Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20120211141042.GA19696@openwall.com>
Date: Sat, 11 Feb 2012 18:10:42 +0400
From: Solar Designer <solar@...nwall.com>
To: owl-dev@...ts.openwall.com
Subject: Re: -Wl,-z,now

On Thu, Feb 09, 2012 at 06:46:10PM +0400, Vasiliy Kulikov wrote:
> To test frequent executions of coreutils out of a bash script I've run
> the following script at vim's source directory:
> 
> 	for i in $(seq 1 100); do ./configure; done
> 
> It should execute significant number of awk, uname, ls, etc.
> 
> The difference in -z,now and without it is negligible - less than a
> second of total 19 min 47 secs.  So, I still think we should globally
> enable it and disable for specific binaries :-)

OK.

> Would perl/etc. really suffer from 10% startup slowdown?  AFAIU the most
> significant slowdown of web services is SQL or bloated script, but not
> an interpreter.

You're right, it rarely matters if we're able to execute a script 3000
or 3300 times per second.  Typical website request rates and bottlenecks
are much lower than that.  At something like 100 req/sec this may amount
to 0.3% of total CPU time spent.  Perhaps we can afford that.

BTW, also similarly relevant is suexec startup time.  So we're paying
this price twice if we're executing a script via suexec.  However,
suexec is so tiny that the impact on its startup time is probably a lot
less (like 1% maybe?)

Alexander

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.