Re: about the advantages of caching metainformation...

On Sat, 4 Mar 2000, Niklas Wiberg wrote:

> Why are the 'Total bytes' numbers for Jigsaw reported negative?

Just an overflow, the real result is 6537455 * 392 = 2562682360
= 21005593 b/s

> /Niklas Wiberg
> 
> 
> Yves Lafon wrote:
> > 
> > I just installed apache 1.3.12 to do some basic tests.
> > 
> > The configuration is the following:
> > 1/ Apache - compiled using gcc2.95.2
> > 
> >    1.3.12, bare configuration, with just two modification:
> >    Port 8000, and ContentDigest on, to give the Content-MD5
> >    header.
> > 
> >    HEAD /Distrib/jigsaw_2.0.4.zip HTTP/1.1
> >    Host: tarantula.inria.fr:8000
> > 
> >    HTTP/1.1 200 OK
> >    Date: Fri, 03 Mar 2000 13:49:12 GMT
> >    Server: Apache/1.3.12 (Unix)
> >    Last-Modified: Fri, 21 Jan 2000 17:21:36 GMT
> >    ETag: "1fd26-63c0ef-388895a0"
> >    Accept-Ranges: bytes
> >    Content-Length: 6537455
> >    Content-MD5: BnTW0ZfYueUDCY1U67jtdg==
> >    Content-Type: application/zip
> > 
> > 2/ Jigsaw 2.1.0
> > 
> >    Running using jdk1.2.2_05.
> >    The configuration is not fair for Jigsaw, it has virtual hosts
> >    3 servers running (including one experimental 2.1 cache)...
> > 
> >    HEAD /Distrib/jigsaw_2.0.4.zip HTTP/1.1
> >    Host: tarantula.inria.fr:8007
> > 
> >    HTTP/1.1 200 OK
> >    Cache-Control: max-age=432000
> >    Date: Fri, 03 Mar 2000 13:48:34 GMT
> >    Content-Length: 6537455
> >    Content-Md5: BnTW0ZfYueUDCY1U67jtdg==
> >    Content-Type: application/zip
> >    Etag: "1vtq0oq:rjaogu80"
> >    Expires: Wed, 08 Mar 2000 13:48:34 GMT
> >    Last-Modified: Fri, 03 Mar 2000 13:24:38 GMT
> >    Server: Jigsaw/2.1.0
> > 
> > The test client is ptester, from the phttpd distribution, still 1.0 but
> > the test is not about HTTP/1.0 / 1.1 but about caching meta information ;)
> > So, we are asking as many time as possible the /Distrib/jigsaw_2.0.4.zip
> > file, the test runs for 2 minutes, with 5 concurrent clients.
> > Here are the results:
> > 
> > 1: Apache
> > ptester -t120 -h138.96.10.3 -p8000 /Distrib/jigsaw_2.0.4.zip
> >     /Distrib/jigsaw_2.0.4.zip /Distrib/jigsaw_2.0.4.zip
> >     /Distrib/jigsaw_2.0.4.zip /Distrib/jigsaw_2.0.4.zip
> > 
> > URL                  NRq  Min Ct Avg Ct Max Ct  Min Tx Avg Tx Max Tx
> > Bytes
> > /Distrib/jigsaw_2.0.4.zip   15  0.0002 0.0003 0.0006  7.2549 8.2440 9.4845 6537754
> > /Distrib/jigsaw_2.0.4.zip   15  0.0002 0.0003 0.0009  6.8569 8.2588 9.2060 6537754
> > /Distrib/jigsaw_2.0.4.zip   15  0.0002 0.0003 0.0017  7.2966 8.2644 9.5115 6537754
> > /Distrib/jigsaw_2.0.4.zip   15  0.0002 0.0002 0.0006  7.5706 8.2792 9.8478 6537754
> > /Distrib/jigsaw_2.0.4.zip   16  0.0002 0.0003 0.0006  4.7803 7.8056 8.7447 6537754
> > 
> > Actual test time.. 125 seconds
> > Total requests.... 76 (0 requests/sec)
> > Total failed...... 0 (0 requests/sec)
> > Total bytes....... 496869304 (3974954 bytes/sec)
> > 
> > Min Tx: 4.7803
> > Max Tx: 9.8478
> > 
> > 2: Jigsaw
> > ptester -t120 -h138.96.10.3 -p8007 /Distrib/jigsaw_2.0.4.zip
> >     /Distrib/jigsaw_2.0.4.zip /Distrib/jigsaw_2.0.4.zip
> >     /Distrib/jigsaw_2.0.4.zip /Distrib/jigsaw_2.0.4.zip
> > 
> > URL                  NRq  Min Ct Avg Ct Max Ct  Min Tx Avg Tx Max Tx
> > Bytes
> > /Distrib/jigsaw_2.0.4.zip   74  0.0002 0.0002 0.0007  0.7829 1.6449 2.8596 6537740
> > /Distrib/jigsaw_2.0.4.zip   81  0.0002 0.0002 0.0010  0.3938 1.5040 2.3876 6537740
> > /Distrib/jigsaw_2.0.4.zip   76  0.0002 0.0002 0.0007  0.6084 1.6051 2.3686 6537740
> > /Distrib/jigsaw_2.0.4.zip   80  0.0002 0.0002 0.0007  0.3588 1.5259 2.5286 6537740
> > /Distrib/jigsaw_2.0.4.zip   81  0.0002 0.0002 0.0007  0.4944 1.5087 2.7251 6537740
> > 
> > Actual test time.. 122 seconds
> > Total requests.... 392 (3 requests/sec)
> > Total failed...... 0 (0 requests/sec)
> > Total bytes....... -1732173216 (-14198141 bytes/sec)
> > 
> > Of course the test was run on the same machine as bandwidth was a
> > limitation (still 10Bt).
> > Why is Apache so slow? because it computes for every request the
> > content-md5 of the file before sending the reply.
> > It is just a proof that you can adapt easily a bench to be good to one
> > particular server (Hey, Jigsaw is 5 time faster than Apache! ;) ).
> > 
> >       /\          - Yves Lafon - World Wide Web Consortium -
> >   /\ /  \        Architecture Domain - Jigsaw Activity Leader
> >  /  \    \/\
> > /    \   /  \   http://www.w3.org/People/Lafon - ylafon@w3.org
> 
> 

      /\          - Yves Lafon - World Wide Web Consortium - 
  /\ /  \        Architecture Domain - Jigsaw Activity Leader
 /  \    \/\    
/    \   /  \   http://www.w3.org/People/Lafon - ylafon@w3.org

Received on Saturday, 4 March 2000 08:38:22 UTC