John,
That is pretty much the same conclusion that I came to as well
which is why Dave asked. The timing is actually at least twice
that because of the way that the manifest is sent to the client
which makes the timing more like 1/3 of a second for Apache/CherryPy
verses little over a second for Apache/cgi-bin. But when one
considers that an install will take several order magnitude longer
then this difference plus only being done once that performance
hit is insignificant. Coupled with the fact that using Apache/cgi-bin
will be less complex and we can improve on the client/server
handshake I think that the design needs to be re-worked.
Expect a new draft next week based on these findings.
Thanks,
John
On 04/ 9/10 05:43 PM, John Plocher wrote:
So, this says that the range is between a couple of tenth's of a
second (0.14-0.16s) for py and half to one second (0.58 - 1s) for cgi.
This is the AutoInstall web server. Presumably, installs are
relatively infrequent and take a relatively long time to complete. In
that context, it seems unlikely that the performance of the web server
will dominate the total performance footprint of the entire use-case.
Does a half second or a second improvement here really matter? Is a
the cgi-vs-py response time difference meaningful when looking at the
total time used by- *and* the frequency of invocation of- the entire
AI sequence? [*] The derivative question in my mind is whether this
is a premature optimization and whether the added complexity of the
web infrastructure and development environment is worth the fraction
of a second improvement.
-John
[*] I don't have a clear understanding of the expected AI webserver
usage - will an "average install" hit the web server millions of times
- or only dozens? Obviously the answer will dictate the weighting
applied to this analysis...
On 04/ 9/10 02:23 PM, John Fischer wrote:
Dave,
apache/mod_wsgi/cherrypy 1, 100, 1000
real 0m0.152s
real 0m14.515s
real 2m34.528s
apache/cgi-bin 1, 100, 1000
real 0m1.065s
real 0m57.964s
real 9m30.514s
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
