Messages in this thread | | | From | "Linux-Kernel" <> | Subject | Re: Finding mysterious 2.0.33 crashes -- Please help | Date | Mon, 16 Feb 1998 12:56:30 -0500 |
| |
--------- > From: Jos Vos <jos@xos.nl> > To: sumner@COLLEGIUM.ADSL.NET.CMU.EDU > Cc: linux-kernel@vger.rutgers.edu > Subject: Re: Finding mysterious 2.0.33 crashes -- Please help > Date: Monday, February 16, 1998 9:50 AM > > G. Sumner Hayes wrote: > > > Has anyone had mysterious hangups in 2.0.33 (total hang, no messages in > > logs, no oops) who is _not_ using an Adaptec SCSI driver? It could help > > narrow the range of code to be checked if that's a common factor in all > > these cases. Also, has anyone had this problem who is _not_ using an > > NE2000 ethernet card? Who is _not_ using a PS/2 mouse? Who is _not_ > > using glibc? > > Yes to all... We have no Adaptec, no NE2000, no PS/2 mouse, and no > glibc > and we have mysterious system halts without _any_ messages very > frequently > (one or two times a week, sometimes 2 times in 2 hours).
[SNIP]
I have 4 machines that each had over 200 days of uptime with 2.0.27 before I upgraded. _Rock Solid_: In three years, I haven't gotten that uptime on any of my cisco 2501's. Way to go Linux. After upgrading to 2.0.33 (because of overlap.c and syn attacks (and others?), someone hit several of our IP address ranges continuously for a few days), I'm averaging less than 2 days of uptime. They machines simply freeze with nothing printed on the screen (newbie-hint type "setterm -blank 0" to keep the screen from blanking), but the harddrives are always running (LED on) when it happens. I'm running remote syslog, and it's not logging anything strange before the crashes. I've done continuous dumps of the serial line between the modem and the serial port (w/ a homemade cable), and there is no strange traffic received from the Internet, so it's not a new type of attack. No sk_buff or "Couldn't get free page" errors are logged, and the machines are only running ~100 processes at peak (so there's probably enough RAM). The peak load is only ~20.0, and it usually stays below 2.5. The machines run pppd (stops then restarts once per day), named ('ndc stats' shows about 10,000 queries per day), DHCPD (serving 30-200 machines that reboot 4 times per day on average (they're Windows :> )), Apache (gets around 10-100 hits per day), Squid (about 30,000 cache hits per day on the heaviest used server, wow), IP Masquerading (masq'd traffic is about 45 Mbytes per day), long list of IP firewall rules (~140 lines, we have to protect those NT machines :> ), IP transparent proxy for four ports (minimal traffic), sshd (for secure port forwarding, makes/breaks a connection about once per hour, has about 1 Kbyte of traffic per hour), xntpd (chimes off of three servers), gated (does OSPF routing using multicasting, the routes change about twice per day when interfaces go up/down and the routing table is ~75 lines long), ftpd (two connections per day to copy big log files), rshd (two connections per day to run ipfwadm to collect stats), crond (wc /etc/crontab is ~220 lines long), rarp, tftp (used to boot several diskless Sun workstations and serve config files for a router, used about once per week), and a few other small servers. None of the crashes coincide with the automated tasks. Each of them routes between the two NE2000 cards and ppp over a modem, and it Masqs only over ppp. They do not run X or any other user applications. I'm running stock Redhat 4.1 with no upgrades except for the kernel (absolutely no glibc, no ecgs, and no gcc 2.8.0). Each machine has a no-name 486 motherboard (4 different motherboards in the 4 systems, but all Intel 486DX/33 processors), 16 Mbytes of RAM (all FPM w/ parity), 2 NE2000 clone ISA cards in each machine (6 different brands, 2 PCI and 6 ISA), internal USR Sportster modems (traded one out for an external to do monitoring, and it didn't affect the problem), and 500 Mbyte Maxtor IDE harddrives connected to four different brands of ISA IDE controllers (yeah, mish-mash junk). I'm running no modules, no SCSI, no Adaptec, no PS/2 mouse, no NFS, no ipx, no smb, no appletalk, no CDROM, no ISDN, and no X. The uptime was especially impressive considering that it runs 75-100 Mbytes into swap while running Squid with only 16 Mbytes of RAM.
I'm in a bit of a lurch, because my users (and the owner of the company!) have come to depend on the Linux machines over the past three years. And, I can't downgrade, because we're still seeing some attacks. The machines are at remote locations (South Africa (4 hour drive from the nearest airport!), France (20 miles from an airport, but still a 4 hour drive at times :> ), Mexico, and one in Alabama, US) so upgrading, managing, testing, and admining is a pain. I spent $350 on my phone bill last month talking someone through upgrading the kernel. I need my 200 day uptime back.
Thank You
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu
| |