window /proc/sys/net/core/rmem_max - maximum receive window /proc/sys/net/core/wmem_default - default send window /proc/sys/net/core/wmem_max - maximum send window improve the performance in Linux? have anyone done this on their servers? The document also describes an algorithm that autonegotiaites the best size for transfering data, and Linux does not seem to have that support built in. In your opinion is it worth investigating this and tweaking the linux servers?? Now in terms of my previous post I have the following comment: The current network infrastructure that we have are using has not changed in a some time, however our traffic has. The host in question has a load factor of less than 1, and it is producing overruns like crazy. I can not attribute this problem to slugish machine, because of the low cpu load. It was suggested that the machine may have an older version of ifconfig. So looking at the /proc I am inclined to believe that he is very right. After editing the output of "cat /proc/net/dev" to fit the screen we get Inter-| Receive face |bytes packets errs drop fifo frame compressed multicast lo: 23416 262 0 0 0 0 0 0 eth0:1068796065 9367666 0 0 0 0 0 0 eth1: 268139 4196 0 0 1 0 0 0 Transmit bytes packets errs drop fifo colls carrier compressed lo: 23416 262 0 0 0 0 0 0 eth0: 3766429230 13754583 2 0 0 0 2 0 eth1: 1008 24 0 0 0 0 0 0 while ifconfig says this. lo Link encap:Local Loopback inet addr:127.0.0.1 Bcast:127.255.255.255 Mask:255.0.0.0 UP LOOPBACK RUNNING MTU:3924 Metric:1 RX packets:23416 errors:262 dropped:0 overruns:0 TX packets:0 errors:0 dropped:0 overruns:23416 eth0 Link encap:10Mbps Ethernet HWaddr 00:20:78:E0:A2:52 inet addr:222.222.8.118 Bcast:222.222.8.127 Mask:255.255.255.240 UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:1074032897 errors:9413029 dropped:0 overruns:0 TX packets:0 errors:0 dropped:0 overruns:2147483647 Interrupt:10 Base address:0xe000 eth1 Link encap:10Mbps Ethernet HWaddr 00:01:02:46:08:B5 inet addr:222.222.8.125 Bcast:222.222.8.127 Mask:255.255.255.240 UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:268319 errors:4196 dropped:0 overruns:0 TX packets:0 errors:0 dropped:0 overruns:1008 Interrupt:12 Base address:0xdc00 And lastly. For kernerls 2.0.36 and 2.14+ what is the most stable driver/ethernet vendor that you recommend from your expereience. Our servers have 3com 3c59x, realtek, smc, etc.. Some work great and other shut down under heavy load. I would love to stick to one vendor with a mature driver. A lot of DoS attacks are targetting this flaw, and end up bringing the interface to its knees. Thanks in advance for any feedback, and if I am on the wrong email list, please let me know, which mailing list would be the best to figure out what is going on. best regards, Adonis On Mon, 11 Dec 2000, Dennis wrote: > At 05:56 PM 12/10/2000, you wrote: > >Hi, > > > >We have a highly (network) loaded server, running redhat 6.2. The card is > >using tulip driver. > > > >This is the output of ifconfig (note:IPs in this post are not real numbers) > >eth0 Link encap:10Mbps Ethernet HWaddr 00:20:78:E0:A2:52 > > inet addr:222.222.8.118 Bcast:222.222.8.127 Mask:255.255.255.240 > > UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 > > RX packets:23883054 errors:181152 dropped:0 overruns:0 > > TX packets:0 errors:0 dropped:0 overruns:95989676 > > Interrupt:10 Base address:0xe000 > > > >We are getting tons of overruns, and I do not know how to fix this > >I tried setting the card via the /etc/conf.module, but nothing changed > > Its LINUX that cant handle the load, the card is just a card. You can A) > use a faster machine b) reduce other tasks running on the system..for > example webservers, PERL scripts anything with substantial disk activity > that may be taking CPU cycles away from processing network traffic, or c) > try a different card. > > Dennis > - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org
