I've written a program that needs to read/write lots of cache
files to
disk. They are around 100k a piece though it can vary quite a bit.
In response to user action I need to load 2-5 of these cache files
and
update the GUI. On windows and linux the read times for these cache
images average around 250ms. However on OS X the mean time to load
rises to 450ms with some exceeding 1200ms.
Are you sure the bottleneck is I/O speed, and not something else like
decoding the image-files?
Have you run a profile on your actual program, to find out where
the time
is really being spent?
Are you sure there isn't a networked disk-volume involved?
Yeah something seems a little weird in your numbers William, we
likely need more information about what is going on. Also no idea
what system you are using (unless I missed it).
On my Mac OS X 10.4.2 system I can run the following grep test
against 10,000 files all 100KB in size (remade my test files to that
size, 100 folders deep with 100 files in each).
[G5:~/test] shawnce% time grep -a -R 666 dir_0 > /dev/null
1.560u 1.408s 0:03.04 97.3% 0+0k 0+1io 0pf+0w
This tells me that my PowerMac Dual G5 2GHz (with one CPU burning
away on Folding@home) is able to open, read, and search 976 MB
(10,000 files * 100KB per file) in 3.04 seconds when coming from the
file cache (had already touched the files in question, no disk IO was
incurred).
That is about 321 MB/s overall and about 3,289 100KB files per
second, generally limited by CPU.
If I do the test again without things in the file cache I get....
[G5:~/test] shawnce% time grep -a -R 666 dir_0 > /dev/null
2.161u 8.148s 0:37.58 27.4% 0+0k 441+545io 0pf+0w
That is about 26 MB/s overall and about 266 100KB files per second.
This means it is taking around 4 milliseconds to read and process
each file, clearly much faster then you reported above (4 ms * 5
files is 20 ms about 20 times faster then what you reported). Unless
I did some math incorrectly...
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Java-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/java-dev/email@hidden
