Re: Ideas on improving single connection bandwidth?
Scott Radvan wrote: On Fri, 09 Mar 2007 16:57:37 -0600 Jeff Bromberger [EMAIL PROTECTED] wrote: Hello, I have listed my dmesg and ifconfig at the end of this post for reference. The problem I'm having is that any ftp or lynx traffic to an external (ie. outside of my LAN) host comes in at 200KB/sec. Meanwhile, my Windows boxes which are on the same switch, router and cable modem, can pull down from the net at over 800KB/sec. I have swapped cables, tried different ports on the switch and done every other type of A/B test to no avail. I have also tried both physical nics on the off chance that I had a hardware issue with one of them. When I transfer between the OpenBSD machine and an XP machine over the gigabit ethernet (again using ftp) I get 7500KB/sec. That's obviously a lot better, but when I transfer between two XP machines (again with ftp) I get 17000KB/sec. But here's where it gets really weird (to me at least). I can run 3 ftp or lynx sessions at the same time and get 200KB/sec on each of them. So obviously the NIC, router, cable modem, etc. are capable of the total throughput that I'm after, but I can't seem to get it in any one connection. The throughput is always so close to 200KB/sec regardless of the client, server, server load, etc. that I can't help but think that there's some kind of throttling limitation in effect here. Of course the fact that the local LAN traffic is not limited to this rate blows that theory, unless the limitation is actually based on the media type (100MBit vs. 1000MBit). I'm pretty much out of ideas! Any thoughts? You might like to check your net.inet.tcp.recvspace and net.inet.tcp.sendspace sysctl settings, as mentioned here in the FAQ, Section 6.6.4. http://www.openbsd.org/faq/faq6.html#Tuning I saw a similar problem on my ADSL2+ line, until changing these particular parameters. With any luck, it will allow you to make full use of your connection speed and resolve the issues you are having. Wow, what a difference. THANK YOU so much. I'm now getting 700KB/sec instead of 200KB/sec. That's with those parameters set to 131072 instead of 16384 the default. I tried 256k and it didn't see to improve any more so I backed it back down. Thanks again! Jeff
panics on amd64 snapshot
I have a dual core Opteron system that I'm trying to make into a mail server for my company to replace a 7 year old Linux box that's on its last leg. I started off using the 3.9 release of the amd64 system and ran into a few problems (keyboard and cdrom didn't work). It was suggested that I move to the latest snapshot, which I did about a week ago. That fixed the keyboard and cdrom problems, so I began configuring the box. I am only running a few packages on this machine: courier imap, postfix, fetchmail, procmail. In the past 2-3 days (which is how long the box has actually been active, i.e. running all the daemons and having mail clients connect to it) I have experienced two kernel panics. I thought that the info from the panic would show up in the dmesg after rebooting, but that data seems to have been corrupted. I could only see bits and pieces of the kernel debugger message from the first panic. This time around, I'm still sitting in the kernel debugger so I am able to run a few commands if anybody has any specific requests. Trace shows: panic() at panic+0x12a amap_wipeout() at amap_wipeout+0x71 uvm_unmap_detach() at uvm_unmap_detach+0x9b sys_munmap() at sys_munmap+0x145 syscall() at syscall+0x25c --- syscall (number 73) --- ps shows the active process was imapd. I am running the bsd.mp kernel from the amd64 snapshot. To ask a different question, for the hardware that I have, what would be the most stable port/version that I could run? Am I better off going with the 3.9 release of the i386 code vs. the current snapshot of the amd64 stuff? My top priorities for this box are stability first and then security second, performance is a distant third since it's just a mail server for a small company. Thanks, Jeff
out of swap error?
Hello, I've got a freshly installed amd64 (dual core opteron) system running the latest snapshot (bsd.mp kernel). I have 1GB of ram and a 1GB swap partition. The machine has next to nothing on it at this point, not even X. The only packages I have installed are postfix, fetchmail and procmail (all from the latest openbsd snapshot). After letting the system sit for about 24 hours I sat down on the console to find a dozen or so out of swap errors like this: UVM: pid 23358 (sshd), uid 0 killed: out of swap UVM: pid 11042 (newsyslog), uid 0 killed: out of swap etc... I had an ssh session already open into the box so I was able to execute top. Top showed very little memory usage (about 10MB) and no swap usage. I've searched the net and usenet for similar issues but the only posters I see having such problems are people trying to run extremely minimal systems like 16MB of ram and no swap partition. Any ideas on how I could be out of swap when I seemingly wasn't using hardly any RAM and no swap? Thanks, Jeff
New system, couple of issues with amd64 MP kernel?
I just built a dual-core opteron system (some of you may recall my bad hardware last week that put a damper on my OpenBSD installation) and everything went very smoothly with the default amd64 /bsd kernel (installing from the v3.9 cd's). However, my understanding is that in order to take advantage of the dual-core properties of my cpu I need to use the /bsd.mp kernel (please correct me if that's wrong). The machine boots fine with the /bsd.mp kernel, however there are two problems: a) the keyboard is dead. b) the cdrom errors out and cannot be mounted. Both of these items function perfectly under the /bsd kernel. Here is a link to a dmesg that shows the /bsd.mp boot followed by the cdrom errors followed by the /bsd boot. http://members.cox.net/supra/dm.txt Any ideas on what is going on? If there is some kind of issue with the amd64 MP kernel at this time (and I'm probably getting way, way ahead of myself) would I be better off performance wise on my particular CPU running the i386 MP kernel or the amd64 non-MP kernel? Thanks, Jeff
Re: New system, couple of issues with amd64 MP kernel?
I just built a dual-core opteron system (some of you may recall my bad hardware last week that put a damper on my OpenBSD installation) and everything went very smoothly with the default amd64 /bsd kernel (installing from the v3.9 cd's). However, my understanding is that in order to take advantage of the dual-core properties of my cpu I need to use the /bsd.mp kernel (please correct me if that's wrong). The machine boots fine with the /bsd.mp kernel, however there are two problems: a) the keyboard is dead. b) the cdrom errors out and cannot be mounted. Both of these items function perfectly under the /bsd kernel. Here is a link to a dmesg that shows the /bsd.mp boot followed by the cdrom errors followed by the /bsd boot. http://members.cox.net/supra/dm.txt At this point, very close to release, you would be much better off trying a snapshot. If you can do this, it will help the project test for release, and you have a much better chance that the bug has been fixed in the meantime. Thanks for the suggestion Tom. I installed the latest amd64 4.0 snapshot and everything seems to be working fine now with both the bsd and bsd.mp kernels. I have submitted a new dmesg, let me know if there is something else useful (for the developers) that I could do to highlight what changed between 3.9 and the 4.0 snapshot. Thanks again, Jef
Kernel panic in openssl on fresh minimal install
I'm a new openbsd user (or I should say I'm attempting to be) and I'm not having a ton of luck here. I bought the cd set (i386) and it arrived yesterday. During the install, the base39.tgz file seemed to be corrupt and the install would crash, the kernel would panic and the machine would reboot. This happened while processing the perl files in the archive FWIW. Anyway, I then switched over to doing an ftp install and that seemed to make it through the packages ok. I only did the base packages and no X stuff. I did select sshd and ntpd to run by default. After completing the install and rebooting, openssl appears to crash . I was going to boot off the cd and change the config to skip sshd from starting, but I wanted to post this to the list first in case anybody else wants me to run an additional debugger command. I'm still sitting at the ddb prompt at the moment. Thanks, Jeff ps. the hardware is a basic desktop w/750MHz AMD duron, 256MB ram, 30GB IDE disk, realtek ethernet, ide cdrom and not much else... - openssl: generating new isakmpd RSA key... Data modified on freelist: word 3 of object 0xd0c0e0f0 size 0x10 previous type UVM amap (0xdea9beef != 0xdeadbeef) Data modified on freelist: word 0 of object 0xd0c0f440 size 0x10 previous type free (0xe388a3c0 != 0xdeadbeef) panic: amap_wipeout: corrupt amap Stopped at Debugger+0x4: leave ddb trace Debugger(d607d210,d607d210,e389ce90,d60b1e38,0) at Debugger+0x4 panic(d0531f67,d60bd604,0,d60bd604,d60bd604) at panic+0x63 amap_wipeout(d60b1e38,0,1,0) at amap_wipeout+0x6b uvm_unmap_detach(d6073cb8,0,0,d6073cb8) at uvm_unmap_detach+0x92 sys_munmap(d60b8b44,e389cf68,e389cf58,1000,5a0) at sys_munmap+0x12c syscall() at syscall+0x2ea --- syscall (number 73) --- 0xdd4b945: ddb ps PID PPID PGRP UID S FLAGS WAIT COMMAND * 3779 18002 180020 7 0x4006 openssl 2713526816 26815 83 30x184 poll ntpd 26815 1268150 3 0x84 poll ntpd 17352 20993 20993 733 0x184 poll syslogd 20993 1 20993 0 3 0x84 netio syslogd 18002 1 18002 0 3 0x4086 pause sh 13 00 0 3 0x100204 crypto_wa crypto 12 00 030x100204 aiodoned aiodoned 11 00 030x100204 syncer update 10 00 030x100204 cleaner cleaner 9 00 030x100204 reaper reaper 8 00 030x100204 pgdaemon pagedaemon 7 00 030x100204 pftm pfpurge 6 00 030x100204 usbevt usb1 5 00 030x100204 usbtsk usbtask 4 00 030x100204 usbevt usb0 3 00 030x100204 apmev apm0 2 00 030x100204 kmalloc kmthread 1 01 030x4084 wait init 0 -10 03 0x80204 scheduler swapper ddb
Re: Kernel panic in openssl on fresh minimal install
This sounds like bad hardware to me. Have you tried installing your CD set elsewhere? Ok, so I made a memtest86 boot cd to test out this machine. At the moment it is 59% of the way through the testing and it's only found 7,016 memory errors, that's not that bad, right? :-) So bad hardware it is... not to get off on a tangent, but does anybody have any good reading links (or care to comment here) on why a BSD system would be more sensitive to bad RAM than a Windows server? I'm having a hard time thinking of Windows as more fault-tolerant than any OS, but it's pretty hard to deny the sudden change in crash frequency. Btw, I'm now 69% of the way through the testing and I've got 303,xxx memory errors. They are racking up so fast now I can't even read the number. I'm literally amazed that this box would run anything. Thanks to all who responded for the helpful prod, I (obviously) wouldn't have jumped straight to this conclusion without some nudging. Jeff
