The current bottleneck with file systems is the speed of getting data
on or off the magnetic surface. So filesystem compression helps, as
less data needs to be physically written or read per image. The CPU
time spent compressing the data is less than the time saved in writing
less data to the surface.
I would be interested to see if the speed up is the same with a solid
state drive, as there is near 'random access' here, unlike with a
magnetic drive where the seek time is one of the bottlenecks. For
example, mechanical hard drives are limited to about 130MB/s, whereas
SSDs can already manage 200MB/s (faster than a first generation SATA
interface at 150MB/s can cope with and one of the drivers behind the
2nd (300MB/s) and 3rd generation (600MB/s) SATA intefaces). The large
size of our image files should make them ideal for use with SSDs.
Quoting "James Holton" <[email protected]>:
I think it important to point out that despite the subject line, Dr.
Scott's statement was:
"I think they process a bit faster too"
Strangely enough, this has not convinced me to re-format my RAID
array with an new file system nor re-write all my software to
support yet another new file format. I guess I am just lazy that
way. Has anyone measured the speed increase? Have macs become
I/O-bound again? In any case, I think it is important to remember
that there are good reasons for leaving image file formats
uncompressed. Probably the most important is the activation barrier
to new authors writing new programs that read them. "fread()" is
one thing, but finding the third-party code for a particular
compression algorithm, navigating a CVS repository and linking to a
library are quite another! This is actually quite a leap for those
of us who never had any formal training in computer science.
Personally, I still haven't figured out how to read pck images, as
it is much easier to write "jiffy" programs for uncompressed data.
For example, if all you want to do is extract a group of pixels
(such as a spot), then you have to decompress the whole image! In
computer speak: fseek() is rendered useless by compression. This
could be why Mar opted not to use the pck compression for their
newer CCD-based detectors?
That said, compressed file systems do appear particularly attractive
if space is limiting. Apparently HFS can do it, but what about
other operating systems? Does anyone have experience with a Linux
file system that both supports compression and doesn't get corrupted
easily?
-James Holton
MAD Scientist
Graeme Winter wrote:
Hi David,
If the data compression is carefully chosen you are right: lossless
jpeg2000 compression on diffraction images works very well, but is a
spot slow. The CBF compression using the byte offset method is a
little less good at compression put massively faster... as you point
out, this is the one used in the pilatus images. I recall that the
.pck format used for the MAR image plates had the same property - it
was quicker to read in a compressed image that the raw equivalent.
So... once everyone is using the CBF standard for their images, with
native lossless compression, it'll save a fair amount in disk space
(=£/$), make life easier for people and - perhaps most importantly -
save a lot of data transfer time.
Now the funny thing with this is that if we compress the images before
we store them, the compression implemented in the file system will be
less effective... oh well, can't win em all...
Cheers,
Graeme
2009/9/18 Waterman, David (DLSLtd,RAL,DIA) <[email protected]>:
Just to comment on this, my friend in the computer game industry insists
that compression begets speed in almost all data handling situations.
This will be worth bearing in mind as we start to have more fine-sliced
Pilatus 6M (or similar) datasets to deal with.
Cheers,
David.
-----Original Message-----
From: CCP4 bulletin board [mailto:[email protected]] On Behalf Of
William G. Scott
Sent: 17 September 2009 22:48
To: [email protected]
Subject: [ccp4bb] I compressed my images by ~ a factor of two, and they
load and process in mosflm faster
If you have OS X 10.6, this will impress your friends and save you some
disk space:
% du -h -d 1 mydata
3.5G mydata
mv mydata mydata.1
sudo ditto --hfsCompression mydata.1 mydata rm -rf mydata.1
% du -h -d 1 mydata
1.8G mydata
This does hfs filesystem compression, so the images are still recognized
by mosflm, et al. I think they process a bit faster too, because half
the information is packed into the resource fork.
This e-mail and any attachments may contain confidential,
copyright and or privileged material, and are for the use of the
intended addressee only. If you are not the intended addressee or
an authorised recipient of the addressee please notify us of
receipt by returning the e-mail and do not use, copy, retain,
distribute or disclose the information in or attached to the e-mail.
Any opinions expressed within this e-mail are those of the
individual and not necessarily of Diamond Light Source Ltd.
Diamond Light Source Ltd. cannot guarantee that this e-mail or any
attachments are free from viruses and we cannot accept liability
for any damage which you may sustain as a result of software
viruses which may be transmitted in or with the message.
Diamond Light Source Limited (company no. 4375679). Registered in
England and Wales with its registered office at Diamond House,
Harwell Science and Innovation Campus, Didcot, Oxfordshire, OX11
0DE, United Kingdom
----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.
