From mboxrd@z Thu Jan 1 00:00:00 1970 From: John David Anglin Subject: Re: debian hppa Date: Thu, 07 Jul 2011 13:26:52 -0400 Message-ID: <4E15EC5C.10702@bell.net> References: <20110527213516.GH21744@dannf.org> <20110527233738.0B95D4FCB@hiauly1.hia.nrc.ca> <20110601021049.GB3984@hiauly1.hia.nrc.ca> <20110707161810.GC12243@dannf.org> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Cc: Thibaut VARENE , John David Anglin , carlos@systemhalted.org, linux-parisc@vger.kernel.org To: dann frazier Return-path: In-Reply-To: <20110707161810.GC12243@dannf.org> List-ID: List-Id: linux-parisc.vger.kernel.org On 7/7/2011 12:18 PM, dann frazier wrote: > On Thu, Jul 07, 2011 at 05:49:36PM +0200, Thibaut VARENE wrote: >> On Thu, Jul 7, 2011 at 1:20 AM, John David Anglin >> wrote: >>> Hi Thibaut, >>> >>> On 13-Jun-11, at 6:19 PM, Thibaut VARENE wrote: >>> >>>>> I would say your other machines need updating even if the process is >>>>> somewhat rocky as that's the only way a broadbased release can be tested. >>>> Well I can assign one machine (the a500, being easy to reboot/fix as >>>> it is) to testing kernels, but for a stable kernel suitable for the >>>> machines I'm e.g. assigning to the GCC Compile Farm[0], what would be >>>> a good SMP kernel version to choose? I'm looking for the same level of >>>> hassle-freeness as 2.6.22.19 since these machines are under relatively >>>> heavy load and I cannot afford to tend to their care on a daily basis >>>> ;-) >>> Sorry for the delay in responding. I think assigning the a500 to the GCC >>> farm >>> provides good visibility. For GCC, the kernel version doesn't matter much. >>> The >>> main issue is stability. >> OK, so I suppose it's fine to stick with 2.6.22.19? I thought there >> were issues with un-implemented syscalls and such, so I don't really >> know... >> This comment was specific to building and testing GCC. It currently takes slightly over 24 hours to build and test GCC trunk on my c3750. > Honestly, the most reliable build machine I've used is my > C3700. I don't recall ever having any build issues with > that. Anything it gives up in speed is definitely outweighed by > having to e.g. retry gcc 5 times before succeeding. By the end of the > port, I'd normally handbuild new gcc uploads here to avoid thrashing > on the real buildds. If given the option, I'd recommend 3 of those > boxes for a buildd ring. > Based on my testing, it's no longer necessary to try 5 times before succeeding with rp3440. I agree that the C37XX machines are very stable and don't suffer from the SMP/cache issues present in the servers. If the ring above could be arranged, that would be excellent. Dave -- John David Anglin dave.anglin@bell.net