Coda File System

Re: Coda Stress Test

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Fri, 19 Apr 2002 12:37:32 -0400
On Thu, Apr 18, 2002 at 02:48:43PM -0400, Casey Helfrich wrote:
> Mike wrote a script that generates 4GB of data (all zeros right now),
> split up into 125kb files.  The script then tried to dump it all into
> Coda.  Right now I have a 6GB hard limit on in the /vicepa partition,
> and cmon shows that it only used 57% of the available space before it
> crashed.

So you you must have written about 30000 files at that point. It doesn't
sound like that really should be much of a problem. How much RVM data
did you configure for this server? Even if every file would take about
2KB of RVM (don't really know the size, just a random approximation),
this should still take only 60MB.

> Below is what I hope will be someuseful information to you.  As you can

Not really useful, RVM is a bit of a black box. This just tells me that
'allocation failed'. Don't know whether this was due to a problem with
the RVM log file, the RDS allocator, or some bug that corrupted memory.

If you apply the attached patch to coda-srv/util/rvmlib.c we can at
least get some info, the returned error code, how much we tried to
allocate and who tried to allocate.

Jan

--- rvmlib.c.orig	Fri Apr 19 12:32:28 2002
+++ rvmlib.c	Fri Apr 19 12:35:59 2002
@@ -207,8 +207,11 @@
 		err = 0;
 		p = rds_malloc(size, _rvm_data->tid, &err);
 		
-		if (err != 0) 
+		if (err != 0) {
+			fprintf(stderr, "rvm_malloc: error %d size %lx file %s line %d\n",
+					err, size, file, line);
 			RVMLIB_ASSERT("error in rvmlib_malloc\n");
+		}
 		RDS_LOG("rdstrace: rec_malloc addr %x size %lx file %s line %d\n",
 			p, size, file, line);
 		return p;
Received on 2002-04-19 12:39:19