1 |
I was just following the conversation but I am convinced now. |
2 |
For the record, I am running two "pretty regular" web-oriented servers: |
3 |
web, db, email, lists, dns, some low band streaming, no fileserving and |
4 |
no nat or any type of routing. |
5 |
|
6 |
I've used nptl before, but I gave up because I had some problems running |
7 |
turck-mmcache on a nptl-only apache. |
8 |
Since mmcache is obsolete and not used on the server anymore I see no |
9 |
problem switching back to a nptl-only environment. |
10 |
|
11 |
Although I think I know all the steps I am supposed to do, I would |
12 |
really appreciate it if some would point out the "migration" steps, just |
13 |
in case I don't miss some important stage. |
14 |
|
15 |
Thank you guys. |
16 |
|
17 |
-----Original Message----- |
18 |
From: J. Ryan Earl [mailto:ryan@×××××××××××××××.com] |
19 |
Sent: 10 august 2005 18:44 |
20 |
To: gentoo-server@l.g.o |
21 |
Subject: Re: [gentoo-server] nptl |
22 |
|
23 |
|
24 |
Billy Holmes wrote: |
25 |
|
26 |
> This makes it easier and faster for the kernel to context switch |
27 |
> between processes (and threads within a process). |
28 |
> |
29 |
> A prime example of this is a test from the LKML, where Ulrich Drepper |
30 |
> wrote he was able to start and stop 100,000 threads in about 2 |
31 |
> seconds. In further messages, Ingo Molnar confirms this and also |
32 |
> points out that before these changes, it would have taken upwards to |
33 |
> 15 minutes for the same thing. |
34 |
> |
35 |
> http://kerneltrap.org/node/422 |
36 |
|
37 |
Absolutely true. NPTL is drastically more scalable than linuxthreads. |
38 |
For instance, spawn 1000 worker threads in a JVM and the scheduling |
39 |
overhead alone will cause the system to become unresponsive across the |
40 |
board. Use NPTL and you can get 10000 threads all running no problem. |
41 |
|
42 |
-ryan |
43 |
-- |
44 |
gentoo-server@g.o mailing list |
45 |
|
46 |
|
47 |
-- |
48 |
gentoo-server@g.o mailing list |