On Tue, 2009-03-10 at 11:34 -0400, Brian Hutchinson wrote: > Thanks for your quick reply. > > Specifically, my laptop is a Dell Latitude D620. I turned off a lot > of the features in Kernel Hacking section (stack checking etc.) > Laptops are a bit notorious for heavy BIOS activity. > I haven't ran cyclictest with the -b option yet as I can't find some > of the .config parameters in this kernel that match the cyclictest > latency checking documentation. Before I beat my head against the > wall too hard I thought I'd ping others and see if what I'm seeing is > normal or if others were able to run on a PC and obtain very low > latency. > this is out of a script I use. sudo chrt -f 80 /usr/bin/time /usr/local/bin/cyclictest -n -q -p 99 -a -t -i 500 -l $NLOOPS -h 10000 -m | tee -a $LOGFILE > I'm trying to come up to speed on the latest 2.6 kernels after having > most experience with 2.4 and early 2.6. I see the TSC isn't used > anymore. I thought about just not using ACPI but I think that will > kill my high res timers. I do see a local ACPI otion so I don't know > if that will make high res timers work ... I'm reading > conflicting/dated info. so I thought I'd consult the experts so they > can show me the way :) > ACPI is fine, deadlock detect is not... Of the available patch sets, the 2.6.24-RT patch set is being used by a few distros. The 2.6.22 RT Kernel Suse ships is equally well tested and performs better, or worse, depending who is getting paid by whom to do the benchmark :) Point is: Use the .22. or the .24 to get best-case numbers for your config and platform, then take that config forward to .26 and beyond, let us know about regressions. I can offer you the configs for 2.6.22 in these packages as tested/optimal. http://ftp.suse.com/pub/projects/kernel/kotd/SLERT10_BRANCH/ Sven > Regards, > > Brain > > > On Tue, Mar 10, 2009 at 10:56 AM, Sven-Thorsten Dietrich > <sven@xxxxxxxxxxxxxxxxxxxxx> wrote: > On Tue, 2009-03-10 at 14:06 +0000, Brian Hutchinson wrote: > > We don't have our target hardware yet so I'm trying to get a > jump on some of the > > newer 2.6 kernel features using a PC. > > > > > What hardware are you testing on? > > > Compiled a 2.6.26.8 with the 2.6.26.8-rt16 patch applied for > a Dell Core Duo > > laptop and when I run cyclictest the max latency is up > around 800us with the > > avg. around 65us. > > > > > You should compare to the 2.6.24-RT patch set first of all, I > would > expect worst-cases to be 25% of what you are seeing. > > If it does not, then your configuration may have too much > debugging > turned on (boot log warns about some of these) - or you are > indeed > getting some SMI interference. > > > I'm running the CFS scheduler in real time mode. > > > > Is this the best I'm going to see on a PC? I've been > reading everything I can > > get my hands on and it looks like SMI and BIOS may hinder me > from getting much > > below what I'm seeing. > > > > > SMIs can be turned off in many cases. > > Some vendors have warranty/support programs tied to SMI > architectures, > in these cases you might be able to press them for a custom > BIOS > (depending on volume and how much hell you want to raise) > > Cheers, > > Sven > > > > TIA! > > > > Regards, > > > > Brian > > > > -- > > To unsubscribe from this list: send the line "unsubscribe > linux-rt-users" in > > the body of a message to majordomo@xxxxxxxxxxxxxxx > > More majordomo info at > http://vger.kernel.org/majordomo-info.html > > > -- To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
