[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
Re: Total Knowledge вопросы для оценки
I already have proxying implemented - it is base part
of the design. What I need to add is load balancing.
As for comparing tntnet to apache... well we all know apache
is slow. We all know that serving static pages with multi-process
(or multithreaded) web server is low-tech these days. Not to
mention that "simple dynamic page" doesn't need to be read
from hard drive, unlike static pages, which affects results. If
you want to get real performance comparisons, try few
different web servers. Try lighttpd. Try thhtpd.
Tommi Mäkitalo wrote:
Am Samstag, 20. Mai 2006 04:13 schrieb Ilya A. Volynets-Evenbakh:
By the way, one thing in terms of clustering that would
be fairly easy to implement in CPPServ is balancing inside
of web server front end.
If web server module tracks sessions and makes sure to
direct requests that belong to same session to same CPPServ
instance, there is no need to implement session migration,
and at the same time it gives sufficient distribution of load
when service usage is so high that such distribution is really
needed.
This does require single webserver to be point of entry into
cluster, but if web serving is the only thing it does, it should
not be a problem - it will either serve static pages, or will
simply forward data between client and CPPServ.
I don't know how much you are aware about my Tntnet (http://www.tntnet.org),
but it might help you with this proxying-thing. Tntnet has a very fast and
scalable http-engine. I have compared Tntnet with Apache2. The result is,
that simple dynamic pages are about 20% faster than static pages with
Apache2. Tested with enabled logging, keep-alive and 3 concurrent connections
with ab2.
Proxying shouldn't be too hard to implement and session-tracking can be
realized with it. Although Tntnet is a alternative to CPPSERV, but for those
who prefer your more servlet-style API, it might help.
Tommi