Re: Linux-1.3.67 - NFS caching

Joshua M. Thompson (invid@optera.com)
Tue, 20 Feb 1996 23:40:07 -0500


>I just made a 1.3.67 available, which fixes some of the problems people
>have had (the old SLIP and IP masquerading patches). It also does

Well...I upgraded three systems to 1.3.67 today:

1. A 486DX/2-80, 32 MB RAM, 1 GB Quantum Fireball 1080 SCSI on Adaptec 1542
controller. Also has a Cyclades 8-port card. This machine is used for shell
access for our users (what few shell accounts we have), runs our web server
(Apache HTTPD 1.0.2), and is our primary DNS server. It also runs the ISDN
link to my house, using pppd 2.2.0e over an ISDN TA connected to the
Cyclades card. A 3com 3C509 connects this machine to our Ethernet.

2. A 486DX/2-66, 16 MB RAM, 1 GB Quantum Fireball 1080 SCSI and 4.2 GB Quantum
Grand Prix SCSI on an Adaptec 1542 controller. Also uses a 3Com 3C509.

3. My personal machine at home. A Triton-based P100 with 16 MB RAM, plus several
Quantum SCSI drives ranging from 1 GB to 160 MB and connected to an Adaptec
2940 SCSI controller. My network card is a 3Com 3C590. (no, my PPP link does
not go to this machine; I have a MicroRouter 900i doing that job).

Machine #2 is still running great. It seems more responsive too.

Machine #3 is also running great. It seems about the same speed, though I
noticed that all my TCP sockets are closing correctly now. :-)

Machine #1 lasted about 10 minutes before I had to hit the reset switch. It
started out OK, but performance quickly went to hell. The load average was
well below 1.0 at all times, and there were only 5 or so sessions open
(three of me and a couple users), but the machine was actually PAUSING for
brief periods of time and then continuing. My personal PPP link that runs
off a Cyclades card on this machine was losing approximately 95% of its
incoming packets. While I was
still able to do anything I ran top, and saw that most of the processes on the
machine were running with negative priorities, though none seemed to be tying
up any signifigant amount of CPU time.

I suspect something on machine #1 was masking interrupts for too long...was
there a recent change to the Cyclades driver? If I could get this darn
machine to work I could give the NFS code a good testing since this machine
does NFS mount several directories from our SCO server (including user home
directories).
For now this machine is back on 1.3.64 and running fine.

--
email: invid@optera.com                           Senior Systems Engineer
 web: http://www.optera.com/~invid        mich.com, Inc.  Farmington, MI
    PGP fingerprint: 90 05 50 DA 0E 18 CD 47  C5 25 83 1F A5 41 17 73
              I do not speak for mich.com from this account.