Mathopd will have higher latency as an image server, because it is
single threaded and doesn't do asynchronous disk I/O. So on a
high-end server, the number of IOs/sec it can generate is limited. AS
does not have this problem since it is multi-threaded.
However, mathopd services requests in a very "fair" manner: in an ab
test, the difference between minimum request time and maximum request
time is very small. On AS, this is not the case, and there are
relatively large variances in request service time, indicating that
some threads are being starved. This could be an AS design issue or
an OS scheduling/resource issue - dunno.
Mathopd is definitely the faster engine, but its design limits its
capacity. Mathopd with asynchronous disk I/O would be awesome.
Mathopd:
--------
Concurrency Level: 15
Time taken for tests: 0.976 seconds
Complete requests: 10000
Failed requests: 0
Total transferred: 2100210 bytes
HTML transferred: 450045 bytes
Requests per second: 10245.90
Transfer rate: 2151.85 kb/s received
Connnection Times (ms)
min avg max
Connect: 0 0 1
Processing: 0 0 1
Total: 0 0 2
AOLServer 3.4:
-------------
Concurrency Level: 15
Time taken for tests: 1.822 seconds
Complete requests: 10000
Failed requests: 0
Total transferred: 2550000 bytes
HTML transferred: 450000 bytes
Requests per second: 5488.47
Transfer rate: 1399.56 kb/s received
Connnection Times (ms)
min avg max
Connect: 0 0 2
Processing: 0 1 210
Total: 0 1 212
>
> On Thu, Jan 08, 2004 at 09:15:49PM -0500, Lamar Owen wrote:
> > There is a good, if inconspicuous, mention of AOLserver in the Feb LJ. On
> > page 46, in the feature on the Magnatune record label, the statement is made:
> > "Apache 2 running PHP and OpenSSL serves all the HTML pages. When Magnatune
> > was Slashdotted, I found that Apache could not keep up with the load for
>
> > images. All HTTP image requests now are off-loaded to AOLserver, which had
> > the lowest latency to serve images at high speeds." Later: "Mathopd [which
> > they use to serve the very large streaming audio files] has more latency than
> > AOLserver, which is why we don't use it to serve small images."
>
> Why would anyone care about the "latency" of serving small images?
> Last I heard a human being viewing images in a browser is not exactly
> senstive to small latencies the way a parellel MPI program might be,
> after all. Or are they talking about absurdly large latency
> differences between Mathopd and AOLserver, like several seconds or
> more?
>
> --
> Andrew Piskorski <[EMAIL PROTECTED]>
> http://www.piskorski.com/
> --
> AOLserver - http://www.aolserver.com/
--
AOLserver - http://www.aolserver.com/
To Remove yourself from this list, simply send an email to <[EMAIL PROTECTED]> with the
body of "SIGNOFF AOLSERVER" in the email message. You can leave the Subject: field of
your email blank.
