To: Felipe Alfaro Solana Cc: Con Kolivas , ck@vds.kolivas.org Bcc: Subject: Re: [ck] RFD: kernel embetterment Reply-To: In-Reply-To: <719B25E7-95AB-4B28-A8E2-F54DBBF40D26@mac.com> On Mon, May 09, 2005 at 03:24:00PM +0200, Felipe Alfaro Solana wrote: > >AFAIK, they impose some performance penalty on the kernel itself and >I guess high-performance, server-oriented guys won't like this kind >of decision. Also, you can't tell them to recompile the kernel in Yeah, you're most likely right. But it takes a lot to tear down a high-end server and a lot of customers have money to buy more and quite a lot of magic can be done in userspace for performance. (I never thought I'd hear myself say "Fix problems by throwing more money and hardware at it", but it seems to be the fastest CADT way in our world..) But whatever it would take to track down losses :) >order to disable the framepointer support since many of them will run >commercial distributions like Red Hat, and this will usually break >the support agreements. Vendors... It's really nice of them to test their kernels, and backport fixes, but it comes at the cost of things still not being fixed upstream in -rc and -mm and -rc-mm and gentoo patch set of the day and even further ties the customers down from doing any patching of their own. >I don't think forking is needed at all... look at -ck patchset: >there's one for workstations and one for server-class machines. I >guess distributions could do the same: distribute a generic kernel >for workstations and another one for production servers. That's one of my issues. -ckserver is not that different, from what I seem to remember, disabling some features and not using some defaults, turns -ck into -ckserver. If the vendors ship out kernels with different target groups, what's up with not making that the mainline behavior, if it's deemed necessary. But I think we've seen that Linux can scale, maybe it'd be more important to keep it that way. Then again, does anyone know how many embedded vendors use vanilla out of the box, or do they patch it some and never release any changes because no one really bothered to ask? Besides, the word on the streets is that 2.6.12 will fail, because we're at -rc3 or -rc4 or -rc4-mm6-bk1024 or whatever and people report filesystems spontaneously combusting all over the board. There are also rumors of kswapd bugging. We just got smp-balanced nice, which I sort of assumed was there; we just got it after eight revisions of the eleventh kernel for god's sake, because it accidentally worked in 2.4 and no one bothered to think twice about this. This should never never ever be even remotely possible to happen. PS. No, I didn't post while I was angry, I just started feeling a bit poignant while writing, and I'm still sticking to Linux :) -- mjt