[cfarm-users] cleanup on gcc119?
/scratch is nice and big on gcc119, but /tmp is pretty full. Filesystem 1024-blocks Free %Used Iused %Iused Mounted on/dev/hd3 2097152 23616 99% 16322 24% /tmp/dev/ramdisk0 67108864 41979172 38% 152696 2% /scratch Steven R. Loomis IBM Global Foundations Technology Team Technical Lead, ICU for C/C++ Phone: 1-720-342-4930 https://ibm.biz/srloomis E-mail: srloo...@us.ibm.com ___ cfarm-users mailing list cfarm-users@lists.tetaneutral.net https://lists.tetaneutral.net/listinfo/cfarm-users
Re: [cfarm-users] peko, please 'nice' your jobs
On Mon, Mar 26, 2018 at 03:26:53PM +0100, Jonathan Wakely via cfarm-users wrote: > Peter's jobs rarely use a huge number of cores (and gcc112 has a lot > of them) so my request was as much that he doesn't just run constant > jobs in a permanent loop, which the machines are not meant to be for. It has only 20 cores. It has 160 threads, but that is not the same thing. Segher ___ cfarm-users mailing list cfarm-users@lists.tetaneutral.net https://lists.tetaneutral.net/listinfo/cfarm-users
Re: [cfarm-users] peko, please 'nice' your jobs
> "Stefan" == Stefan Ring via cfarm-users >writes: Hi, > On Mon, Mar 26, 2018 at 3:56 PM, Jonathan Wakely via cfarm-users > wrote: >> I can't remember ever using gcc112 when it wasn't running several jobs >> for "peko", all run without nice. >> >> Please remember these are shared machines, and running jobs >> back-to-back all day, every day doesn't seem necessary. > Unfortunately, nice buys you close to nothing on machines with > multiple virtual threads per core (Hyper Threading or SMT), and the > POWER machines are heavily multi-threaded. So the only good remedy is > to start fewer simultaneous jobs than there are physical cores in a > machine. I've scaled down from make -j16 to -j8. -- Bye, Peter Korsgaard ___ cfarm-users mailing list cfarm-users@lists.tetaneutral.net https://lists.tetaneutral.net/listinfo/cfarm-users
Re: [cfarm-users] peko, please 'nice' your jobs
On Mon, Mar 26, 2018 at 10:15 AM, Peter Korsgaard via cfarm-userswrote: >> "Jonathan" == Jonathan Wakely via cfarm-users >> writes: > > > I can't remember ever using gcc112 when it wasn't running several jobs > > for "peko", all run without nice. > > Sorry, fixed. > > > Please remember these are shared machines, and running jobs > > back-to-back all day, every day doesn't seem necessary. > > This is an autobuilder (randconfig) setup for Buildroot, but I have > scaled back the number of jobs to 1/4, hopefully that is OK? Peter, If you are running an autobuilder that is running continually, then you should request a dedicated VM from OSUOSL. The GNU Compile Farm system is okay for intermittent builders and testers, but not intended for a user to monopolize the system -- even part of the system -- for an autobuilder that runs continuously. Thanks, David ___ cfarm-users mailing list cfarm-users@lists.tetaneutral.net https://lists.tetaneutral.net/listinfo/cfarm-users
Re: [cfarm-users] peko, please 'nice' your jobs
[resent with correct from] > "Jonathan" == Jonathan Wakely via cfarm-users >writes: > I can't remember ever using gcc112 when it wasn't running several jobs > for "peko", all run without nice. Sorry, fixed. > Please remember these are shared machines, and running jobs > back-to-back all day, every day doesn't seem necessary. This is an autobuilder (randconfig) setup for Buildroot, but I have scaled back the number of jobs to 1/4, hopefully that is OK? -- Bye, Peter Korsgaard ___ cfarm-users mailing list cfarm-users@lists.tetaneutral.net https://lists.tetaneutral.net/listinfo/cfarm-users