Wojciech Puchar wrote:
>>Converting Squid to a forking model is not feasible. If you want such a
>>
>
> each process per connection is terrible - tried oops on linux once.
> it's easy to reach 4090 processes - linux maximum.
He's referring to a forking model...not a threading model. They are
different. Threads share memory and everything else with their parent,
while a forked process does not. Processes are independent once they
have been spawned and have no knowledge of the other processes unless it
is explicitly communicated via shared memory or some other IPC. Oops is
a threaded server, not a forking server. Apache is a pre-forking server.
> threads are good with OS in which they are well implemented (not linux)
Actually threads are quite good under Linux in 2.4 (and not bad under
2.2). I've been able to run Oops at about the same speed on Linux 2.4
as on FreeBSD (which is one of the development platforms for
Oops)--though still not as fast as a well-tuned Squid. Yes, threads on
Linux are still technically more like 'light weight processes' than true
threads...so a bit heavier than on some OSs. But they are actually
lower overhead than threads under Solaris and some of the other
proprietary Unices--at least that's what I've been able to find out in
my research of the subject. Perhaps you've seen some data to suggest
otherwise (if so, links are helpful). Anyway, threads in Linux are not
bad at all, and I'm not sure where you would have gotten the idea that
they are.
--
Joe Cooper <joe@swelltech.com>
Affordable Web Caching Proxy Appliances
http://www.swelltech.com
Received on Fri Aug 17 2001 - 18:39:26 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:01:43 MST