Re: Hard-limits for kern.maxproc
Re: Hard-limits for kern.maxproc
- Subject: Re: Hard-limits for kern.maxproc
- From: mm w <email@hidden>
- Date: Fri, 6 Feb 2009 09:04:41 -0800
"One would hope that Apple's Mac OS X designers and developers will
eventually provide the same modern kernel resource allocation and
configuration features that are already provided in other Unix/Linux
operating systems if Apple hopes to be able to sell Mac OS X Server
into any kind of serious production environment."
Hello, this comment is somehow fair enought, but I would like to add
some comments about it:
1- that always depends on what your processes are doing: aka I could
have 10 really griddy processes
against thousands
2- I think this work has been started for a while, it requires a
better traceability of a proc, if you look at the different version
src, you may reach my point concerning this and some proc arg not yet
implemented in various areas (from 10.4.0 to 10.5.x)
3- Darwin firstable, has been designed by for the Desktop version and
have a really stable OS,
x-server is a really shy try in the server world, it's a tough
decision to say I will challenge the world leaders
since several years on this market, this is the same than challenging
Microsoft on a Desktop OS.
secondable, Darwin is really young in comparaison of AIX AND Sun
Solaris, plus those NIX are designed to be servers, and it's not yet
the case for Darwin that's shared src between both projects.
so for now, the main point for this discussion is: "I have a problem
regarding server tuning for my needs",
I try three ways:
- "highjack" kern_proc thru a kmem based prog, working, uggly
- "highjack" kern_proc thru a kext, working, uggly
- recompile Cyrus working nice, with some net-filtering connections
(an another really nice tool,
and I am really not a sysadmin, I just did this, spare time, by pure
intellectual interest)
- 4eme, I didn't try it: change your imap server.
Conclusion, the dynamic tuning is not a "fact" to correct software mistakes?
in other hand, should I (kernel) be more "unstable" (now how many
threads by proc) to correct userspace design mistakes, even if you
have a dynamic tuning you have to decide of a limit, and sure this
depends on
"what is the job of the machine?"
so, I am sorry when someone who is supposed to be engineer says: "I
will wait for the next generation of computers", or maybe in ten
years... to fix the problem, are you crazy or lazy?, you know I was
used to work with people who were building their hardware just for
testing their Algorytms that didn't fit with the market-proposed
technology that you can find know in a supermarket, so you can imagine
if those people did the same, "ho I will wait maybe in 10/20 years I
could verify my theory, for now I m going in vacation"
Cheers!
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-kernel mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden