call for testers

One of the items that has been a major performance deficiency in the
Windows AFS client has been the high use of cpu time in order to
determine that an entry does not exist within a directory.  This cost is
a side effect of the design of the AFS3 directory format which
incorporates a case sensitive hash table.  As the Windows platform is
case-insensitive, if the hash table lookup fails to find the requested
file, it becomes necessary to perform a linear search of the entire
directory in order to determine if there is a single case-insensitive
match, an ambiguous match, or no match at all.

The high cost of these lookups is a killer for applications that make
use of search paths that include directories within AFS.  The Java
CLASSPATH, INCLUDE, LIB, and PATH search paths are typical examples.

The most recent daily builds include a new B+ tree implementation.  The
Windows AFS client builds a case insensitive B+ tree for each directory
out of the contents of the AFS3 directory data.  The B+ tree is then
used for all lookups.  The result is significant reduction in search time. 

The daily builds can be obtained from

\\afs\athena.mit.edu\user\j\a\jaltman\Public\OpenAFS\

Any build 1-5-20070822 or later will do.

Please test and report any bugs you experience.

Thanks.

Jeffrey Altman
Secure Endpoints Inc.