[opensuse] Re: fdisk calculations
Steffen Winterfeldt a écrit :
2012217344/512/46/47 = 1817.813...
2012217344 is the disk size. And that's independent of the 'geometry'.
why should it be? the geometry size is the only available and OS independant. In fact this number gives 1919 mebibytes (1919*2^20=2012217344) isn't 1919 prime? I wonder how all this is built. Where are used the 900.096 bytes left? is the 2012217344 including extra bytes in sectors for sync/alignment and whatsoever the drive need? but we are here on flash drive, not magnetic, so what? thanks jdd -- http://www.dodin.net http://valerie.dodin.org http://www.youtube.com/watch?v=t-eic8MSSfM http://www.facebook.com/profile.php?id=1412160445 -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
On Monday March 16 2009, jdd wrote:
...
In fact this number gives 1919 mebibytes (1919*2^20=2012217344) isn't 1919 prime?
Speculation is not required (nor pesky arithmetic): % factor 1919 1919: 19 101
...
thanks jdd
Randall Schulz -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
On Mon, 16 Mar 2009, jdd wrote:
Steffen Winterfeldt a écrit :
2012217344/512/46/47 = 1817.813...
2012217344 is the disk size. And that's independent of the 'geometry'.
why should it be? the geometry size is the only available and OS independant.
Geometry values are just gimmick and not really used for anything but for confusing messages.
In fact this number gives 1919 mebibytes (1919*2^20=2012217344) isn't 1919 prime?
I wonder how all this is built. Where are used the 900.096 bytes left?
As has already been posted, 1919 is 101*19; so if you are worried about that disk space, change the geometry to something like 101 heads x 19 sectors/track. Steffen -- Der frühe Wirt holt sich den Wurm. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Steffen Winterfeldt a écrit :
As has already been posted, 1919 is 101*19; so if you are worried about that disk space, change the geometry to something like 101 heads x 19 sectors/track.
not really worried, only curious. flash memories are made from discrete chips, so shouldn't the value be some kind of exact bit count? anyway I aksed on the fdisk list - the first answer speaks about bug, I will follow jdd -- http://www.dodin.net http://valerie.dodin.org http://www.youtube.com/watch?v=t-eic8MSSfM http://www.facebook.com/profile.php?id=1412160445 -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Steffen Winterfeldt schrieb:
On Mon, 16 Mar 2009, jdd wrote:
Steffen Winterfeldt a écrit :
2012217344/512/46/47 = 1817.813...
2012217344 is the disk size. And that's independent of the 'geometry'.
why should it be? the geometry size is the only available and OS independant.
Geometry values are just gimmick and not really used for anything but for confusing messages.
No, especially for SSDs careful selection of Geometry values can be important: http://thunk.org/tytso/blog/2009/02/20/aligning-filesystems-to-an-ssds-erase... According to the nr. of heads* nr. sectors/head should a multiple of the erase-block size (most often 128MB), and all partitions should be aligned to cylinder boundaries. Especially the first partition of a disk is usually not aligned by default, because the first track, not cylinder, is excluded from it when you specify 1 as the starting cylinder. Not respecting the erase block size leads to file system block split over two erase blocks and therefore to worse performance and more erase blocks erased than necessary and so to shorter life time of the device.
[...]
Herbert -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
On Mon, Mar 16, 2009 at 10:30 AM, Herbert Graeber <hgraeber@opensuse.org> wrote:
Steffen Winterfeldt schrieb:
On Mon, 16 Mar 2009, jdd wrote:
Steffen Winterfeldt a écrit :
2012217344/512/46/47 = 1817.813...
2012217344 is the disk size. And that's independent of the 'geometry'.
why should it be? the geometry size is the only available and OS independant.
Geometry values are just gimmick and not really used for anything but for confusing messages.
No, especially for SSDs careful selection of Geometry values can be important:
http://thunk.org/tytso/blog/2009/02/20/aligning-filesystems-to-an-ssds-erase...
According to the nr. of heads* nr. sectors/head should a multiple of the erase-block size (most often 128MB), and all partitions should be aligned to cylinder boundaries. Especially the first partition of a disk is usually not aligned by default, because the first track, not cylinder, is excluded from it when you specify 1 as the starting cylinder.
Not respecting the erase block size leads to file system block split over two erase blocks and therefore to worse performance and more erase blocks erased than necessary and so to shorter life time of the device.
[...]
Herbert
Herbert, I agree with you for the first partition. After that I don't think tools like fdisk attempt to align with cylinder boundaries anymore. Given your comment above and the email I just posted 5 minutes ago about the new 4K alignment issues we will be seeing within the year, it sounds like it is time for partitioning tools to totally drop the concept of aligning with cylinders and start asking questions more relevant to the hardware. Specifically "what is the physical sector size (unit of atomicity) for the device? And what is the offset into the the device represented by the very beginning of the drive. Greg -- Greg Freemyer Head of EDD Tape Extraction and Processing team Litigation Triage Solutions Specialist http://www.linkedin.com/in/gregfreemyer First 99 Days Litigation White Paper - http://www.norcrossgroup.com/forms/whitepapers/99%20Days%20whitepaper.pdf The Norcross Group The Intersection of Evidence & Technology http://www.norcrossgroup.com -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Greg Freemyer schrieb:
On Mon, Mar 16, 2009 at 10:30 AM, Herbert Graeber <hgraeber@opensuse.org> wrote:
Steffen Winterfeldt schrieb:
On Mon, 16 Mar 2009, jdd wrote:
Steffen Winterfeldt a écrit :
2012217344/512/46/47 = 1817.813...
2012217344 is the disk size. And that's independent of the 'geometry'.
why should it be? the geometry size is the only available and OS independant.
Geometry values are just gimmick and not really used for anything but for confusing messages.
No, especially for SSDs careful selection of Geometry values can be important:
http://thunk.org/tytso/blog/2009/02/20/aligning-filesystems-to-an-ssds-erase...
According to the nr. of heads* nr. sectors/head should a multiple of the erase-block size (most often 128MB), and all partitions should be aligned to cylinder boundaries. Especially the first partition of a disk is usually not aligned by default, because the first track, not cylinder, is excluded from it when you specify 1 as the starting cylinder.
Not respecting the erase block size leads to file system block split over two erase blocks and therefore to worse performance and more erase blocks erased than necessary and so to shorter life time of the device.
[...]
I agree with you for the first partition. After that I don't think tools like fdisk attempt to align with cylinder boundaries anymore.
fdisk and other partitioning tools can work without aligning to cylinders. But when creating new partitions they usually respect cylinder boundaries. When you need special block sizes, eg. for SSDs or partitions on raid devices, you can use the -H and -S options to override the values 63 for sectors/track and 255 for heads.
Given your comment above and the email I just posted 5 minutes ago about the new 4K alignment issues we will be seeing within the year, it sounds like it is time for partitioning tools to totally drop the concept of aligning with cylinders and start asking questions more relevant to the hardware. Specifically "what is the physical sector size (unit of atomicity) for the device? And what is the offset into the the device represented by the very beginning of the drive It is not possible to drop the geometry completely, because this build deeply into the BIOS and the kernel.
Theoretically it would be possible to query the device for it's block size and select a proper geometry for partitioning. But this does not work, because many devices lie about their real geometry to be compatible to older Windows OSs. Herbert -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
On Mon, Mar 16, 2009 at 11:20 AM, Herbert Graeber <hgraeber@opensuse.org> wrote:
Greg Freemyer schrieb:
On Mon, Mar 16, 2009 at 10:30 AM, Herbert Graeber <hgraeber@opensuse.org> wrote:
Steffen Winterfeldt schrieb:
On Mon, 16 Mar 2009, jdd wrote:
Steffen Winterfeldt a écrit :
2012217344/512/46/47 = 1817.813...
2012217344 is the disk size. And that's independent of the 'geometry'.
why should it be? the geometry size is the only available and OS independant.
Geometry values are just gimmick and not really used for anything but for confusing messages.
No, especially for SSDs careful selection of Geometry values can be important:
http://thunk.org/tytso/blog/2009/02/20/aligning-filesystems-to-an-ssds-erase...
According to the nr. of heads* nr. sectors/head should a multiple of the erase-block size (most often 128MB), and all partitions should be aligned to cylinder boundaries. Especially the first partition of a disk is usually not aligned by default, because the first track, not cylinder, is excluded from it when you specify 1 as the starting cylinder.
Not respecting the erase block size leads to file system block split over two erase blocks and therefore to worse performance and more erase blocks erased than necessary and so to shorter life time of the device.
[...]
I agree with you for the first partition. After that I don't think tools like fdisk attempt to align with cylinder boundaries anymore.
fdisk and other partitioning tools can work without aligning to cylinders. But when creating new partitions they usually respect cylinder boundaries. When you need special block sizes, eg. for SSDs or partitions on raid devices, you can use the -H and -S options to override the values 63 for sectors/track and 255 for heads.
Given your comment above and the email I just posted 5 minutes ago about the new 4K alignment issues we will be seeing within the year, it sounds like it is time for partitioning tools to totally drop the concept of aligning with cylinders and start asking questions more relevant to the hardware. Specifically "what is the physical sector size (unit of atomicity) for the device? And what is the offset into the the device represented by the very beginning of the drive It is not possible to drop the geometry completely, because this build deeply into the BIOS and the kernel.
Theoretically it would be possible to query the device for it's block size and select a proper geometry for partitioning. But this does not work, because many devices lie about their real geometry to be compatible to older Windows OSs.
Herbert
From a performance perspective alignment with physical atomic write areas is critical, put I can see where that could break some boot code. It is very convoluted. I just posted to the lkml-ide list that
At a minimum tools like fdisk need to start asking the user about these parameters. The best partition layouts simply don't correlate to CHS values anymore. I don't know what the solution is when the physical device is claiming a sectors per track value that is not a multiple of its true atomic read/write region. ie. like the 4k sector hdd manufacturers are currently planning. they should try to encourage hardware manufacturers of the new 4K sector hard drives to report 56 sectors per track. I think that would drastically simplify things. Greg -- Greg Freemyer Head of EDD Tape Extraction and Processing team Litigation Triage Solutions Specialist http://www.linkedin.com/in/gregfreemyer First 99 Days Litigation White Paper - http://www.norcrossgroup.com/forms/whitepapers/99%20Days%20whitepaper.pdf The Norcross Group The Intersection of Evidence & Technology http://www.norcrossgroup.com -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
participants (5)
-
Greg Freemyer -
Herbert Graeber -
jdd -
Randall R Schulz -
Steffen Winterfeldt