Coda File System

Re: RVM_INIT error: Current time before last recorded - check kernel date

From: Ben Gertzfield <che_at_debian.org>
Date: 01 Dec 1998 11:25:28 -0800
>>>>> "jaharkes" == jaharkes  <jaharkes_at_cs.cmu.edu> writes:

    jaharkes> Hi Ben, RVM uses (probably amongst others) timestamps to
    jaharkes> guarantee consistency, and is therefore slightly picky
    jaharkes> about what time the gettimeofday call returns.

Ah, okay. :)

    jaharkes> In your case, as you've just rebooted into a new kernel,
    jaharkes> the time is taken from the CMOS clock, and it must have
    jaharkes> been a couple of minutes behind your system clock. We
    jaharkes> keep all servers (and clients) in sync by using xntpd,
    jaharkes> and we write the current system time to the cmos during
    jaharkes> system shutdown to avoid problems.

This system is xntpd'd as well. Odd..

    jaharkes> btw. about the lack of debug symbols, I've checked and
    jaharkes> the binaries were compiled with debugging, but somehow
    jaharkes> got stripped before being packaged up.

Oh! That'd do it. I guess I should switch to the official Debian
packages..

    Ben> I still think the "check kernel date" error in the subject
    Ben> above is pretty vague, though.

    jaharkes> Yeah, it might be possible to turn it into something
    jaharkes> like

    jaharkes> "RVM_Init: detected an anomaly in the space-time
    jaharkes> continuum, last RVM operation was Mon Nov 30 1998,
    jaharkes> 18:00:00, current time is Mon Nov 30 1998, 17:56:00."

    jaharkes> However that might still be confusing, for non-trekkies.

I have no problem with it ;)

    Ben> Also, in an unrelated question, is it usual for codasrv to be
    Ben> taking up 116 megs of ram? :)

    jaharkes> The RVM data partition is copied into the server process
    jaharkes> memory image. It is the way RVM is implemented. We might
    jaharkes> be able to use mmap to avoid using both disk- and
    jaharkes> swap-space, but it is not yet possible to memory map a
    jaharkes> raw partition on linux. And we would have to make doubly
    jaharkes> sure it gives the same consistency guarantees. How many
    jaharkes> programs can you `kill-9' and restart without losing
    jaharkes> information?

Hmmmm. Once it is possible to mmap a raw partition under Linux, it
would be extremely useful to make this an option; the ability to run
an NFS server on a box with not-so-much RAM is nice, and being able to
run Coda on a box with 32 or 64 megs of RAM would be nice.

    jaharkes> As not all parts of the RVM mapping are used very
    jaharkes> intensively, most of the memory occupied by RVM will get
    jaharkes> swapped out.

Oh, then that's not so bad. I'm glad I've got a lot of swap. :)

Ben

-- 
Brought to you by the letters E and B and the number 2.
"Step away from the car. This car is protected by Viper." -- TMBG
Debian GNU/Linux -- where do you want to go tomorrow? http://www.debian.org/
I'm on FurryMUCK as Che, and EFNet and YiffNet IRC as Che_Fox.
Received on 1998-12-01 14:59:00