ubuntu-boot team mailing list archive

[Scott James Remnant <scott@xxxxxxxxxxxxx>]

There's been a fair amount of tuning going on in udev upstream; while
this is largely concerned with memory usage, it should have a small but
measurable effect on boot performance as well.

First up, the difference in memory overhead:

	udev 124	udev 136	change
VmPeak  2556 kB		2328 kB		-228 kB (91%)
VmSize  2532 kB		2324 kB		-208 kB (91%)
VmHWM   1324 kB		 620 kB		-704 kB (46%)
VmRSS   1320 kB		 616 kB		-704 kB (46%)
VmData   536 kB		 304 kB		-232 kB (56%)

Definite improvement here, using something like half the actual resident
memory with a general reduction in peak, etc.


Timings taken by replacing udevd only:

	udev 124	udev 136

MEAN    6.90s		6.39s
MEDIAN  8.08s		7.64s
MODE    8.63s		7.68s
STDDEV  2.43s		2.28s

Slightly more sprightly then, events take a little less time on average
to be handled; the overhead inside the udev daemon is reduced, so this
is what we'd expect.

(Fewer missing uevents too, it occurred to me that these are simply
those that happen before udevmonitor gets fully started)


Numbers from manual instrumenting:

			udev 124	udev 136	change

udev in initramfs:        0.01s		0.01s		0 (100%)
udev in system:           0.05s		0.02s		-0.03s (40%)

processing in initramfs:  2.60s		2.81s		+0.21s (108%)
processing in system:    10.47s		9.95s		-0.52s (95%)

These results are actually what we'd expect.  In the initramfs, the new
udev makes little difference (the slight increase could be noise)
because there's few rules to parse there.

In the full system, there's a general reduction in time, due to the
larger number of rules -- and the increased efficiency of the new udev.

Scott
-- 
Scott James Remnant
scott@xxxxxxxxxxxxx

Attachment: signature.asc
Description: This is a digitally signed message part