From: Mark Daniel [Mark.Daniel@wasd.vsm.com.au]
Sent: Wednesday, September 05, 2001 8:19 PM
To: Info-VAX@Mvb.Saic.Com
Subject: Re: Alpha/VMS V Sun/Solaris

I made the configuration changes suggested by Dave ('fileflags' to 3)
and here are the results (couldn't be bothered activating CSWS). 
Curiously, I remember when the original results were generated OSU
*seemed* low I checked the configuration files a couple of times and
that multi-threading was enabled for the image.  These looked OK to me
(and apparently were).  If such a seemingly anomalous result crops up
again alarm bells might ring and I'll bounce it off whomever.  The
original results were generated with VMS 7.2-1, these with 7.3 (same
TCP/IP release, 5.1), which may account for some variation between the
two sets.  As emphasized in the WASD doco these results are *indicative*
only.

 C A C H E   D I S A B L E D        C A C H E   E N A B L E D
     requests/second                    requests/second
 Response Concurrent WASD OSU      Response Concurrent WASD OSU
 0K       1           88   60      0K       1          283  213
 0K       10         111   67      0K       10         441  278
 64K      1           14   14      64K      1           82   34
 64K      10          33   23      64K      10          98   41

Just by way of comparison with the previous results I reverted the
'fileflags' parameter to 7 and obtained the following.

   C A C H E   E N A B L E D
       requests/second
 Response Concurrent WASD OSU
 0K       1          270   56
 0K       10         423   82
 64K      1           78   21
 64K      10          95   27

These confirm Dave's analysis regarding the effect/circumvention of the
bug.

David Jones wrote:
> 
> Mark Daniel wrote:
> >
> > Can't resist ...
> >
> >   http://wasd.vsm.com.au/ht_root/doc/htd/htd_1800.html
> 
> The first 2 tables on that web page show:
> 
>     C A C H E   D I S A B L E D            C A C H E   E N A B L E D
>         requests/second                        request/second
> Response Concurrent WASD OSU Apache     Response Concurrent WASD OSU Apache
> 0K       1          58   41  14         0K       1          301  46  15
> 0K       10         89   47  42         0K       10         382  64  21
> 64K      1          39   14  18         64K      1          83   18  13
> 64K      10         43   16  39         64K      10         98   21  31
> 
> Try running the OSU test with fileflags set 2 it's pre-3.9 of 3 instead of
> 7.  When bit 2 is set, the cache refresh logic includes a test against
> the file's expiration date field so that 'expired' files are given an
> effective refresh time of zero.  A cache entry refresh is nearly as expensive
> as reading the file since the file has to be accessed to check for a
> changed modification date.  Due to a bug in the code, files with no expiration
> date are treated as expired so you don't see much improvement from the
> cache.  Keeping file flags at 3 skips the code path containg the bug so you get
> full performance of the cache.
> 
> David L. Jones               |      Phone:    (614) 292-6929
> Ohio State University        |      Internet:
> 140 W. 19th St. Rm. 231a     |               jonesd@er6s1.eng.ohio-state.edu
> Columbus, OH 43210           |               vman+@osu.edu
> 
> Disclaimer: I'm looking for marbles all day long.

-- 
Non sinere illegitamus carborundum.