On Sunday 23 January 2005 11:54, Christian Wicke wrote: > Hello, > > I experience a huge performance loss when I run my programms inside a > UML-child. Creating 1,000 files took eighteen times as long as on the > UML-host.
> A simple shell script to count up to 100,000 took three times as long. > Is this normal? More or less... you don't describe your environment, from the host config I guess it's SKAS however... Do you set TMPDIR to a tmpfs or ramfs mount point? Without those tricks, any test is pointless. Creating a process inside UML takes a lot longer of the host... try to rewrite the scripts in perl or C, or any scripting language, and you'll see that the difference reduce a lot. The problem is that starting the "touch" process is a lot slower than creating the file itself! Try this: $ > a to create that file, and times will probably improve (I'm not sure, since probably the shell forks anyway to execute a command, so we loose anyway)... On the other side, you'd get a 18x difference even if you ran a loop with of 100.000 /bin/echo... or maybe in that case the difference would be even bigger. In many real workloads, however, things are a lot better... pure userspace computation, without page faults, sees no difference inside UML vs. inside the host, and a kernel compilation gives only something littler than 2x as slowdown ratio... (I don't have even a order of figures about this, there was something in the list)... > I run debian sarge with Linux 2.6.8 with skas-patch on the host and debian > woody with Linux 2.4.25 for the uml-guest. The configuration files are > attached. > Host and guest use reiserfs. The guest has a lvm-device as UDB; so writing > should only go through the filesystem layer of the guest. No, until you use UBD you also go through the block device layer (the UBD driver is not different from a IDE driver, it only uses the host VFS instead of playing with hardware). That's not the problem, however. > I checked what overhead causes lvm: It is less than 10%, so this is not the > problem. > Swapping isn't the problem either. The host uses almoust no swap-memory. You should check for the guest too, but I have no reason to think the guest is swapping. > > Creating 1000 files: > UML-host: > # time ./test_create_file.sh > > real 0m1.761s > user 0m0.577s > sys 0m1.145s > > UML-guest: > # time ./test_create_file.sh > > real 0m32.769s > user 0m2.970s > sys 0m29.750s > > #cat test_create_file.sh > x=1; while [ $x -le 1000 ]; do x=$((x+1)); touch file_$x; done > > Counting to 100000: > UML-host: > # time ./test_count.sh > > real 0m2.780s > user 0m2.716s > sys 0m0.059s > > UML-guest: > # time ./test_count.sh > > real 0m8.675s > user 0m7.510s > sys 0m1.180s > > # cat test_count.sh > x=1; while [ $x -le 100000 ]; do x=$((x+1)); done > I start the UML-guest with > /usr/bin/SCREEN -d -m -- su -c HOME=/var/local/afprod2 /opt/uml/bin/linux > mem=222M umid=afprod2 eth0=tuntap,tap12 ubd0=/dev/mapper/vg1-afprod2 > con0=fd:0,fd:1 con=pts afprod2 -- Paolo Giarrusso, aka Blaisorblade Linux registered user n. 292729 http://www.user-mode-linux.org/~blaisorblade ------------------------------------------------------- This SF.Net email is sponsored by: IntelliVIEW -- Interactive Reporting Tool for open source databases. Create drag-&-drop reports. Save time by over 75%! Publish reports on the web. Export to DOC, XLS, RTF, etc. Download a FREE copy at http://www.intelliview.com/go/osdn_nl _______________________________________________ User-mode-linux-user mailing list User-mode-linux-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user
