* Re: writing disk image to file
2003-06-09 3:53 writing disk image to file James Miller
@ 2003-06-09 5:52 ` Ray Olszewski
2003-06-10 19:40 ` Stephen Samuel
1 sibling, 0 replies; 4+ messages in thread
From: Ray Olszewski @ 2003-06-09 5:52 UTC (permalink / raw)
To: linux-newbie
At 10:53 PM 6/8/2003 -0500, James Miller wrote:
>I've got a small disk whose contents I'd like to save as an image file, in
>case I ever want to write it back to the disk it came from. This is a 4MB
>flash disk. From what I gather in my Linux reading, I should do something
>like the following: hook this tiny IDE disk up as, say /dev/hdb and the do
>something like "dd if=/dev/hdb of=dev-hdb-image.img" ? ? ? (the question
>marks outside the quotes indicates things I'm not sure about). This will
>write an image file to, I suppose the current dir - correct?
Yes. As always, there are unstated assumptions ... for example, the user
executing the dd command needs read access to the hdb device (not its
filesystem, the block device itself) and write access to the pwd (=
"present working directory", Linux/Unix jargon for what you call "the
current dir"). But these unstated assumptions are nothing special ...
though in practice they usually mean you need to run dd as root.
> Is the
>output file's name somewhat a matter of indifference, as it seems to me?
>E.g., wouldn't necessarily require any extension (might be called simply
>"dev-hdb-image"), though an extension might help me keep track of just
>what sort of file it is.
Unlike Windows, in Linux (and Unix) extensions have no special meanings to
the system; they are just part of the filename, there for the convenience
of the user. There are conventions -- .c means a C source program, .pl a
Perl program, .cpp a C++ program, .sh a shell script ... but these are
conventions only. Though some applications (like the gcc compiler) enforce
some extension rules about the files they work with, there is nothing here
that corresponds to Windows using the extension to identify the application
that is used to open a file.
Pick whatever you want to ... though .img is an extension commonly used for
this sort of image file (typically in the context of floppy images).
>And, finally, the question marks: I see
>sometimes "bs=512" or some such following the proposed file name which, as
>I get it, refers to block size. I'm not really clear on block sizes and
>what they do, but I have run accross information indicating that, in the
>case of an image file, it's not too important and can be left out. Can
>anyone illumine my benightedness on this matter?
Picking a block size that matches the size of the physical device in the
transfer (in this example, that would be hdb) can speed up the transfer
process. But I haven't seen it as having any other practical significance,
and even that effect is small with today's fast systems.
>Thanks, James
>
>PS Ray, I recall you asking at some point not long ago about copying
>one disk to a disk of a different size. In Rute User's Guide he writes
>"If they (disks) are not the same size, you will have to use tar or
>mirrordir to replicate the file system exactly." I guess you got that
>task accomplished, but did you do it using this means, or by some other?
At least for the system I was cloning, doing a dd worked fine, with the
destination drive being larger than the source drive. I was later able to
use fdisk to create additional partitions in the unused area of the
destination drive.
-
To unsubscribe from this list: send the line "unsubscribe linux-newbie" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.linux-learn.org/faqs
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: writing disk image to file
2003-06-09 3:53 writing disk image to file James Miller
2003-06-09 5:52 ` Ray Olszewski
@ 2003-06-10 19:40 ` Stephen Samuel
2003-06-10 19:57 ` Mike Dresser
1 sibling, 1 reply; 4+ messages in thread
From: Stephen Samuel @ 2003-06-10 19:40 UTC (permalink / raw)
To: James Miller; +Cc: Linux-Newbie list
James Miller wrote:
> I've got a small disk whose contents I'd like to save as an image file, in
> case I ever want to write it back to the disk it came from. This is a 4MB
> flash disk. From what I gather in my Linux reading, I should do something
> like the following: hook this tiny IDE disk up as, say /dev/hdb and the do
> something like "dd if=/dev/hdb of=dev-hdb-image.img" ? ? ? (the question
Yep. That would work fine as is.
.....
> what sort of file it is. And, finally, the question marks: I see
> sometimes "bs=512" or some such following the proposed file name which, as
> I get it, refers to block size. I'm not really clear on block sizes and
> what they do, but I have run accross information indicating that, in the
> case of an image file, it's not too important and can be left out. Can
> anyone illumine my benightedness on this matter?
Block sizing indicates how big of a chunk DD will read and write the data
in (dd can also split them up into ibs=___ and obs=___ -- in which case it
will create separate input and output buffers of different sizes).
The main purpose of the BS parameters was to take care of I/O devices for
which the write size made a difference. tape drives with variable block
sizes or streaming come to mind. Depending on the tape drive, you could
sometimes end up with very different performance and even effective tape
capacities depending on how big your write block sizes were.
In the current day, the biggest value to bs= is taking advantage of the
HD's caching capabilities. I.E. sometimes reading in chunks as big as (or
much bigger than) the track/cylinder size can result in a notable performance
increase. It also cuts down on system overhead.
My quick test (below) showed a barely noticable performance increase, but a
more noticable cut (percentage wise) in terms of CPU usage. In this case,
I'm comparing a 512byte buffer with a 5meg buffer:
Note that I'm testing read performance only, not write perfromance (no spare
disk partitions to mess with)
[root etc]# time /usr/bin/time dd if=/dev/hda5 of=/dev/null bs=512
2056256+0 records in
2056256+0 records out
real 1m12.906s
user 0m2.430s
sys 0m11.050s
[root etc]# time /usr/bin/time dd if=/dev/hda5 of=/dev/null bs=5120k
200+1 records in
200+1 records out
real 1m12.585s
user 0m0.010s
sys 0m10.020s
In this case, the 10000 times factor in writes resulted in the user CPU time
dropping from 2.4 seconds to 10miliseconds while the system CPU consumption
only dropped by 10%
>
> Thanks, James
>
> PS Ray, I recall you asking at some point not long ago about copying
> one disk to a disk of a different size. In Rute User's Guide he writes
> "If they (disks) are not the same size, you will have to use tar or
> mirrordir to replicate the file system exactly." I guess you got that
> task accomplished, but did you do it using this means, or by some other?
What will happen if you copy (say) a 1GB partition to a 4GB partition
is that it will more or less work, but you still have a filesystem that
has data indicating that there's only 1GB of disk space in the partition --
in other words, it will ignore the extra 3GB of disk space. The main
solutions to that problem are either:
1) create (mkfs) a new (empty) partitions and then bit copy the FILES (at the
user level) onto the new partition (e.g. with tar), or
2) bit-copy the partition over, and then manipulate the File System data directly
to be consistent with the extra available disk space. parted is the tool that
I usually use for things like that.
--
Stephen Samuel +1(604)876-0426 samuel@bcgreen.com
http://www.bcgreen.com/~samuel/
Powerful committed communication, reaching through fear, uncertainty and
doubt to touch the jewel within each person and bring it to life.
-
To unsubscribe from this list: send the line "unsubscribe linux-newbie" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.linux-learn.org/faqs
^ permalink raw reply [flat|nested] 4+ messages in thread