Re: Any ideas at all about network problem?
Brad Knowles wrote: > Other than that, you should try swapping out as much hardware as > you can -- the cards, the cables, etc If possible, you should > also test with other computers (in case the problem is with one > specific machine when it is running 5.0). Swapping 5.0 out for 4.7 will also work. 8-). Seriously: go through the intermediate versions of FreeBSD. Do you have a local CVS tree? Have you ever played the guessing game "higher or lower"? It only takes you at most log2(N)+1 "guesses", and you will have the day the change was made. Then you "cvs diff -r -r", and you will have the code that changed that day. Revert the change in the bus, rl, or other code that's in the code path, and you have your fix. Very easy. > Just because something appears to work perfectly in another OS is > not an indication that there is not anything wrong with that setup. Only with the second OS, when it fails to workaround the problem that the first OS works around without being asked... > However, since there are many potential software components that > could be involved, while testing each component individually between > now and then should theoretically be doable in 10 tests (as > previously mentioned), the combinatorial explosion will be > exceptionally nasty. You only need the delta for the day it was introduced, vs. the day before. Unless it was the KSE import that did it, it's very likely that it will be very obvious, in context. On an 800MHz box, you are talking 6 hours to do this, including all the tests that you have to run. If you have a fast box, you can cut this in half. Or if you have three fast boxes (so you can do branch prediction up and down on two while testing on the third), you can cut this to about 40 minutes. That's less than an hour, and far less time than it's taken the original poster to post messages in this thread, describing the problem. BTW: Good reason to buy hardware for your senior engineers there, and not complain when they ask for it, isn't it? 8-) 8-). -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
At 10:33 PM -0500 2002/12/04, Craig Reyenga wrote: Unfortunately, I have no extra hardware available to me, so I can't experiment with switches and whatnot. Also, wouldn't some sort of software experimentation be more appropriate, considering that my existing setup works _perfetcly_ in 4.7? At the very least, try hard-wiring the configuration at both ends to be 100Mbps full-duplex. There's a chance that 4.7 and 5.0 will handle auto-negotiation differently. Other than that, you should try swapping out as much hardware as you can -- the cards, the cables, etc If possible, you should also test with other computers (in case the problem is with one specific machine when it is running 5.0). Just because something appears to work perfectly in another OS is not an indication that there is not anything wrong with that setup. If that was the case, there would never be a need for any replacement for any Microsoft OSes, because many vendors stop trying to debug the problem when they can prove that things work just fine under Windows. I'm not sure what to do; should I be trying various versions of if_rl.c? Or is there something else that I should be trying? If you really want to try swapping software, you'll have to do a binary search on each potential piece of software involved. However, since there are many potential software components that could be involved, while testing each component individually between now and then should theoretically be doable in 10 tests (as previously mentioned), the combinatorial explosion will be exceptionally nasty. -- Brad Knowles, <[EMAIL PROTECTED]> "They that can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety." -Benjamin Franklin, Historical Review of Pennsylvania. GCS/IT d+(-) s:+(++)>: a C++(+++)$ UMBSHI$ P+>++ L+ !E W+++(--) N+ !w--- O- M++ V PS++(+++) PE- Y+(++) PGP>+++ t+(+++) 5++(+++) X++(+++) R+(+++) tv+(+++) b+() DI+() D+(++) G+() e++> h--- r---(+++)* z(+++) To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Unfortunately, I have no extra hardware available to me, so I can't experiment with switches and whatnot. Also, wouldn't some sort of software experimentation be more appropriate, considering that my existing setup works _perfetcly_ in 4.7? I'm not sure what to do; should I be trying various versions of if_rl.c? Or is there something else that I should be trying? -Craig - Original Message - From: "Brad Knowles" <[EMAIL PROTECTED]> To: <[EMAIL PROTECTED]> Cc: "Brad Knowles" <[EMAIL PROTECTED]>; "Craig Reyenga" <[EMAIL PROTECTED]>; "Terry Lambert" <[EMAIL PROTECTED]>; <[EMAIL PROTECTED]>; "Christopher J Olson" <[EMAIL PROTECTED]> Sent: Tuesday, December 03, 2002 19:32 Subject: Re: Any ideas at all about network problem? > At 12:31 PM +0200 2002/12/03, [EMAIL PROTECTED] wrote: > > > The two machines involved are connected by a crossover cable: > > I've heard of lots of problems with machines using cross-over > cables. Can you connect the machines through a switch, and ensure > that they are hard-wired to 100Base-TX full duplex at both ends, as > opposed to auto-negotiating? > > > I'll try a different cable this evening when I get home. Is there > > a minimum length? The cable is currently 2m long. I'm prepared > > to do any other debugging people here can suggest to make it work > > faster. FWIW my single CPU workstaion at the office running > > 4.7-STABLE with an fxp0 NIC does not suffer the same throughput > > reduction. > > I've also heard of lots of problems with some machines when the > cable is too short, at least in certain combinations. Try > successively longer lengths of cable, at least up to 20-30m. > > -- > Brad Knowles, <[EMAIL PROTECTED]> > > "They that can give up essential liberty to obtain a little temporary > safety deserve neither liberty nor safety." > -Benjamin Franklin, Historical Review of Pennsylvania. > > GCS/IT d+(-) s:+(++)>: a C++(+++)$ UMBSHI$ P+>++ L+ !E W+++(--) N+ !w--- > O- M++ V PS++(+++) PE- Y+(++) PGP>+++ t+(+++) 5++(+++) X++(+++) R+(+++) > tv+(+++) b+() DI+() D+(++) G+() e++> h--- r---(+++)* z(+++) > > To Unsubscribe: send mail to [EMAIL PROTECTED] > with "unsubscribe freebsd-current" in the body of the message > To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Brad Knowles wrote: > At 12:31 PM +0200 2002/12/03, [EMAIL PROTECTED] wrote: > > > The two machines involved are connected by a crossover cable: > > I've heard of lots of problems with machines using cross-over > cables. Can you connect the machines through a switch, and ensure > that they are hard-wired to 100Base-TX full duplex at both ends, as > opposed to auto-negotiating? They do this already. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Joshua Goodall wrote: > On Wed, Dec 04, 2002 at 01:32:58AM +0100, Brad Knowles wrote: > > At 12:31 PM +0200 2002/12/03, [EMAIL PROTECTED] wrote: > > > > > The two machines involved are connected by a crossover cable: > > > > I've heard of lots of problems with machines using cross-over > > cables. Can you connect the machines through a switch, and ensure > > that they are hard-wired to 100Base-TX full duplex at both ends, as > > opposed to auto-negotiating? > > Just to add to this a-la-carte selection of possibilities; I recently > debugged a NIC throughput problem which was resolved by a flash BIOS > upgrade. This was an ASUS A7VL-VM motherboard, which appeared to do > unpleasant things to PCI IRQs. Wipes egg off face: it seems that most of the bottleneck was an old drive in the one machine, although systat showed it as only 67% busy. If I fetch a file off the machine with the slow disk twice, the second time it's served from cache and the transfer rate is 60.9 megabits per second (test file is 90984206 bytes). The receiving machine's disk is reported as 10% busy by systat. Much better, but I'd still expect the ethernet to be saturated since transferring data to faster disk using the transaction size from systat yields about 143 megabits per second. [brane] ~ $ dd if=/dev/zero of=t bs=126k count=705 705+0 records in 705+0 records out 90961920 bytes transferred in 5.100551 secs (17833744 bytes/sec) I'll try a long cable tonight and see if I can borrow a switch from work for the evening. Ian To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
On Wed, Dec 04, 2002 at 01:32:58AM +0100, Brad Knowles wrote: > At 12:31 PM +0200 2002/12/03, [EMAIL PROTECTED] wrote: > > > The two machines involved are connected by a crossover cable: > > I've heard of lots of problems with machines using cross-over > cables. Can you connect the machines through a switch, and ensure > that they are hard-wired to 100Base-TX full duplex at both ends, as > opposed to auto-negotiating? Just to add to this a-la-carte selection of possibilities; I recently debugged a NIC throughput problem which was resolved by a flash BIOS upgrade. This was an ASUS A7VL-VM motherboard, which appeared to do unpleasant things to PCI IRQs. Joshua -- Joshua Goodall [EMAIL PROTECTED] "Your byte hit ratio is weak, old man" "If you cache me now, I will dump more core than you can possibly imagine" To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
At 12:31 PM +0200 2002/12/03, [EMAIL PROTECTED] wrote: The two machines involved are connected by a crossover cable: I've heard of lots of problems with machines using cross-over cables. Can you connect the machines through a switch, and ensure that they are hard-wired to 100Base-TX full duplex at both ends, as opposed to auto-negotiating? I'll try a different cable this evening when I get home. Is there a minimum length? The cable is currently 2m long. I'm prepared to do any other debugging people here can suggest to make it work faster. FWIW my single CPU workstaion at the office running 4.7-STABLE with an fxp0 NIC does not suffer the same throughput reduction. I've also heard of lots of problems with some machines when the cable is too short, at least in certain combinations. Try successively longer lengths of cable, at least up to 20-30m. -- Brad Knowles, <[EMAIL PROTECTED]> "They that can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety." -Benjamin Franklin, Historical Review of Pennsylvania. GCS/IT d+(-) s:+(++)>: a C++(+++)$ UMBSHI$ P+>++ L+ !E W+++(--) N+ !w--- O- M++ V PS++(+++) PE- Y+(++) PGP>+++ t+(+++) 5++(+++) X++(+++) R+(+++) tv+(+++) b+() DI+() D+(++) G+() e++> h--- r---(+++)* z(+++) To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Sure, I'm not sure what to tell you though. If you can tell me what info you need, then I'll find it for you. I sense a small game of chicken meets egg forming here. -Craig - Original Message - From: "Cliff L. Biffle" <[EMAIL PROTECTED]> > > Well... > My Realtek card in my 5.0 workstation is fully capable of saturating a 100mbps > link. It's an 8139; the only thing I did that might have been unusual was > forcing the media/mediaopts after bad experiences with rl's autodetect in the > past. Perhaps if you give me more information about your setup I could > better reproduce it. :-\ > > -Cliff L. Biffle > To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
On Monday 02 December 2002 12:09 pm, Craig Reyenga wrote: > Right on. I hope that you find something because right now it seems > so hopeless. I'd have to say that this is the strangest problem that I've > ever had with FreeBSD. Well... My Realtek card in my 5.0 workstation is fully capable of saturating a 100mbps link. It's an 8139; the only thing I did that might have been unusual was forcing the media/mediaopts after bad experiences with rl's autodetect in the past. Perhaps if you give me more information about your setup I could better reproduce it. :-\ -Cliff L. Biffle To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Brad Knowles writes: > At 3:32 PM -0700 2002/12/02, Cliff L. Biffle wrote: > > > One thing I've used in the past that improves Realtek throughput is forcing > > the media type and duplex setting on both ends of the connection. Autodetect > > in the 8139s seems to be unreliable at times. > > This is true for most 10/100 Base-T implementations I've seen. > None of them have been able to reliably auto-detect. Any time I hear > someone complain of network throughput problems, this is one of the > first things I have them check. However, this would not seem to be > the case in this instance, unless 4.7 and 5.x are not handling > auto-detect in the same way. Phk made some large (and somewhat controversial, at the time) changes to the mii support code in -current about 7 months ago. Before blaming the mii code, I'd feel better if we knew a hardcoded connection worked better. Drew To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Brad Knowles wrote: > At 12:55 AM -0500 2002/12/02, Craig Reyenga wrote: > > > I just tried a 3com 3c905 NIC (my roommate's) and it _also_ > > transfers slowly (about 3.5MB/sec, so just under half of what i used to > > get with my realtek in -stable). It also spit out a few messages: > > According to all the source modules I've read regarding RealTek > cards, they're about the biggest pieces of hardware garbage that has > ever been inflicted on the free/open community. However, a 3Com card > should be a little better. Have you tried replacing both RealTek > cards with 3Com, to see if that changes things? I have a similar problem with my xl0 and ed0 which has been around for just about as long as I can remember. I'd originally thought it was a cable problem, but since I still get transfer rates in excess of my access line speed, I'd never really debugged it. The two machines involved are connected by a crossover cable: 1. FreeBSD 5.0-CURRENT #7: Mon Dec 2 19:03:06 SAST 2002 Dual PII-266, 196MB xl0: <3Com 3c905-TX Fast Etherlink XL> port 0xe400-0xe43f irq 2 at device 8.0 on pci0 xl0: Ethernet address: 00:60:08:0b:2e:c8 miibus0: on xl0 [brane-dead] /usr/src/sys/i386/conf # ifconfig xl0 xl0: flags=8843 mtu 1500 inet 196.7.162.26 netmask 0xfffc broadcast 196.7.162.27 ether 00:60:08:0b:2e:c8 media: Ethernet 100baseTX status: active 2. FreeBSD 4.7-STABLE #3: Sun Nov 24 18:29:27 SAST 2002 Dual PIII-700, 256MB xl0: <3Com 3c905-TX Fast Etherlink XL> port 0xd800-0xd83f irq 12 at device 10.0 on pci0 xl0: Ethernet address: 00:60:08:10:10:31 miibus0: on xl0 [brane] ~ $ ifconfig xl0 xl0: flags=8843 mtu 1500 inet 196.7.162.25 netmask 0xfffc broadcast 196.7.162.27 ether 00:60:08:10:10:31 media: Ethernet 100baseTX status: active When I had two SMC Ultra NICs in these machines (both 4.x-STABLE at the time) I got transfer rates of about 1.1 megabits per second. Using the same cable, but with 3c905 NICs I get 25.3 megabits per second and without ipfw I get 26.7 megabits per second. 'top' showed the ftpd on the CURRENT machine spending a lot of time in *Giant and both machines had a lot of idle CPU time >50%. I'll try a different cable this evening when I get home. Is there a minimum length? The cable is currently 2m long. I'm prepared to do any other debugging people here can suggest to make it work faster. FWIW my single CPU workstaion at the office running 4.7-STABLE with an fxp0 NIC does not suffer the same throughput reduction. Ian To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
At 3:32 PM -0700 2002/12/02, Cliff L. Biffle wrote: One thing I've used in the past that improves Realtek throughput is forcing the media type and duplex setting on both ends of the connection. Autodetect in the 8139s seems to be unreliable at times. This is true for most 10/100 Base-T implementations I've seen. None of them have been able to reliably auto-detect. Any time I hear someone complain of network throughput problems, this is one of the first things I have them check. However, this would not seem to be the case in this instance, unless 4.7 and 5.x are not handling auto-detect in the same way. -- Brad Knowles, <[EMAIL PROTECTED]> "They that can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety." -Benjamin Franklin, Historical Review of Pennsylvania. GCS/IT d+(-) s:+(++)>: a C++(+++)$ UMBSHI$ P+>++ L+ !E W+++(--) N+ !w--- O- M++ V PS++(+++) PE- Y+(++) PGP>+++ t+(+++) 5++(+++) X++(+++) R+(+++) tv+(+++) b+() DI+() D+(++) G+() e++> h--- r---(+++)* z(+++) To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
On Monday 02 December 2002 03:05 am, Brad Knowles wrote: > According to all the source modules I've read regarding RealTek > cards, they're about the biggest pieces of hardware garbage that has > ever been inflicted on the free/open community. However, a 3Com card > should be a little better. Have you tried replacing both RealTek > cards with 3Com, to see if that changes things? One thing I've used in the past that improves Realtek throughput is forcing the media type and duplex setting on both ends of the connection. Autodetect in the 8139s seems to be unreliable at times. In my testing here I can get upwards of 1100kBps over 10baseT as long as I set both ends to full-duplex 10baseT/UTP. (This is from a 5.0-DP2 box to a 4.5-RELEASE box, 8139's on both ends; scp and nfs show about the same throughput.) iirc, Craig was seeing closer to 800kBps. I'm trying to find a patch cable (or a crimper) to try 100mbps. -Cliff L. Biffle To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
At 12:55 AM -0500 2002/12/02, Craig Reyenga wrote: I just tried a 3com 3c905 NIC (my roommate's) and it _also_ transfers slowly (about 3.5MB/sec, so just under half of what i used to get with my realtek in -stable). It also spit out a few messages: According to all the source modules I've read regarding RealTek cards, they're about the biggest pieces of hardware garbage that has ever been inflicted on the free/open community. However, a 3Com card should be a little better. Have you tried replacing both RealTek cards with 3Com, to see if that changes things? I'd really rather not play around with different versions of FreeBSD to fix this problem, because this computer is where I keep all of my stuff, and with exams, I just won't have the time. Yes I know that I "shouldn't be using 5.0 then" but a problem is a problem and it should be fixed. Yeah, well. Regrettably, while 5.x-CURRENT needs as many testers as it can get before 5.0-RELEASE, it is not something that people should be using for critical work. I agree that this is a problem, but what we need right now are people who can help us find problems (which you've done) but then can also go the next steps and help us locate the problem (which you are unwilling or unable to do). As soon as I get a couple of other things squared away, I'm going to take the leap myself and start trying to run -CURRENT, but with the commitment that I will do everything I can to locate any bugs that I find. But this means that I need to make sure that my FreeBSD box is doing useful (but not critical) stuff, so that I can give it a semi "real world" test. At the very least, try DP1. If that doesn't work, then the changes happened somewhere earlier, and they will be more difficult to track down. Either that, or the problem is actually somewhere else, and you're only seeing the results through your network transfers. -- Brad Knowles, <[EMAIL PROTECTED]> "They that can give up essential liberty to obtain a little temporary safety deserve neither liberty nor safety." -Benjamin Franklin, Historical Review of Pennsylvania. GCS/IT d+(-) s:+(++)>: a C++(+++)$ UMBSHI$ P+>++ L+ !E W+++(--) N+ !w--- O- M++ V PS++(+++) PE- Y+(++) PGP>+++ t+(+++) 5++(+++) X++(+++) R+(+++) tv+(+++) b+() DI+() D+(++) G+() e++> h--- r---(+++)* z(+++) To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Right on. I hope that you find something because right now it seems so hopeless. I'd have to say that this is the strangest problem that I've ever had with FreeBSD. -Craig - Original Message - From: "Cliff L. Biffle" <[EMAIL PROTECTED]> To: "Craig Reyenga" <[EMAIL PROTECTED]>; <[EMAIL PROTECTED]> Sent: Monday, December 02, 2002 13:11 Subject: Re: Any ideas at all about network problem? On Monday 02 December 2002 10:47 am, Craig Reyenga wrote: > Ok, I'm convinced. Clearly I'm the one that has to do the testing > because I seem to be the lucky guy with the problem. I'm actually on my way to the office now to set up a test scenario with our 5-current boxen. We've got a whole scad of 8139s, 8129s, 3c905s, and some old Davicom-based cards that gave me endless trouble under 4.x. I'll let you know what I find. For reference, the 8139 in my 5-current box here works at full speed, but I'm on 10mbps; we have more 100baseT equipment at work. -Cliff L. Biffle To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
On Monday 02 December 2002 10:47 am, Craig Reyenga wrote: > Ok, I'm convinced. Clearly I'm the one that has to do the testing > because I seem to be the lucky guy with the problem. I'm actually on my way to the office now to set up a test scenario with our 5-current boxen. We've got a whole scad of 8139s, 8129s, 3c905s, and some old Davicom-based cards that gave me endless trouble under 4.x. I'll let you know what I find. For reference, the 8139 in my 5-current box here works at full speed, but I'm on 10mbps; we have more 100baseT equipment at work. -Cliff L. Biffle To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Ok, I'm convinced. Clearly I'm the one that has to do the testing because I seem to be the lucky guy with the problem. The super weird thing about all of this is that cpu usage is very minimal during transfers. The 905 card was weird too: it actually ran at 8MB/sec for about 4 sec, then the kernel gave a few TX errors and then it stayed at or below 3.7MB/sec. I have also tried my friends laptop as an alternate client and it showed identical behaviour as my usual WinXP box. My setup has no switches or anything, just a white cable. If I need to give more information, just ask and I'll try to fetch it. I've already given a full dmesg and a few other things, so I'm not sure what to say at this point. Thanks in advance, -Craig - Original Message - From: "Terry Lambert" <[EMAIL PROTECTED]> > Craig Reyenga wrote: > > I just tried a 3com 3c905 NIC (my roommate's) and it _also_ > > transfers slowly (about 3.5MB/sec, so just under half of what i used to > > get with my realtek in -stable). It also spit out a few messages: > > [ ... ] > > > I'd really rather not play around with different versions of FreeBSD to > > fix this problem, because this computer is where I keep all of my stuff, > > and with exams, I just won't have the time. Yes I know that I "shouldn't be > > using 5.0 then" but a problem is a problem and it should be fixed. > > Your alternative to doing the necessary binary search is to > provide enough information that someone else can repeat the > reduced performance you are getting with your hardware, so > that they can perform the binary search on your behalf. > > FWIW, the root cause is likely a result of something in the > last 8 months, which means log2(240)+1 = 8 compiles to find > the problem on your hardware; if, in the last 2.5 years, > which we know to be the case, it's log2(2.5*365)+1 = 10 > compiles. > > You already have hardware to test the kernels out on, to see > if a particular version has the problem, so you're the logical > candidate to do the compiling. Given a 1GHz machine, we are > probably talking about 6 hours elapsed time, given a local CVS > tree, and compiling and testing occurring serially. > > If you don't want to do the work, you are going to have to > provide a better characterization of the problem so that it > can be repeated by someone who's willing to do it on your > behalf, or out of curiousity; most people who could deal with > it for you aren't the types to buy RealTek or 3C905 ethernet > cards, which the driver comments suggest are badly designed > hardware. > > -- Terry > > To Unsubscribe: send mail to [EMAIL PROTECTED] > with "unsubscribe freebsd-current" in the body of the message > To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
On Mon, Dec 02, 2002 at 05:18:34PM +0300, Igor Roboul wrote: > Maybe this is _CABLE_? oops :-( All works fine with 4-STABLE? -- Igor Roboul, System administrator at Speech Technology Center http://www.speechpro.com http://www.speechpro.ru To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
On Mon, Dec 02, 2002 at 12:55:39AM -0500, Craig Reyenga wrote: > I just tried a 3com 3c905 NIC (my roommate's) and it _also_ > transfers slowly (about 3.5MB/sec, so just under half of what i used to > get with my realtek in -stable). It also spit out a few messages: Maybe this is _CABLE_? -- Igor Roboul, System administrator at Speech Technology Center http://www.speechpro.com http://www.speechpro.ru To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
On Mon, Dec 02, 2002 at 03:15:22AM -0800, Terry Lambert wrote: > Cliff Sarginson wrote: > > On Mon, Dec 02, 2002 at 01:27:34AM -0800, Terry Lambert wrote: > > > [ ... bad throughput on bad NICs ... ] > > > > Mmmm. I use these RTL cheapo nics. I accept the fact they have a bad > > reputation. However I have used them for some time, and they have > > behaved impeccably. I have noticed no change in throughput on 5.0 > > either, and since I have downloaded loads off stuff which comes through > > my firewall and then via the LAN to the 5.0 test machine I think I would > > have noticed if it was slower than normal. > > Having said that I have not checked the throughput, but then nothing has > > happened to make me want to check it. I run 4-7 Stable updated at least > > once a week, and the latest patched 5.0 from a few days ago. > > You may want to check it. You can at least confirm whether his > problem is his alone, or if it's shared by others. I suspect > that it will be specific to him, actually, given other factors > (i.e. it might be specifically an rl0 <-> rl0 only problem, etc.). > Ok, next time I do some largish downloads I will collect the stats to see if there is anything significant -- but as I say subjectively everything feels just about the same. > > > The network also runs through a fairly cheapo switch. > > It could be a duplex problem, but none of that code is different > from 4.7, and he says it works there, dso that's not going to be > it. > > I expect that it's his specific cards from a specific vendor, > and the only one who will be able to find the change that causes > his problem is him (he's the only one complaining about it, at > this point). > Yes I guess. They are pretty widely used despite their reputation ! The problem is the price differential between them and a super-whizzy card is pretty phenomenal, at least here in Holland. Having said that if I perceived it as a problem I would save my pennies up... -- Regards Cliff Sarginson The Netherlands [ This mail has been checked as virus-free ] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Cliff Sarginson wrote: > On Mon, Dec 02, 2002 at 01:27:34AM -0800, Terry Lambert wrote: > > [ ... bad throughput on bad NICs ... ] > > Mmmm. I use these RTL cheapo nics. I accept the fact they have a bad > reputation. However I have used them for some time, and they have > behaved impeccably. I have noticed no change in throughput on 5.0 > either, and since I have downloaded loads off stuff which comes through > my firewall and then via the LAN to the 5.0 test machine I think I would > have noticed if it was slower than normal. > Having said that I have not checked the throughput, but then nothing has > happened to make me want to check it. I run 4-7 Stable updated at least > once a week, and the latest patched 5.0 from a few days ago. You may want to check it. You can at least confirm whether his problem is his alone, or if it's shared by others. I suspect that it will be specific to him, actually, given other factors (i.e. it might be specifically an rl0 <-> rl0 only problem, etc.). > The network also runs through a fairly cheapo switch. It could be a duplex problem, but none of that code is different from 4.7, and he says it works there, dso that's not going to be it. I expect that it's his specific cards from a specific vendor, and the only one who will be able to find the change that causes his problem is him (he's the only one complaining about it, at this point). -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
On Mon, Dec 02, 2002 at 01:27:34AM -0800, Terry Lambert wrote: > [ ... bad throughput on bad NICs ... ] > Mmmm. I use these RTL cheapo nics. I accept the fact they have a bad reputation. However I have used them for some time, and they have behaved impeccably. I have noticed no change in throughput on 5.0 either, and since I have downloaded loads off stuff which comes through my firewall and then via the LAN to the 5.0 test machine I think I would have noticed if it was slower than normal. Having said that I have not checked the throughput, but then nothing has happened to make me want to check it. I run 4-7 Stable updated at least once a week, and the latest patched 5.0 from a few days ago. The network also runs through a fairly cheapo switch. Maybe I just get lucky, touch wood (or my anti-static wrist band :). My problem is that extensive use of my hoover sometimes makes some of the computers stop, but I don't think a slightly overloaded electrical system can be solved in software... -- Regards Cliff Sarginson The Netherlands [ This mail has been checked as virus-free ] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
[ ... bad throughput on bad NICs ... ] Nate Lawson wrote: > > FWIW, the root cause is likely a result of something in the > > last 8 months, which means log2(240)+1 = 8 compiles to find > > the problem on your hardware; if, in the last 2.5 years, > > which we know to be the case, it's log2(2.5*365)+1 = 10 > > compiles. > > My guess is mii_tick() latency. Off the top of my head, I would blame one of: 1.79don't strip the Ethernet header for input packets 1.72If we get 0x back when reading the status register, assume the card has gone away instead of spinning in the interrupt handler. This stops my machine from hanging when I eject a rl(4)-based cardbus card. 1.71Fix if_timer logic to make sure that there is always a timeout pending if there are packets queued for transmission. 1.69Add DEVICE_POLLING support to the "rl" driver. The diffs are very similar to the ones for the "sis" driver. If I had to place a bet, I would bet it was 1.69, even though he said it happens with/without polling, he didn't say whether he disabled it as a compilation option or just by using the sysctl. If he used the sysctl method, that means an extra registration function call in rl_intr(), each time an interrupt happens, which is nothing (but it does affect the benchmarks, making polling look betther than it really is, in a side-by-side comparison), and the rl_poll code gets called on all the clock ticks, potentially skewing the hardware state: the ether_poll code should probably also check that IFF_POLLING is set. If it's not that, then my next favorite to blame would be 1.79, which seems to me is a *massive* change. I pick both of those because they are relatively driver-agnostic changes, and he said they happened on both drivers. My third choice, is "something else in the system was changed", which makes it better to do the binary search. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
On Mon, 2 Dec 2002, Terry Lambert wrote: > Craig Reyenga wrote: > > I just tried a 3com 3c905 NIC (my roommate's) and it _also_ > > transfers slowly (about 3.5MB/sec, so just under half of what i used to > > get with my realtek in -stable). It also spit out a few messages: > > [ ... ] > > > I'd really rather not play around with different versions of FreeBSD to > > fix this problem, because this computer is where I keep all of my stuff, > > and with exams, I just won't have the time. Yes I know that I "shouldn't be > > using 5.0 then" but a problem is a problem and it should be fixed. > > Your alternative to doing the necessary binary search is to > provide enough information that someone else can repeat the > reduced performance you are getting with your hardware, so > that they can perform the binary search on your behalf. > > FWIW, the root cause is likely a result of something in the > last 8 months, which means log2(240)+1 = 8 compiles to find > the problem on your hardware; if, in the last 2.5 years, > which we know to be the case, it's log2(2.5*365)+1 = 10 > compiles. My guess is mii_tick() latency. -Nate To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Craig Reyenga wrote: > I just tried a 3com 3c905 NIC (my roommate's) and it _also_ > transfers slowly (about 3.5MB/sec, so just under half of what i used to > get with my realtek in -stable). It also spit out a few messages: [ ... ] > I'd really rather not play around with different versions of FreeBSD to > fix this problem, because this computer is where I keep all of my stuff, > and with exams, I just won't have the time. Yes I know that I "shouldn't be > using 5.0 then" but a problem is a problem and it should be fixed. Your alternative to doing the necessary binary search is to provide enough information that someone else can repeat the reduced performance you are getting with your hardware, so that they can perform the binary search on your behalf. FWIW, the root cause is likely a result of something in the last 8 months, which means log2(240)+1 = 8 compiles to find the problem on your hardware; if, in the last 2.5 years, which we know to be the case, it's log2(2.5*365)+1 = 10 compiles. You already have hardware to test the kernels out on, to see if a particular version has the problem, so you're the logical candidate to do the compiling. Given a 1GHz machine, we are probably talking about 6 hours elapsed time, given a local CVS tree, and compiling and testing occurring serially. If you don't want to do the work, you are going to have to provide a better characterization of the problem so that it can be repeated by someone who's willing to do it on your behalf, or out of curiousity; most people who could deal with it for you aren't the types to buy RealTek or 3C905 ethernet cards, which the driver comments suggest are badly designed hardware. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
I just tried a 3com 3c905 NIC (my roommate's) and it _also_ transfers slowly (about 3.5MB/sec, so just under half of what i used to get with my realtek in -stable). It also spit out a few messages: xl0: <3Com 3c905-TX Fast Etherlink XL> port 0xe800-0xe83f irq 5 at device 8.0 on pci0 xl0: Ethernet address: 00:60:08:91:4c:fe miibus0: on xl0 nsphy0: on miibus0 nsphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto xl0: transmission error: 90 xl0: tx underrun, increasing tx start threshold to 120 bytes xl0: transmission error: 90 xl0: tx underrun, increasing tx start threshold to 180 bytes xl0: transmission error: 90 xl0: tx underrun, increasing tx start threshold to 240 bytes xl0: transmission error: 90 xl0: tx underrun, increasing tx start threshold to 300 bytes I'd really rather not play around with different versions of FreeBSD to fix this problem, because this computer is where I keep all of my stuff, and with exams, I just won't have the time. Yes I know that I "shouldn't be using 5.0 then" but a problem is a problem and it should be fixed. -Craig - Original Message - From: "Terry Lambert" <[EMAIL PROTECTED]> To: "Craig Reyenga" <[EMAIL PROTECTED]> Cc: <[EMAIL PROTECTED]> Sent: Monday, December 02, 2002 00:06 Subject: Re: Any ideas at all about network problem? > Craig Reyenga wrote: > > It worked fine in 4.7 and all previous versions, just DP2 dunno about DP1. > > Well, you will have to back up to a version of the source code > before DP2 that didn't have the problem, perform a binary search > to find the exact delta that caused the problem, and examine the > code differences in order to find the problem change, and why it > causes the problem. > > Personally, I'd start with DP1, but that's because I have a > CDROM locally, and the CVS tree is not always buildable, since > there is no software enforcement of buildability before a change > is committed. > > There are almost 2 years worth of changes in the things which > were not brough back to the -STABLE branches from -CURRENT, so > diffing 4.7 and DP2 isn't likely to get you anywhere, I think. > > -- Terry > To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Craig Reyenga wrote: > It worked fine in 4.7 and all previous versions, just DP2 dunno about DP1. Well, you will have to back up to a version of the source code before DP2 that didn't have the problem, perform a binary search to find the exact delta that caused the problem, and examine the code differences in order to find the problem change, and why it causes the problem. Personally, I'd start with DP1, but that's because I have a CDROM locally, and the CVS tree is not always buildable, since there is no software enforcement of buildability before a change is committed. There are almost 2 years worth of changes in the things which were not brough back to the -STABLE branches from -CURRENT, so diffing 4.7 and DP2 isn't likely to get you anywhere, I think. -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
It worked fine in 4.7 and all previous versions, just DP2 dunno about DP1. -Craig - Original Message - From: "Terry Lambert" <[EMAIL PROTECTED]> To: "Craig Reyenga" <[EMAIL PROTECTED]> Cc: <[EMAIL PROTECTED]> Sent: Sunday, December 01, 2002 23:38 Subject: Re: Any ideas at all about network problem? > Craig Reyenga wrote: > > In a recent thread started by me, named "Network is crazy slow in DP2" > > I wrote that I'm getting substantially lower speeds than I should be > > over my 100mbit link (realtek 8139 on both sides). > > DP1 have the problem? > > 4.7? > > cvs diff DP2 DP1? > > -- Terry > > To Unsubscribe: send mail to [EMAIL PROTECTED] > with "unsubscribe freebsd-current" in the body of the message > To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
Re: Any ideas at all about network problem?
Craig Reyenga wrote: > In a recent thread started by me, named "Network is crazy slow in DP2" > I wrote that I'm getting substantially lower speeds than I should be > over my 100mbit link (realtek 8139 on both sides). DP1 have the problem? 4.7? cvs diff DP2 DP1? -- Terry To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message