Author Login
Post Reply
Michael Meyer wrote:
> On mere build servers running linux one could disable
> the second core at boot time permanently to reduce the
> build time, if no other processes are running on this
> server.
>
> But on workstations the second core is often used by
> the application being developed if it is
> multithreaded, so there might be no way in linux to
> reduce the build time wih two cores?
we dont want to turn core #2, off, or even cores #3 and #4. Not when my
memory/process footprint has 3 vmware images (windows for exchange,
centos for release builds, rhel4 for testing and then the IDE on my main
machine)
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
9184 slo 18 0 2443m 802m 15m S 0 13.4 84:51.39 java
9129 slo 5 -10 770m 579m 561m S 0 9.7 38:32.89 vmware-vmx
8403 slo 5 -10 768m 570m 556m S 16 9.5 850:09.04 vmware-vmx
8361 slo 15 0 891m 368m 29m S 2 6.2 63:29.62
firefox-bin
9144 slo 5 -10 501m 304m 290m S 0 5.1 6:12.59 vmware-vmx
8253 slo 15 0 373m 251m 22m S 0 4.2 12:02.73
thunderbird-bin
29911 slo 18 0 1641m 177m 12m S 0 3.0 7:42.59 java
5708 root 15 0 314m 167m 59m S 0 2.8 30:08.87 Xorg
24284 slo 17 0 1560m 141m 8692 S 0 2.4 0:35.69 java
19907 slo 15 0 645m 92m 61m S 0 1.6 0:04.22 soffice.bin
11023 slo 15 0 609m 69m 36m S 0 1.2 22:51.60 amarokapp
8304 slo 15 0 384m 54m 28m S 0 0.9 1:47.77 vmware
I had a look at the code and there's no obvious synchronisation in the
javac code. If there is a delay, it may be in the output/listener logic,
which could well be synchronized. On a multicpu system the cost of
acquiring a lock may be higher.
I'd love to track this down, but nothing obvious springs to mind.
-could you try building under vmware/xen? We do that for releases to
ensure the release builds are with clean source, no matter what I'm editing.
-steve
---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscribe@(protected)
For additional commands, e-mail: user-help@(protected)