I'm in the process of bringing up a Sparc 20, that is configured with
256 MB of RAM, a Ross Quad/72 (4 72 Mhz cpu), and Solaris 2.3 (with lots
of bug patches- ~70 or so!) that is expected to support at least 300 
simultaneous logins as an AFS client. This machine has an AFS disk cache of 
430 MB, though the cache size entry I'm using is 390,000K (.9 x 430 MB),
as recommended by the manual.

Does anyone have any recommendations about what afsd flags to use? 
Specifically, I'm interested in what values, if any, to use for the  
-dcache, -volumes, and -chunksize options.

The default "LARGE" option in the /etc/init.d/afs.rc file is:

LARGE="-stat 2800 -dcache 2400 -daemons 5 -volumes 128"

Since the -chunksize option is omitted, the default of 64K is used. Any
recommendations here?

According to the Command Reference Manual for AFS 3.3, the default number of
V files in a disk cache is the greater of one of the following:

        - 100
        - cachesize/chunksize * 1.5
                (390,000/64 * 1.5 = 9,140)
        - cachesize/10,000 
                (390,000/10,000 = 39)

Since we've got a rather large cache partition (at least I think so!) of 
430 MB, 128 V files seems too small to me if I opt for  the LARGE option
above, while 9,140 seems too large if I go with the default by omitting a
-volumes flag. If we go with the 9,140 and each V file contains 64K, this 
would exceed the 430MB of available disk space by 141 MB!

Any advice about this will be much appreciated. Thanks in advance.

Sincerely,
Steven McElwee

 --
 -----------------------------------------------------------------------------
 Steven McElwee          |         Email -->      |  [EMAIL PROTECTED]  
 Academic Computing      |   <-- US Snail Mail    |  
 Duke University         |------------------------|---------------------------
 401 North Building      |  (919) 660-6914 (Work) |  (919) 684-8651 (Fax)
 Durham, NC 27706        |                        |  (919) 612-1955 (Cellular)
 -----------------------------------------------------------------------------





Reply via email to