Re: Abysmal RECV network performance
Hi, I'm guessing that the tulip driver is not setting the chip up correctly. I've seen this happen with other tulip variants (21143) when tries to autonegotiate. if you do an ifconfig eth1 you will see numerous carrier and crc errors. Set the tulip_debug flag to 2 or 3 in /etc/modules.conf and see what gets said. A newer version of the driver may help you. You might try the one on sourceforge. Also, I've only ever seen full 100BaseT speeds with decent adapters, like 21143 based tulips, Intel eepros, and vortex/boomerang 3com cards. A lot of the cheaper controllers just won't get there. skd On Mon, May 28, 2001 at 03:47:22AM +, John William wrote: > Can someone please help me troubleshoot this problem - I am getting abysmal > (see numbers below) network performance on my system, but the poor > performance seems limited to receiving data. Transmission is OK. > > The computer in question is a dual Pentium 90 machine. The machine has > RedHat 7.0 (kernel 2.2.16-22 from RedHat). I have compiled 2.2.19 (stock) > and 2.4.3 (stock) for the machine and used those for testing. I had a > NetGear FA310TX card that I used with the "tulip" driver and a 3Com 3CSOHO > card (Hurricane chipset) that I used with the "3c59x" driver. I used the > netperf package to test performance (latest version, but I don't have the > version number off-hand). The numbers netperf is giving me seem to correlate > well to FTP statistics I see to the box. > > I have a second machine (P2-350) with a NetGear FA311 (running 2.4.3 and the > "natsemi" driver) that I used to talk with the Pentium 90 machine. The two > machines are connected through a NetGear FS105 10/100 switch. I also tried > using a 10BT hub (see below). > > When connected, the switch indicated 100 Mbps, full duplex connections to > both cards. This matches the speed indicator lights on both cards. I have > run the miidiag program in the past to verify that the cards are actually > set to full duplex, but I didn't run it again this time (this isn't the > first time I have tried to chase this problem down). > > For the purposes of this message, call the P2-350 machine "A" and the dual > P-90 machine "B". I ran the following tests: > > Machine "A" to localhost 754.74 Mbps > > Kernel 2.2.19SMP > Machine "B" to localhost 80.63 Mbps > Machine "B" to "A" (tulip)55.38 Mbps > Machine "A" to "B" (tulip)10.60 Mbps > Machine "A" to "B" (3c95x)12.10 Mbps > > Kernel 2.4.3 SMP > Machine "B" to localhost 83.87 Mbps > Machine "B" to "A" (tulip)68.07 Mbps > Machine "A" to "B" (tulip)1.62Mbps > Machine "A" to "B" (3c95x)2.37Mbps > > Kernel 2.2.16-22 (RedHat kernel) > Machine "B" to localhost 92.29 Mbps > Machine "B" to "A" (tulip)57.34 Mbps > Machine "A" to "B" (tulip)9.98Mbps > Machine "A" to "B" (3c95x)9.05Mbps > > Now, with both "A" and "B" plugged into a 10BT hub: > > Kernel 2.2.19SMP > Machine "B" to "A" (tulip)6.96Mbps > Machine "A" to "B" (tulip)6.89Mbps > > At the end of the runs, I do not see any messages in syslog that would > indicate a problem. Using the switch, there were no collisions but looking > at /sbin/ifconfig there were a lot of "Frame:" errors on receive. "A lot" > means ~30% of the total packets received. This happened with both cards and > all kernels. > > The conclusions I draw from this data are: > > 1) Both machines connecting to localhost (data not going out over the wire) > give reasonable numbers and are considerably above what I actually see going > over the network (as would be expected). > 2) The P-90 machine seems to have good transmit speed over both cards and > all kernels. Transmit performance is close to the localhost numbers, so I > can believe them. In the past, I have compared the performance of the FA310 > to the 3ComSOHO card and there did not seem to be any measurable performance > difference between the two. > 3) Both the FA310 and the 3ComSOHO card have similar receive speeds, leading > me to believe that the problem lies with either the machine or the kernel > and not the individual cards or drivers. > 4) Booting the machine as a uni-processor machine (with a non-SMP 2.2.16 > kernel) did not change anything, so it does not appear to be a problem with > SMP. > 5) Kernel 2.4.3 receive performance is significantly lower than either 2.2.x > kernel, so that tends to point to some fundamental problem in the kernel. > 6) As I understand it, the 3Com card has some hardware acceleration for > checksumming, and this is a slow machine, so why is the performance almost > identical to the FA310? > > So, my questions are: > > What kind of performance should I be seeing with a P-90 on a 100Mbps > connection? I was expecting something in the range of 40-70 Mbps - certainly > not 1-2 Mbps. > > What can I do to track this problem down? Has anyone else had problems like > this? > > Thanks in advance
Re: Abysmal RECV network performance
John William wrote: > > >Depends on what is driving it... An application I built can only push > >about > >80 Mbps bi-directional on PII 550Mhz machines. It is not the most > >efficient program in > >the world, but it isn't too bad either... > > > >I missed the rest of this thread, so maybe you already mentioned it, but > >what is the bottleneck? Is your CPU running at 100%? > > > >Greatly increasing the buffers both in the drivers and in the sockets > >does wonders for higher-speed connections, btw. > > > >Ben > > I don't know what the bottleneck is. What I'm seeing is ~60Mbps transmit > speed and anywhere from 1 to 12Mpbs receive speed on a couple 10/100 cards > using the 2.2.16, 2.2.19 and 2.4.3 kernels. > > I have tried increasing the size of the RX ring buffer and it did not seem > to make any difference. It appears that there is some sort of overrun or > other problem. There is a significant slowdown between the 2.2.x and 2.4.x > kernels. > > However, just tonight, while really hammering on the system, I started to > get some messages like "eth1: Oversized Ethernet frame spanned multiple > buffers, status 7fff8301!". Any ideas what could be causing that? Nope, I'd take it up with the driver developers. For what it's worth, the Intel Ether-Express Pro cards are the only ones I've found yet that really work right at high speeds. Intel's e100 driver seems to work really well for me, but the eepro driver also works well with most versions of the eepro cards I've used... I have had definate problems with the natsemi (locked up), tulip (won't autonegotiate multi-port cards correctly, or something), rtl8139 (would lock up, haven't tried recent drivers though) I used to assume that Linux had the best/fastest networking support around, but the reality is that I've had a really hard time finding hardware/drivers that works at high speeds (60Mbps+, bi-directional). -- Ben Greear <[EMAIL PROTECTED]> <[EMAIL PROTECTED]> President of Candela Technologies Inc http://www.candelatech.com ScryMUD: http://scry.wanfear.com http://scry.wanfear.com/~greear - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
>Depends on what is driving it... An application I built can only push >about >80 Mbps bi-directional on PII 550Mhz machines. It is not the most >efficient program in >the world, but it isn't too bad either... > >I missed the rest of this thread, so maybe you already mentioned it, but >what is the bottleneck? Is your CPU running at 100%? > >Greatly increasing the buffers both in the drivers and in the sockets >does wonders for higher-speed connections, btw. > >Ben I don't know what the bottleneck is. What I'm seeing is ~60Mbps transmit speed and anywhere from 1 to 12Mpbs receive speed on a couple 10/100 cards using the 2.2.16, 2.2.19 and 2.4.3 kernels. I have tried increasing the size of the RX ring buffer and it did not seem to make any difference. It appears that there is some sort of overrun or other problem. There is a significant slowdown between the 2.2.x and 2.4.x kernels. However, just tonight, while really hammering on the system, I started to get some messages like "eth1: Oversized Ethernet frame spanned multiple buffers, status 7fff8301!". Any ideas what could be causing that? - John _ Get your FREE download of MSN Explorer at http://explorer.msn.com - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
John William wrote: > > >I've seen many reports like this where the NIC is invalidly in > >full-duplex more while the router is in half-duplex mode. > > [root@copper diag]# ./tulip-diag eth1 -m > tulip-diag.c:v2.08 5/15/2001 Donald Becker ([EMAIL PROTECTED]) > http://www.scyld.com/diag/index.html > Index #1: Found a Lite-On 82c168 PNIC adapter at 0xfe00. > Port selection is MII, full-duplex. > Transmit started, Receive started, full-duplex. > The Rx process state is 'Waiting for packets'. > The Tx process state is 'Idle'. > The transmit threshold is 512. > MII PHY found at address 1, status 0x782d. > MII PHY #1 transceiver registers: >1000 782d 7810 01e1 41e1 0001 > > 4000 38c8 0010 0002 >0001 . > [root@copper diag]# ./mii-diag eth1 > Basic registers of MII PHY #1: 1000 782d 7810 01e1 41e1 0001 . > The autonegotiated capability is 01e0. > The autonegotiated media type is 100baseTx-FD. > Basic mode control register 0x1000: Auto-negotiation enabled. > You have link beat, and everything is working OK. > Your link partner advertised 41e1: 100baseTx-FD 100baseTx 10baseT-FD > 10baseT. >End of basic transceiver informaion. > > On the NetGear switch, I have indicator lights for 100baseT-FD on both > connections used for testing. So it appears to me that everything is working > correctly (hardware). > > I keep coming back to a problem with the kernel, or that somehow I have two > cards (FA310 and 3CSOHO) defective in almost exactly the same way, but only > on receive. If it were a hardware problem, why would I only get poor > performance in one direction and not both? > > Does anyone have network performance numbers for a comparable machine (P-90 > class)? I'm thinking I should expect 50-70Mbps on a PCI 10/100 ethernet card > from a P-90 class machine, right? Depends on what is driving it... An application I built can only push about 80 Mbps bi-directional on PII 550Mhz machines. It is not the most efficient program in the world, but it isn't too bad either... I missed the rest of this thread, so maybe you already mentioned it, but what is the bottleneck? Is your CPU running at 100%? Greatly increasing the buffers both in the drivers and in the sockets does wonders for higher-speed connections, btw. Ben -- Ben Greear <[EMAIL PROTECTED]> <[EMAIL PROTECTED]> President of Candela Technologies Inc http://www.candelatech.com ScryMUD: http://scry.wanfear.com http://scry.wanfear.com/~greear - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
>I've seen many reports like this where the NIC is invalidly in >full-duplex more while the router is in half-duplex mode. [root@copper diag]# ./tulip-diag eth1 -m tulip-diag.c:v2.08 5/15/2001 Donald Becker ([EMAIL PROTECTED]) http://www.scyld.com/diag/index.html Index #1: Found a Lite-On 82c168 PNIC adapter at 0xfe00. Port selection is MII, full-duplex. Transmit started, Receive started, full-duplex. The Rx process state is 'Waiting for packets'. The Tx process state is 'Idle'. The transmit threshold is 512. MII PHY found at address 1, status 0x782d. MII PHY #1 transceiver registers: 1000 782d 7810 01e1 41e1 0001 4000 38c8 0010 0002 0001 . [root@copper diag]# ./mii-diag eth1 Basic registers of MII PHY #1: 1000 782d 7810 01e1 41e1 0001 . The autonegotiated capability is 01e0. The autonegotiated media type is 100baseTx-FD. Basic mode control register 0x1000: Auto-negotiation enabled. You have link beat, and everything is working OK. Your link partner advertised 41e1: 100baseTx-FD 100baseTx 10baseT-FD 10baseT. End of basic transceiver informaion. On the NetGear switch, I have indicator lights for 100baseT-FD on both connections used for testing. So it appears to me that everything is working correctly (hardware). I keep coming back to a problem with the kernel, or that somehow I have two cards (FA310 and 3CSOHO) defective in almost exactly the same way, but only on receive. If it were a hardware problem, why would I only get poor performance in one direction and not both? Does anyone have network performance numbers for a comparable machine (P-90 class)? I'm thinking I should expect 50-70Mbps on a PCI 10/100 ethernet card from a P-90 class machine, right? - John _ Get your FREE download of MSN Explorer at http://explorer.msn.com - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
> >the Netgear FA311/2 (tulip). Found that the link lost > >connectivity because of card lockups and transmit timeout > >failures - and some of these were silent. However, I moved > >to the 3C905C (3c59x driver) which behaved like a champ, and > I'm a little confused here - do you mean the FA310TX ("tulip" driver) or the > FA311/2 ("natsemi" driver)? I have not had any connection problems with > either the FA310 or the FA311 cards. I haven't noticed any speed problems > with the FA311 card, but I haven't benchmarked it, either. The FA310 is so > horribly slow, I couldn't help but notice. Unfortunately, the same is true > of the 3cSOHO. Sorry, meant to describe both, (natsemi and tulip, but latter on older DEC chip). > I looked at tcpdump to try and figure it out, and it appeared that the P-90 > was taking a very long time to ACK some packets. I am not a TCP/IP guru by > any stretch, but my guess at the time was that the packets that were taking > forever to get ACK'ed were the ones causing a framing error on the P-90, but > again, I'm not an expert. > The only unusual stat is the framing errors. There are a lot of them under > heavy receive load. The machine will go for weeks without a single framing > error, but if I blast some netperf action at it (or FTP send to it, etc.) > then I get about 1/3 of the incoming packets (to the P-90) with framing > errors. I see no other errors at all except a TX overrun error (maybe 1 in > 10 packets). Tried to reproduce this problem last night on my machines at home (kernel 2.4.4, 500MHz K7/400Mhz K6). Just doing FTP and netperf tests, didnt see any significant variation between rcv and tx sides. Admittedly different machines, and between 3C905C and a FA310TX (tulip). However, if the problem was purely kernel protocol under load, it should have showed. Also, am not seeing significant frame errors - 1 in 10K, definitely not seeing anything remotely like 30%. If 1/3 of your packets are being dropped with frame errs, you'll see lots of retransmissions, horrible performance, no question. But I would expect frame errors to be due to things like the speed not being negotiated correctly(?), or if the board isnt sitting quite right (true - thats the only experience I remember of recv code path being error prone compared to tx), but that should affect all the kernel versions you ran on that host.. Am pretty clueless about media level issues, but it would help to identify whats causing the framing errors. Not much help, I know.. thanks, Nivedita --- Nivedita Singhvi(503) 578-4580 Linux Technology Center [EMAIL PROTECTED] IBM Beaverton, OR [EMAIL PROTECTED] - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
the Netgear FA311/2 (tulip). Found that the link lost connectivity because of card lockups and transmit timeout failures - and some of these were silent. However, I moved to the 3C905C (3c59x driver) which behaved like a champ, and I'm a little confused here - do you mean the FA310TX (tulip driver) or the FA311/2 (natsemi driver)? I have not had any connection problems with either the FA310 or the FA311 cards. I haven't noticed any speed problems with the FA311 card, but I haven't benchmarked it, either. The FA310 is so horribly slow, I couldn't help but notice. Unfortunately, the same is true of the 3cSOHO. Sorry, meant to describe both, (natsemi and tulip, but latter on older DEC chip). I looked at tcpdump to try and figure it out, and it appeared that the P-90 was taking a very long time to ACK some packets. I am not a TCP/IP guru by any stretch, but my guess at the time was that the packets that were taking forever to get ACK'ed were the ones causing a framing error on the P-90, but again, I'm not an expert. The only unusual stat is the framing errors. There are a lot of them under heavy receive load. The machine will go for weeks without a single framing error, but if I blast some netperf action at it (or FTP send to it, etc.) then I get about 1/3 of the incoming packets (to the P-90) with framing errors. I see no other errors at all except a TX overrun error (maybe 1 in 10 packets). Tried to reproduce this problem last night on my machines at home (kernel 2.4.4, 500MHz K7/400Mhz K6). Just doing FTP and netperf tests, didnt see any significant variation between rcv and tx sides. Admittedly different machines, and between 3C905C and a FA310TX (tulip). However, if the problem was purely kernel protocol under load, it should have showed. Also, am not seeing significant frame errors - 1 in 10K, definitely not seeing anything remotely like 30%. If 1/3 of your packets are being dropped with frame errs, you'll see lots of retransmissions, horrible performance, no question. But I would expect frame errors to be due to things like the speed not being negotiated correctly(?), or if the board isnt sitting quite right (true - thats the only experience I remember of recv code path being error prone compared to tx), but that should affect all the kernel versions you ran on that host.. Am pretty clueless about media level issues, but it would help to identify whats causing the framing errors. Not much help, I know.. thanks, Nivedita --- Nivedita Singhvi(503) 578-4580 Linux Technology Center [EMAIL PROTECTED] IBM Beaverton, OR [EMAIL PROTECTED] - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
I've seen many reports like this where the NIC is invalidly in full-duplex more while the router is in half-duplex mode. [root@copper diag]# ./tulip-diag eth1 -m tulip-diag.c:v2.08 5/15/2001 Donald Becker ([EMAIL PROTECTED]) http://www.scyld.com/diag/index.html Index #1: Found a Lite-On 82c168 PNIC adapter at 0xfe00. Port selection is MII, full-duplex. Transmit started, Receive started, full-duplex. The Rx process state is 'Waiting for packets'. The Tx process state is 'Idle'. The transmit threshold is 512. MII PHY found at address 1, status 0x782d. MII PHY #1 transceiver registers: 1000 782d 7810 01e1 41e1 0001 4000 38c8 0010 0002 0001 . [root@copper diag]# ./mii-diag eth1 Basic registers of MII PHY #1: 1000 782d 7810 01e1 41e1 0001 . The autonegotiated capability is 01e0. The autonegotiated media type is 100baseTx-FD. Basic mode control register 0x1000: Auto-negotiation enabled. You have link beat, and everything is working OK. Your link partner advertised 41e1: 100baseTx-FD 100baseTx 10baseT-FD 10baseT. End of basic transceiver informaion. On the NetGear switch, I have indicator lights for 100baseT-FD on both connections used for testing. So it appears to me that everything is working correctly (hardware). I keep coming back to a problem with the kernel, or that somehow I have two cards (FA310 and 3CSOHO) defective in almost exactly the same way, but only on receive. If it were a hardware problem, why would I only get poor performance in one direction and not both? Does anyone have network performance numbers for a comparable machine (P-90 class)? I'm thinking I should expect 50-70Mbps on a PCI 10/100 ethernet card from a P-90 class machine, right? - John _ Get your FREE download of MSN Explorer at http://explorer.msn.com - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
John William wrote: I've seen many reports like this where the NIC is invalidly in full-duplex more while the router is in half-duplex mode. [root@copper diag]# ./tulip-diag eth1 -m tulip-diag.c:v2.08 5/15/2001 Donald Becker ([EMAIL PROTECTED]) http://www.scyld.com/diag/index.html Index #1: Found a Lite-On 82c168 PNIC adapter at 0xfe00. Port selection is MII, full-duplex. Transmit started, Receive started, full-duplex. The Rx process state is 'Waiting for packets'. The Tx process state is 'Idle'. The transmit threshold is 512. MII PHY found at address 1, status 0x782d. MII PHY #1 transceiver registers: 1000 782d 7810 01e1 41e1 0001 4000 38c8 0010 0002 0001 . [root@copper diag]# ./mii-diag eth1 Basic registers of MII PHY #1: 1000 782d 7810 01e1 41e1 0001 . The autonegotiated capability is 01e0. The autonegotiated media type is 100baseTx-FD. Basic mode control register 0x1000: Auto-negotiation enabled. You have link beat, and everything is working OK. Your link partner advertised 41e1: 100baseTx-FD 100baseTx 10baseT-FD 10baseT. End of basic transceiver informaion. On the NetGear switch, I have indicator lights for 100baseT-FD on both connections used for testing. So it appears to me that everything is working correctly (hardware). I keep coming back to a problem with the kernel, or that somehow I have two cards (FA310 and 3CSOHO) defective in almost exactly the same way, but only on receive. If it were a hardware problem, why would I only get poor performance in one direction and not both? Does anyone have network performance numbers for a comparable machine (P-90 class)? I'm thinking I should expect 50-70Mbps on a PCI 10/100 ethernet card from a P-90 class machine, right? Depends on what is driving it... An application I built can only push about 80 Mbps bi-directional on PII 550Mhz machines. It is not the most efficient program in the world, but it isn't too bad either... I missed the rest of this thread, so maybe you already mentioned it, but what is the bottleneck? Is your CPU running at 100%? Greatly increasing the buffers both in the drivers and in the sockets does wonders for higher-speed connections, btw. Ben -- Ben Greear [EMAIL PROTECTED] [EMAIL PROTECTED] President of Candela Technologies Inc http://www.candelatech.com ScryMUD: http://scry.wanfear.com http://scry.wanfear.com/~greear - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
Depends on what is driving it... An application I built can only push about 80 Mbps bi-directional on PII 550Mhz machines. It is not the most efficient program in the world, but it isn't too bad either... I missed the rest of this thread, so maybe you already mentioned it, but what is the bottleneck? Is your CPU running at 100%? Greatly increasing the buffers both in the drivers and in the sockets does wonders for higher-speed connections, btw. Ben I don't know what the bottleneck is. What I'm seeing is ~60Mbps transmit speed and anywhere from 1 to 12Mpbs receive speed on a couple 10/100 cards using the 2.2.16, 2.2.19 and 2.4.3 kernels. I have tried increasing the size of the RX ring buffer and it did not seem to make any difference. It appears that there is some sort of overrun or other problem. There is a significant slowdown between the 2.2.x and 2.4.x kernels. However, just tonight, while really hammering on the system, I started to get some messages like eth1: Oversized Ethernet frame spanned multiple buffers, status 7fff8301!. Any ideas what could be causing that? - John _ Get your FREE download of MSN Explorer at http://explorer.msn.com - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
John William wrote: Depends on what is driving it... An application I built can only push about 80 Mbps bi-directional on PII 550Mhz machines. It is not the most efficient program in the world, but it isn't too bad either... I missed the rest of this thread, so maybe you already mentioned it, but what is the bottleneck? Is your CPU running at 100%? Greatly increasing the buffers both in the drivers and in the sockets does wonders for higher-speed connections, btw. Ben I don't know what the bottleneck is. What I'm seeing is ~60Mbps transmit speed and anywhere from 1 to 12Mpbs receive speed on a couple 10/100 cards using the 2.2.16, 2.2.19 and 2.4.3 kernels. I have tried increasing the size of the RX ring buffer and it did not seem to make any difference. It appears that there is some sort of overrun or other problem. There is a significant slowdown between the 2.2.x and 2.4.x kernels. However, just tonight, while really hammering on the system, I started to get some messages like eth1: Oversized Ethernet frame spanned multiple buffers, status 7fff8301!. Any ideas what could be causing that? Nope, I'd take it up with the driver developers. For what it's worth, the Intel Ether-Express Pro cards are the only ones I've found yet that really work right at high speeds. Intel's e100 driver seems to work really well for me, but the eepro driver also works well with most versions of the eepro cards I've used... I have had definate problems with the natsemi (locked up), tulip (won't autonegotiate multi-port cards correctly, or something), rtl8139 (would lock up, haven't tried recent drivers though) I used to assume that Linux had the best/fastest networking support around, but the reality is that I've had a really hard time finding hardware/drivers that works at high speeds (60Mbps+, bi-directional). -- Ben Greear [EMAIL PROTECTED] [EMAIL PROTECTED] President of Candela Technologies Inc http://www.candelatech.com ScryMUD: http://scry.wanfear.com http://scry.wanfear.com/~greear - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
Hi, I'm guessing that the tulip driver is not setting the chip up correctly. I've seen this happen with other tulip variants (21143) when tries to autonegotiate. if you do an ifconfig eth1 you will see numerous carrier and crc errors. Set the tulip_debug flag to 2 or 3 in /etc/modules.conf and see what gets said. A newer version of the driver may help you. You might try the one on sourceforge. Also, I've only ever seen full 100BaseT speeds with decent adapters, like 21143 based tulips, Intel eepros, and vortex/boomerang 3com cards. A lot of the cheaper controllers just won't get there. skd On Mon, May 28, 2001 at 03:47:22AM +, John William wrote: Can someone please help me troubleshoot this problem - I am getting abysmal (see numbers below) network performance on my system, but the poor performance seems limited to receiving data. Transmission is OK. The computer in question is a dual Pentium 90 machine. The machine has RedHat 7.0 (kernel 2.2.16-22 from RedHat). I have compiled 2.2.19 (stock) and 2.4.3 (stock) for the machine and used those for testing. I had a NetGear FA310TX card that I used with the tulip driver and a 3Com 3CSOHO card (Hurricane chipset) that I used with the 3c59x driver. I used the netperf package to test performance (latest version, but I don't have the version number off-hand). The numbers netperf is giving me seem to correlate well to FTP statistics I see to the box. I have a second machine (P2-350) with a NetGear FA311 (running 2.4.3 and the natsemi driver) that I used to talk with the Pentium 90 machine. The two machines are connected through a NetGear FS105 10/100 switch. I also tried using a 10BT hub (see below). When connected, the switch indicated 100 Mbps, full duplex connections to both cards. This matches the speed indicator lights on both cards. I have run the miidiag program in the past to verify that the cards are actually set to full duplex, but I didn't run it again this time (this isn't the first time I have tried to chase this problem down). For the purposes of this message, call the P2-350 machine A and the dual P-90 machine B. I ran the following tests: Machine A to localhost 754.74 Mbps Kernel 2.2.19SMP Machine B to localhost 80.63 Mbps Machine B to A (tulip)55.38 Mbps Machine A to B (tulip)10.60 Mbps Machine A to B (3c95x)12.10 Mbps Kernel 2.4.3 SMP Machine B to localhost 83.87 Mbps Machine B to A (tulip)68.07 Mbps Machine A to B (tulip)1.62Mbps Machine A to B (3c95x)2.37Mbps Kernel 2.2.16-22 (RedHat kernel) Machine B to localhost 92.29 Mbps Machine B to A (tulip)57.34 Mbps Machine A to B (tulip)9.98Mbps Machine A to B (3c95x)9.05Mbps Now, with both A and B plugged into a 10BT hub: Kernel 2.2.19SMP Machine B to A (tulip)6.96Mbps Machine A to B (tulip)6.89Mbps At the end of the runs, I do not see any messages in syslog that would indicate a problem. Using the switch, there were no collisions but looking at /sbin/ifconfig there were a lot of Frame: errors on receive. A lot means ~30% of the total packets received. This happened with both cards and all kernels. The conclusions I draw from this data are: 1) Both machines connecting to localhost (data not going out over the wire) give reasonable numbers and are considerably above what I actually see going over the network (as would be expected). 2) The P-90 machine seems to have good transmit speed over both cards and all kernels. Transmit performance is close to the localhost numbers, so I can believe them. In the past, I have compared the performance of the FA310 to the 3ComSOHO card and there did not seem to be any measurable performance difference between the two. 3) Both the FA310 and the 3ComSOHO card have similar receive speeds, leading me to believe that the problem lies with either the machine or the kernel and not the individual cards or drivers. 4) Booting the machine as a uni-processor machine (with a non-SMP 2.2.16 kernel) did not change anything, so it does not appear to be a problem with SMP. 5) Kernel 2.4.3 receive performance is significantly lower than either 2.2.x kernel, so that tends to point to some fundamental problem in the kernel. 6) As I understand it, the 3Com card has some hardware acceleration for checksumming, and this is a slow machine, so why is the performance almost identical to the FA310? So, my questions are: What kind of performance should I be seeing with a P-90 on a 100Mbps connection? I was expecting something in the range of 40-70 Mbps - certainly not 1-2 Mbps. What can I do to track this problem down? Has anyone else had problems like this? Thanks in advance for any help you can offer. - John _ Get your FREE download of MSN Explorer at
Re: Abysmal RECV network performance
>From: Nivedita Singhvi <[EMAIL PROTECTED]> >To: [EMAIL PROTECTED] >CC: [EMAIL PROTECTED] >Subject: Re: Abysmal RECV network performance >Date: Mon, 28 May 2001 23:45:28 -0700 (PDT) >While we didnt use 2.2 kernels at all, we did similar tests >on 2.4.0 through 2.4.4 kernels, on UP and SMP. I've used >a similar machine (PII 333MHz) as well as faster (866MHz) >machines, and got pretty nifty (> 90Mbs) throughput on >netperf tests (tcp stream, no disk I/O) over a 100Mb full >duplex link. (not sure if there are any P-90 issues). > >Throughput does drop with small MTU, very small packet sizes, >small socket buffer sizes, but only at extremes, for the most >part throughput was well over 70Mbs. (this is true for single >connections, you dont mention how many connections you were >scaling to, if any). Sorry, yes - I was doing single connection tests with no changes to the default Netperf settings. Each machine was running a copy of the server and then I ran the test with "netperf -H x.x.x.x" to each machine (or just "netperf" for a localhost speed check). >However, we did run into serious performance problems with >the Netgear FA311/2 (tulip). Found that the link lost >connectivity because of card lockups and transmit timeout >failures - and some of these were silent. However, I moved >to the 3C905C (3c59x driver) which behaved like a champ, and >we didnt see the problems any more, so have stuck to that card. >This was back in the 2.4.0 time frame, and there have many >patches since then to various drivers, so not sure if the >problem(s) have been resolved or not (likely to have been, >extensively reported). Both your cards might actually be >underperforming.. I'm a little confused here - do you mean the FA310TX ("tulip" driver) or the FA311/2 ("natsemi" driver)? I have not had any connection problems with either the FA310 or the FA311 cards. I haven't noticed any speed problems with the FA311 card, but I haven't benchmarked it, either. The FA310 is so horribly slow, I couldn't help but notice. Unfortunately, the same is true of the 3cSOHO. While I am willing to accept that both the FA310 and FA311 cards are underperforming, I think it is more than a little strange that the 3cSOHO card would turn in the same performance numbers. Also, keep in mind that I was only seeing horrible receive performance, TX performance seemed to be ok. I didn't post FTP numbers (both machines are running FTP servers). While the FTP performance numbers are probably not as "scientific" as Netperf, they do seem to agree from what I have observered. I.e. retrieving files from the P-90 machine is ok (~3MB/sec) but sending files to it is very slow (~100K/sec). This roughly agrees with the Netperf numbers I saw. FTP transfers to the FA311 machine (P2-350) are OK in both directions. >Are you seeing any errors reported in /var/log/messages? >Are you monitoring your connection via tcpdump, for example? >You might sometimes see long gaps in transmission...Are >there any abnormal numbers in /proc/net/ stats? I dont remember >seeing that high frame errors, although there were a few. No, I don't see anything in /var/log/messages. I looked at tcpdump to try and figure it out, and it appeared that the P-90 was taking a very long time to ACK some packets. I am not a TCP/IP guru by any stretch, but my guess at the time was that the packets that were taking forever to get ACK'ed were the ones causing a framing error on the P-90, but again, I'm not an expert. The only unusual stat is the framing errors. There are a lot of them under heavy receive load. The machine will go for weeks without a single framing error, but if I blast some netperf action at it (or FTP send to it, etc.) then I get about 1/3 of the incoming packets (to the P-90) with framing errors. I see no other errors at all except a TX overrun error (maybe 1 in 10 packets). >HW checksumming for the kind of test you are doing (tcp, mostly >fast path) will not buy you any real performance gain, the >checksum is actually consumed by the user-kernel copy routine. Ok, I'll take your word for it. The P-90 isn't a very fast machine to begin with, so I thought it could use all the HW assistance it could get (that and the 3cSOHO card was really cheap :-). I am very disappointed that TCP/IP performance on this machine is so lousy, but the problem is clearly with the kernel - just look at the performance numbers for 2.4.3 vs 2.2.19 (or 2.2.16). Those numbers aren't exactly great, but they are a lot better than 2.4.3. >You can also run the tests on a profiling kernel and compare >results... > >Nivedita > >--- >Nivedita Singhvi(503) 578-4580 >Linux Technology Center [EMAIL PROTECTED] >IBM Beaverton, OR
Re: Abysmal RECV network performance
> Can someone please help me troubleshoot this problem - > I am getting abysmal (see numbers below) network performance > on my system, but the poor performance seems limited to receiving > data. Transmission is OK. [ snip ] > What kind of performance should I be seeing with a P-90 > on a 100Mbps connection? I was expecting something in the > range of 40-70 Mbps - certainly not 1-2 Mbps. > > What can I do to track this problem down? Has anyone else > had problems like this? While we didnt use 2.2 kernels at all, we did similar tests on 2.4.0 through 2.4.4 kernels, on UP and SMP. I've used a similar machine (PII 333MHz) as well as faster (866MHz) machines, and got pretty nifty (> 90Mbs) throughput on netperf tests (tcp stream, no disk I/O) over a 100Mb full duplex link. (not sure if there are any P-90 issues). Throughput does drop with small MTU, very small packet sizes, small socket buffer sizes, but only at extremes, for the most part throughput was well over 70Mbs. (this is true for single connections, you dont mention how many connections you were scaling to, if any). However, we did run into serious performance problems with the Netgear FA311/2 (tulip). Found that the link lost connectivity because of card lockups and transmit timeout failures - and some of these were silent. However, I moved to the 3C905C (3c59x driver) which behaved like a champ, and we didnt see the problems any more, so have stuck to that card. This was back in the 2.4.0 time frame, and there have many patches since then to various drivers, so not sure if the problem(s) have been resolved or not (likely to have been, extensively reported). Both your cards might actually be underperforming.. Are you seeing any errors reported in /var/log/messages? Are you monitoring your connection via tcpdump, for example? You might sometimes see long gaps in transmission...Are there any abnormal numbers in /proc/net/ stats? I dont remember seeing that high frame errors, although there were a few. HW checksumming for the kind of test you are doing (tcp, mostly fast path) will not buy you any real performance gain, the checksum is actually consumed by the user-kernel copy routine. You can also run the tests on a profiling kernel and compare results... Nivedita --- Nivedita Singhvi(503) 578-4580 Linux Technology Center [EMAIL PROTECTED] IBM Beaverton, OR [EMAIL PROTECTED] - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
Can someone please help me troubleshoot this problem - I am getting abysmal (see numbers below) network performance on my system, but the poor performance seems limited to receiving data. Transmission is OK. [ snip ] What kind of performance should I be seeing with a P-90 on a 100Mbps connection? I was expecting something in the range of 40-70 Mbps - certainly not 1-2 Mbps. What can I do to track this problem down? Has anyone else had problems like this? While we didnt use 2.2 kernels at all, we did similar tests on 2.4.0 through 2.4.4 kernels, on UP and SMP. I've used a similar machine (PII 333MHz) as well as faster (866MHz) machines, and got pretty nifty ( 90Mbs) throughput on netperf tests (tcp stream, no disk I/O) over a 100Mb full duplex link. (not sure if there are any P-90 issues). Throughput does drop with small MTU, very small packet sizes, small socket buffer sizes, but only at extremes, for the most part throughput was well over 70Mbs. (this is true for single connections, you dont mention how many connections you were scaling to, if any). However, we did run into serious performance problems with the Netgear FA311/2 (tulip). Found that the link lost connectivity because of card lockups and transmit timeout failures - and some of these were silent. However, I moved to the 3C905C (3c59x driver) which behaved like a champ, and we didnt see the problems any more, so have stuck to that card. This was back in the 2.4.0 time frame, and there have many patches since then to various drivers, so not sure if the problem(s) have been resolved or not (likely to have been, extensively reported). Both your cards might actually be underperforming.. Are you seeing any errors reported in /var/log/messages? Are you monitoring your connection via tcpdump, for example? You might sometimes see long gaps in transmission...Are there any abnormal numbers in /proc/net/ stats? I dont remember seeing that high frame errors, although there were a few. HW checksumming for the kind of test you are doing (tcp, mostly fast path) will not buy you any real performance gain, the checksum is actually consumed by the user-kernel copy routine. You can also run the tests on a profiling kernel and compare results... Nivedita --- Nivedita Singhvi(503) 578-4580 Linux Technology Center [EMAIL PROTECTED] IBM Beaverton, OR [EMAIL PROTECTED] - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Abysmal RECV network performance
From: Nivedita Singhvi [EMAIL PROTECTED] To: [EMAIL PROTECTED] CC: [EMAIL PROTECTED] Subject: Re: Abysmal RECV network performance Date: Mon, 28 May 2001 23:45:28 -0700 (PDT) snip While we didnt use 2.2 kernels at all, we did similar tests on 2.4.0 through 2.4.4 kernels, on UP and SMP. I've used a similar machine (PII 333MHz) as well as faster (866MHz) machines, and got pretty nifty ( 90Mbs) throughput on netperf tests (tcp stream, no disk I/O) over a 100Mb full duplex link. (not sure if there are any P-90 issues). Throughput does drop with small MTU, very small packet sizes, small socket buffer sizes, but only at extremes, for the most part throughput was well over 70Mbs. (this is true for single connections, you dont mention how many connections you were scaling to, if any). Sorry, yes - I was doing single connection tests with no changes to the default Netperf settings. Each machine was running a copy of the server and then I ran the test with netperf -H x.x.x.x to each machine (or just netperf for a localhost speed check). However, we did run into serious performance problems with the Netgear FA311/2 (tulip). Found that the link lost connectivity because of card lockups and transmit timeout failures - and some of these were silent. However, I moved to the 3C905C (3c59x driver) which behaved like a champ, and we didnt see the problems any more, so have stuck to that card. This was back in the 2.4.0 time frame, and there have many patches since then to various drivers, so not sure if the problem(s) have been resolved or not (likely to have been, extensively reported). Both your cards might actually be underperforming.. I'm a little confused here - do you mean the FA310TX (tulip driver) or the FA311/2 (natsemi driver)? I have not had any connection problems with either the FA310 or the FA311 cards. I haven't noticed any speed problems with the FA311 card, but I haven't benchmarked it, either. The FA310 is so horribly slow, I couldn't help but notice. Unfortunately, the same is true of the 3cSOHO. While I am willing to accept that both the FA310 and FA311 cards are underperforming, I think it is more than a little strange that the 3cSOHO card would turn in the same performance numbers. Also, keep in mind that I was only seeing horrible receive performance, TX performance seemed to be ok. I didn't post FTP numbers (both machines are running FTP servers). While the FTP performance numbers are probably not as scientific as Netperf, they do seem to agree from what I have observered. I.e. retrieving files from the P-90 machine is ok (~3MB/sec) but sending files to it is very slow (~100K/sec). This roughly agrees with the Netperf numbers I saw. FTP transfers to the FA311 machine (P2-350) are OK in both directions. Are you seeing any errors reported in /var/log/messages? Are you monitoring your connection via tcpdump, for example? You might sometimes see long gaps in transmission...Are there any abnormal numbers in /proc/net/ stats? I dont remember seeing that high frame errors, although there were a few. No, I don't see anything in /var/log/messages. I looked at tcpdump to try and figure it out, and it appeared that the P-90 was taking a very long time to ACK some packets. I am not a TCP/IP guru by any stretch, but my guess at the time was that the packets that were taking forever to get ACK'ed were the ones causing a framing error on the P-90, but again, I'm not an expert. The only unusual stat is the framing errors. There are a lot of them under heavy receive load. The machine will go for weeks without a single framing error, but if I blast some netperf action at it (or FTP send to it, etc.) then I get about 1/3 of the incoming packets (to the P-90) with framing errors. I see no other errors at all except a TX overrun error (maybe 1 in 10 packets). HW checksumming for the kind of test you are doing (tcp, mostly fast path) will not buy you any real performance gain, the checksum is actually consumed by the user-kernel copy routine. Ok, I'll take your word for it. The P-90 isn't a very fast machine to begin with, so I thought it could use all the HW assistance it could get (that and the 3cSOHO card was really cheap :-). I am very disappointed that TCP/IP performance on this machine is so lousy, but the problem is clearly with the kernel - just look at the performance numbers for 2.4.3 vs 2.2.19 (or 2.2.16). Those numbers aren't exactly great, but they are a lot better than 2.4.3. You can also run the tests on a profiling kernel and compare results... Nivedita --- Nivedita Singhvi(503) 578-4580 Linux Technology Center [EMAIL PROTECTED] IBM Beaverton, OR [EMAIL PROTECTED] Thanks for the assistance. Based on the benchmark information I have, I would say that there is a problem with the kernel and would like to pursue getting that fixed. I just can't justify why 2.4.3 should be 600% slower than
Abysmal RECV network performance
Can someone please help me troubleshoot this problem - I am getting abysmal (see numbers below) network performance on my system, but the poor performance seems limited to receiving data. Transmission is OK. The computer in question is a dual Pentium 90 machine. The machine has RedHat 7.0 (kernel 2.2.16-22 from RedHat). I have compiled 2.2.19 (stock) and 2.4.3 (stock) for the machine and used those for testing. I had a NetGear FA310TX card that I used with the "tulip" driver and a 3Com 3CSOHO card (Hurricane chipset) that I used with the "3c59x" driver. I used the netperf package to test performance (latest version, but I don't have the version number off-hand). The numbers netperf is giving me seem to correlate well to FTP statistics I see to the box. I have a second machine (P2-350) with a NetGear FA311 (running 2.4.3 and the "natsemi" driver) that I used to talk with the Pentium 90 machine. The two machines are connected through a NetGear FS105 10/100 switch. I also tried using a 10BT hub (see below). When connected, the switch indicated 100 Mbps, full duplex connections to both cards. This matches the speed indicator lights on both cards. I have run the miidiag program in the past to verify that the cards are actually set to full duplex, but I didn't run it again this time (this isn't the first time I have tried to chase this problem down). For the purposes of this message, call the P2-350 machine "A" and the dual P-90 machine "B". I ran the following tests: Machine "A" to localhost754.74 Mbps Kernel 2.2.19SMP Machine "B" to localhost80.63 Mbps Machine "B" to "A" (tulip) 55.38 Mbps Machine "A" to "B" (tulip) 10.60 Mbps Machine "A" to "B" (3c95x) 12.10 Mbps Kernel 2.4.3 SMP Machine "B" to localhost83.87 Mbps Machine "B" to "A" (tulip) 68.07 Mbps Machine "A" to "B" (tulip) 1.62Mbps Machine "A" to "B" (3c95x) 2.37Mbps Kernel 2.2.16-22 (RedHat kernel) Machine "B" to localhost92.29 Mbps Machine "B" to "A" (tulip) 57.34 Mbps Machine "A" to "B" (tulip) 9.98Mbps Machine "A" to "B" (3c95x) 9.05Mbps Now, with both "A" and "B" plugged into a 10BT hub: Kernel 2.2.19SMP Machine "B" to "A" (tulip) 6.96Mbps Machine "A" to "B" (tulip) 6.89Mbps At the end of the runs, I do not see any messages in syslog that would indicate a problem. Using the switch, there were no collisions but looking at /sbin/ifconfig there were a lot of "Frame:" errors on receive. "A lot" means ~30% of the total packets received. This happened with both cards and all kernels. The conclusions I draw from this data are: 1) Both machines connecting to localhost (data not going out over the wire) give reasonable numbers and are considerably above what I actually see going over the network (as would be expected). 2) The P-90 machine seems to have good transmit speed over both cards and all kernels. Transmit performance is close to the localhost numbers, so I can believe them. In the past, I have compared the performance of the FA310 to the 3ComSOHO card and there did not seem to be any measurable performance difference between the two. 3) Both the FA310 and the 3ComSOHO card have similar receive speeds, leading me to believe that the problem lies with either the machine or the kernel and not the individual cards or drivers. 4) Booting the machine as a uni-processor machine (with a non-SMP 2.2.16 kernel) did not change anything, so it does not appear to be a problem with SMP. 5) Kernel 2.4.3 receive performance is significantly lower than either 2.2.x kernel, so that tends to point to some fundamental problem in the kernel. 6) As I understand it, the 3Com card has some hardware acceleration for checksumming, and this is a slow machine, so why is the performance almost identical to the FA310? So, my questions are: What kind of performance should I be seeing with a P-90 on a 100Mbps connection? I was expecting something in the range of 40-70 Mbps - certainly not 1-2 Mbps. What can I do to track this problem down? Has anyone else had problems like this? Thanks in advance for any help you can offer. - John _ Get your FREE download of MSN Explorer at http://explorer.msn.com - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Abysmal RECV network performance
Can someone please help me troubleshoot this problem - I am getting abysmal (see numbers below) network performance on my system, but the poor performance seems limited to receiving data. Transmission is OK. The computer in question is a dual Pentium 90 machine. The machine has RedHat 7.0 (kernel 2.2.16-22 from RedHat). I have compiled 2.2.19 (stock) and 2.4.3 (stock) for the machine and used those for testing. I had a NetGear FA310TX card that I used with the tulip driver and a 3Com 3CSOHO card (Hurricane chipset) that I used with the 3c59x driver. I used the netperf package to test performance (latest version, but I don't have the version number off-hand). The numbers netperf is giving me seem to correlate well to FTP statistics I see to the box. I have a second machine (P2-350) with a NetGear FA311 (running 2.4.3 and the natsemi driver) that I used to talk with the Pentium 90 machine. The two machines are connected through a NetGear FS105 10/100 switch. I also tried using a 10BT hub (see below). When connected, the switch indicated 100 Mbps, full duplex connections to both cards. This matches the speed indicator lights on both cards. I have run the miidiag program in the past to verify that the cards are actually set to full duplex, but I didn't run it again this time (this isn't the first time I have tried to chase this problem down). For the purposes of this message, call the P2-350 machine A and the dual P-90 machine B. I ran the following tests: Machine A to localhost754.74 Mbps Kernel 2.2.19SMP Machine B to localhost80.63 Mbps Machine B to A (tulip) 55.38 Mbps Machine A to B (tulip) 10.60 Mbps Machine A to B (3c95x) 12.10 Mbps Kernel 2.4.3 SMP Machine B to localhost83.87 Mbps Machine B to A (tulip) 68.07 Mbps Machine A to B (tulip) 1.62Mbps Machine A to B (3c95x) 2.37Mbps Kernel 2.2.16-22 (RedHat kernel) Machine B to localhost92.29 Mbps Machine B to A (tulip) 57.34 Mbps Machine A to B (tulip) 9.98Mbps Machine A to B (3c95x) 9.05Mbps Now, with both A and B plugged into a 10BT hub: Kernel 2.2.19SMP Machine B to A (tulip) 6.96Mbps Machine A to B (tulip) 6.89Mbps At the end of the runs, I do not see any messages in syslog that would indicate a problem. Using the switch, there were no collisions but looking at /sbin/ifconfig there were a lot of Frame: errors on receive. A lot means ~30% of the total packets received. This happened with both cards and all kernels. The conclusions I draw from this data are: 1) Both machines connecting to localhost (data not going out over the wire) give reasonable numbers and are considerably above what I actually see going over the network (as would be expected). 2) The P-90 machine seems to have good transmit speed over both cards and all kernels. Transmit performance is close to the localhost numbers, so I can believe them. In the past, I have compared the performance of the FA310 to the 3ComSOHO card and there did not seem to be any measurable performance difference between the two. 3) Both the FA310 and the 3ComSOHO card have similar receive speeds, leading me to believe that the problem lies with either the machine or the kernel and not the individual cards or drivers. 4) Booting the machine as a uni-processor machine (with a non-SMP 2.2.16 kernel) did not change anything, so it does not appear to be a problem with SMP. 5) Kernel 2.4.3 receive performance is significantly lower than either 2.2.x kernel, so that tends to point to some fundamental problem in the kernel. 6) As I understand it, the 3Com card has some hardware acceleration for checksumming, and this is a slow machine, so why is the performance almost identical to the FA310? So, my questions are: What kind of performance should I be seeing with a P-90 on a 100Mbps connection? I was expecting something in the range of 40-70 Mbps - certainly not 1-2 Mbps. What can I do to track this problem down? Has anyone else had problems like this? Thanks in advance for any help you can offer. - John _ Get your FREE download of MSN Explorer at http://explorer.msn.com - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/