Coda File System

Re: Metadata partition size

From: Frank Van Damme <>
Date: Mon, 8 Sep 2003 17:53:23 +0200
On Monday 08 September 2003 16:23, Jan Harkes wrote:
> On Mon, Sep 08, 2003 at 01:30:39PM +0200, Frank Van Damme wrote:
> > On Sunday 07 September 2003 23:41, Jan Harkes wrote:
> > > Ok, cool. Yeah the whole RVM data partition is loaded (mmapped) into
> > > virtual memory. Which makes the server pretty swap heavy.
> >
> > Why is this needed? Users need twice the amount of disk space the
> > metadata actually takes.
> Just the way RVM works. The data partition contains the last known
> consistent state. The application only modifies the in-memory copy and
> logs it's actions to the RVM log. Once in a while the logged
> transactions are committed to the underlying RVM data file.

Not too convenient from an end user perspective imho. Why not just write the 
metadata to disk and still use a log (like a journaled file system)?

> Private mmaps helps a bit because unmodified pages don't have to be
> written to swap under memory pressure as they are reloaded from RVM
> data.
> > No luck...
> Then I don't really know what's wrong. If the auth2 daemon is running we
> should at least be able to get a binding even if the password is
> incorrect. Maybe strace/tcpdump can give a clue where clog is trying to
> send it's packets to.

Strace ends with: 

time(NULL)                              = 1063036281
time(NULL)                              = 1063036281
time(NULL)                              = 1063036281
write(2, "Invalid login (RPC2_NOBINDING (F"..., 36Invalid login 
) = 36
munmap(0x40014000, 4096)                = 0
semget(1, 4096, IPC_CREAT|IPC_EXCL|IPC_NOWAIT|0x401680c0|0200) = -1 ENOSYS 
(Function not implemented)

But maybe I just ill-configured my realm name or something... can I modify my 
realm name somewhere?

Frank Van Damme 
~~~~~~~~~~~~~~~    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
"Je pense, donc je suis breveté."
Received on 2003-09-08 11:57:40