On Fri, 14 Feb 2003, jvand2 wrote: > Does your redhat have a "timer patch" similar to the one for SuSE (2.4 > kernel)?
i dont think so, im using the 2.4.9-38 kernel, with no patches. > If so we found that it does add a some amount of overhead in the long run. In > some > of our testing for a process with lots of sequential I/O we save on average > 20% of the wall clock runtime buy using the standard kernel with out the > timer patch. > > Also I did not see from the other posts if you were using LVM. If so we found > that stripe accross all the PV's made on average 50% improvement. You > might try an LVM (even if you do not need it for space) with at least 2 PV's. the / and /usr are ext3, without lvm. > > In some emails about Linux with LVM on the shark to IBM they said a block > size of 4KB and a chunk size of 1MB should give the best performance. I > have not tried a change to the LVM extent to 1MB yet, so I do not know if it > helps. It might be something you can look into. > > On Friday 14 February 2003 16:41, you wrote: > > On Fri, 14 Feb 2003, David Boyes wrote: > > > > Another test we ran yesterday where to tar (WITHOUT gzip, > > > > only tar) 100 > > > > Mb, we ran the test about 30 times, without modifing the > > > > enviroment, the > > > > 70-80% times we get about 1.4s, but 20-30% times the time > > > > vary from 2s to > > > > 8s, without any reason, then, the "specialist" from ibm told > > > > us that we > > > > must try with suse (kernel 2.2), because "maybe" the kernel > > > > from redhat > > > > have some problems. > > > > what do you think? > > > > > > I think you need a different "specialist". This one is playing Easter > > > Egg Hunt with you and giving you bogus information. > > > > > > If anything, going back to a 2.2 kernel will make it worse, as that > > > release lacked several important improvements. The variation in times > > > is not too unusual if there are other virtual machines active on the > > > same VM image -- CP is doing something else and has to timeslice. > > > > when i said the same he told us that he only wants to test if the problem > > is at linux level or at zvm level > > > > zvm has two weeks installed, there is only the osasf, rtm and tcpip vm up, > > but they have no work to do, there is no significat network traffic. > > > > > Do you have a VM performance tool installed, ie FCON or ESAMON, or even > > > RTM/VM? Wall clock time isn't really going to tell you very much about > > > what's happening in your environment. > > > > we have rtm/vm > > > > > -- db > -- Alejandro Leyva Rabinovich. Jefe de la Unidad Departamental de Soporte Ticnico (Administracisn de Mainframe). Direccisn General de Informatica. Secretarma de Finanzas. Gobierno del Distrito Federal.
