Radified Community Forums - Classic Ghost fact-checking session: anyone bored enough to join in?

Dan Goodell

Special Guest

Offline

Posts: 552
N California

Back to top

Re: Classic Ghost fact-checking session: anyone bored enough to join in?
Reply #15 - Mar 1^st, 2010 at 2:18am

Vista is the first Windows operation system to abandon internal CHS addressing of disks.It uses LBA sector addressing internally instead.

I think this is misleading. XP, Win2K, and even Windows 98 also used LBA addressing internally. And if you use LBA addressing, any correlation with CHS is purely incidental. In fact, none of these OS's requires CHS-alignment.

Just to prove that, I tested by restoring a Win98 image to a totally non-aligned partition. I manually created a partition that began at CHS 0/32/35 (LBA 2050), with the partition's beginning and ending points neither CHS-aligned nor 2048-aligned. I then used Ghost 2003 to restore an old Win98se image to this wacky partition. Win98 booted and ran just fine. (For those reading between the lines . . . yes, Ghost 2003 has no quarrel with non-standard partitions!)

Similarly, I have had no trouble restoring 2K/XP images to 2048-aligned partitions, nor any trouble restoring Vista/Win7 images to CHS-aligned partitions. Vista/Win7 don't just tolerate CHS-aligned partitions, they accept them just as naturally as any other partition. Because all these OS's use LBA addressing, the only thing they care about is which LBA sector the partition starts from--and that can be anywhere.

To put things in perspective, I think it helps to make a distinction between the operating system and the disk partitioning utility. The utility Microsoft tossed in with Win98 or XP--whether it be fdisk or diskpart--was designed to create CHS-aligned partitions, regardless of what OS you were planning on using them with. The utility shipped with Vista and Win7 is designed to create 2048-aligned partitions, likewise regardless of what OS they're to be used with.

When Microsoft made this change, abandoning CHS, it was forced to decide (arbitrarily) on a new number for how many sectors would represent "boundaries" for partitions.As it turns out, they settled on 2048.2048 because at 512 bytes per sector, 512 bytes * 2048 sectors = 1,048,576 bytes, or exactly 1 megabyte.

The reason Microsoft switched to 2048-alignment wasn't arbitrary. There's a white paper somewhere on Microsoft's website that details the reasoning, but in a nutshell it was predicated by the impending shift to 4KB sectors (vs. the current 512-byte sectors) on supra-terabyte hard disks, as well as a way to effectuate more efficient disk transfers. Disk controllers can read or write large blocks of data more efficiently if those blocks line up on boundaries that are related to the size of the blocks. For the sake of efficiency, Microsoft wants to shift not only the alignment of partition beginning and ending points, but I've heard the latest NTFS spec also shifted the location of the MFT and swap file within the partition for more efficient disk transfers, even if you use CHS-aligned partitions.

Note that these reasons really had nothing to do with Vista, per se. It's all about hardware transfer rates. Vista just happened to be in the right place at the right time--the next OS to ship after Microsoft made the switch to this new disk strategy. Hence, I refuse to call these "Vista partitions". (To me, a "Vista partition" is the partition the Vista OS is installed on.) 2048-aligned partitions are no more necessary to Vista than CHS-aligned partitions--and work equally well with XP as they do with Vista.

I suppose you could just as easily call the non-first-sector areas of EBR/MBR boot tracks "MBR/EBR slack space" or "the MBR/EBR demilitarized zone" if you wanted. Personally, I've adopted the term "boot track" because not only is it apparently the most popular name for that area . . .

I guess that depends on who you hang out with. I've rarely heard it called a "boot track". I've been snooping and tinkering inside disk sectors for over 25 years, and the most common name I've heard is "Track 0". That, of course, is a direct holdover from CHS terminology (where "track" and "head" are synonymous), even though the concept lost any correlation to reality once IDE drives and translating controllers hit the market 20 years ago.

"MBR slack space" . . . I like that! That's the best term I've heard for it.

IP Logged

pishposh

Dude

Offline

I Love Radified!

Posts: 24
Cambria, California

Back to top

Re: Classic Ghost fact-checking session: anyone bored enough to join in?
Reply #16 - Mar 1^st, 2010 at 11:47am

Quote:

Vista is the first Windows operation system to abandon internal CHS addressing of disks.It uses LBA sector addressing internally instead.

I think this is misleading. XP, Win2K, and even Windows 98 also used LBA addressing internally. And if you use LBA addressing, any correlation with CHS is purely incidental. In fact, none of these OS's requires CHS-alignment.

Just to prove that, I tested by restoring a Win98 image to a totally non-aligned partition. I manually created a partition that began at CHS 0/32/35 (LBA 2050), with the partition's beginning and ending points neither CHS-aligned nor 2048-aligned. I then used Ghost 2003 to restore an old Win98se image to this wacky partition. Win98 booted and ran just fine. (For those reading between the lines . . . yes, Ghost 2003 has no quarrel with non-standard partitions!)

Similarly, I have had no trouble restoring 2K/XP images to 2048-aligned partitions, nor any trouble restoring Vista/Win7 images to CHS-aligned partitions. Vista/Win7 don't just tolerate CHS-aligned partitions, they accept them just as naturally as any other partition. Because all these OS's use LBA addressing, the only thing they care about is which LBA sector the partition starts from--and that can be anywhere.

To put things in perspective, I think it helps to make a distinction between the operating system and the disk partitioning utility. The utility Microsoft tossed in with Win98 or XP--whether it be fdisk or diskpart--was designed to create CHS-aligned partitions, regardless of what OS you were planning on using them with. The utility shipped with Vista and Win7 is designed to create 2048-aligned partitions, likewise regardless of what OS they're to be used with.

Okay. This is news to me because I was under the impression that the MBR, VBR, and IO.SYS/NTLDR code from those operating systems were hanging on to older CHS constraints.

Just out of curiosity, did you check the off-kilter test partition you created after restoring the W98 image to it? To make certain Ghost didn't alter its boundaries to cylinder boundaries? Ghost is not supposed to change an existing partition's boundaries when restoring a partition image to it; however I've never tested that claim against a destination partition whose boundaries are already "misaligned" (to verify Ghost doesn't try to "correct" them). If you did, and W98 still worked, then I've learned something new today -- so thanks. (Just finished reading http://radified.com/cgi-bin/yabb2/YaBB.pl?num=1265301270 and http://www.anandtech.com/storage/showdoc.aspx?i=3691&p=1).

P.S. Anandtech's article describes Western Digital's two solutions for using non-4K-aware OS'es with 4K HDDs, one of which is its partition re-alignment tool. Based on your knowledge of this topic, couldn't one just start with an unpartitioned 4K drive, use a Vista/7 install CD partitioning tool to create one's partition(s), and then install an OS like XP on that drive, for exactly the same optimum performance results?

Also, this made me realize: if you have in deed verified Ghost 2003 doesn't re-align partition boundaries when restoring partition images to existing "unaligned" partitions, this would make Ghost 2003 compatible with 4K drives provided one only dealt in partition imaging and only restored partition images to existing partitions created by 4K-aware partitioning tools.

Quote:

Quote:

When Microsoft made this change, abandoning CHS, it was forced to decide (arbitrarily) on a new number for how many sectors would represent "boundaries" for partitions.As it turns out, they settled on 2048.2048 because at 512 bytes per sector, 512 bytes * 2048 sectors = 1,048,576 bytes, or exactly 1 megabyte.

The reason Microsoft switched to 2048-alignment wasn't arbitrary. There's a white paper somewhere on Microsoft's website that details the reasoning, but in a nutshell it was predicated by the impending shift to 4KB sectors (vs. the current 512-byte sectors) on supra-terabyte hard disks, as well as a way to effectuate more efficient disk transfers. Disk controllers can read or write large blocks of data more efficiently if those blocks line up on boundaries that are related to the size of the blocks. For the sake of efficiency, Microsoft wants to shift not only the alignment of partition beginning and ending points, but I've heard the latest NTFS spec also shifted the location of the MFT and swap file within the partition for more efficient disk transfers, even if you use CHS-aligned partitions.

One thing I don't understand. If the new megabyte alignment scheme (starting partition 1 at LBA sector 2048) is based on upcoming 4K sector HDDs, why did they choose sector 2048 instead of 64? 512*64/4096=8 4K sectors exactly. Yet they chose 2048, throwing almost a megabyte away. (With 4K sectors, 2048 sectors throws 8 MB away!) Could it be we're both correct, then? That 2048 is based on 4K sectors and simplifying partition sizing to increments of exactly 1 MB?

Quote:

Note that these reasons really had nothing to do with Vista, per se. It's all about hardware transfer rates. Vista just happened to be in the right place at the right time--the next OS to ship after Microsoft made the switch to this new disk strategy. Hence, I refuse to call these "Vista partitions". (To me, a "Vista partition" is the partition the Vista OS is installed on.) 2048-aligned partitions are no more necessary to Vista than CHS-aligned partitions--and work equally well with XP as they do with Vista.

So noted. Also, the terminology I'm seeing is "megabyte alignment". So, there's no need for those people calling it "Vista partitioning" for a want of a better term. One either has CHS/cylinder alignment (63) or megabyte alignment (2048).

P.S. So noted re: "track 0" instead of "boot track".

IP Logged

pishposh

Dude

Offline

I Love Radified!

Posts: 24
Cambria, California

Back to top

Re: Classic Ghost fact-checking session: anyone bored enough to join in?
Reply #17 - Mar 1^st, 2010 at 11:48am

Incidentally:

As far as Ghost 11.5.1 (and maybe earlier versions of 11 -- haven't checked), it is in deed compatible with 4K sector HDDs. From Symantec's Ghost Imaging Foundation document:

-align=chs, -align=1mb - Lets you override the way in which the partitions are aligned when an individual partition or disk full of partitions is restored. This switch aligns the partition to the boundary as follows: CHS - Aligns to a track or cylinder boundary; 1MB - Aligns with a boundary of 1 MB. By default, a partition is aligned on the destination computer as it was on the source computer. Note: The 1MB alignment option supports Windows Vista.

IP Logged

Dan Goodell

Special Guest

Offline

Posts: 552
N California

Back to top

Re: Classic Ghost fact-checking session: anyone bored enough to join in?
Reply #18 - Mar 2^nd, 2010 at 4:21am

did you check the off-kilter test partition you created after restoring the W98 image to it? To make certain Ghost didn't alter its boundaries to cylinder boundaries? Ghost is not supposed to change an existing partition's boundaries when restoring a partition image to it; however I've never tested that claim against a destination partition whose boundaries are already "misaligned" (to verify Ghost doesn't try to "correct" them).

Yes, I verified the boundaries of my wacky partition were not altered.

IMHE, I've never seen Ghost 2003 try to change the boundaries when filling an existing partition with the contents from an image. Keep in mind that Ghost 2003 could not have known about 2048-aligned partitions, either--they should be as foreign to Ghost 2003 as my wacky, handmade partition. Yet I have been using Ghost 2003 to image or restore 2048-aligned partitions at will since the advent of Vista, and Ghost has never tried to "correct" any of those partitions, either.

couldn't one just start with an unpartitioned 4K drive, use a Vista/7 install CD partitioning tool to create one's partition(s), and then install an OS like XP on that drive, for exactly the same optimum performance results?

In theory, that ought to work . . . except you need to allow that the installer program might have its own requirements, which need not necessarily be the same as the OS itself. That's why for my test I chose to restore from an image instead of installing from a CD--to make sure I was testing the OS's requirements instead of the installer's requirements. (That, and because it was also quicker.)

(Aside: over the years I've become all too familiar with this distinction from working with various other programs. For instance, a client has a cherished program that was made in the Win98 era. It runs perfectly well in Win2K or XP, if you can get it installed--the setup routine refuses to run unless it sees Win98. It's not a problem with the program, it's a problem with the installer.)

I fully expect a Win98 installation CD would complain about non-CHS aligned partitions. In fact, Microsoft KB931760 suggests the XP installer has a similar limitation. But I would emphasize that the installer and the OS are two different things, and I think my experiment proves that.

this would make Ghost 2003 compatible with 4K drives provided one only dealt in partition imaging and only restored partition images to existing partitions created by 4K-aware partitioning tools.

I think that's probably true, but until I've had a chance to start tinkering with 4K drives I wouldn't want to say for sure.

(Setting aside, for the moment, any argument over the definition of "compatible". For instance, Ghost 2003 will image/restore a Win7 OS, but since it doesn't know about BCD it can't edit it. A BCD edit isn't always necessary, but because it might be in certain circumstances, would one then say Ghost 2003 isn't compatible with Win7? The literalness of that has been debated in other threads elsewhere on this forum.)

If the new megabyte alignment scheme (starting partition 1 at LBA sector 2048) is based on upcoming 4K sector HDDs, why did they choose sector 2048 instead of 64? 512*64/4096=8 4K sectors exactly.

You need to see the bigger picture. The performance gains aren't when you're transferring a few sectors at a time, Microsoft wants a standard that holds up well when transferring huge gobs of data. Think in terms of 4MB disk caches, or RAID-5 arrays, for example. To quote Microsoft, "A starting offset of 2,048 sectors covers most stripe unit size scenarios."

"Megabyte-aligned" . . . another good, descriptive term!

IP Logged

pishposh

Dude

Offline

I Love Radified!

Posts: 24
Cambria, California

Back to top

Re: Classic Ghost fact-checking session: anyone bored enough to join in?
Reply #19 - Mar 2^nd, 2010 at 6:40am

Quote:

did you check the off-kilter test partition you created after restoring the W98 image to it? To make certain Ghost didn't alter its boundaries to cylinder boundaries? Ghost is not supposed to change an existing partition's boundaries when restoring a partition image to it; however I've never tested that claim against a destination partition whose boundaries are already "misaligned" (to verify Ghost doesn't try to "correct" them).

Yes, I verified the boundaries of my wacky partition were not altered.

Thanks, good to know that for sure.

Quote:

IMHE, I've never seen Ghost 2003 try to change the boundaries when filling an existing partition with the contents from an image. Keep in mind that Ghost 2003 could not have known about 2048-aligned partitions, either--they should be as foreign to Ghost 2003 as my wacky, handmade partition. Yet I have been using Ghost 2003 to image or restore 2048-aligned partitions at will since the advent of Vista, and Ghost has never tried to "correct" any of those partitions, either.

Right, and I just wanted to verify that Ghost's advertised normal behavior (not re-aligning existing partitions when restoring partition images) held true when those existing partitions weren't cylinder-aligned. (You never know: it's quite possible Ghost's authors could have advertised Ghost as not re-aligning existing partitions when restoring partition images, but coded it to do so anyway if and only if the existing destination partitions weren't cylinder-aligned, as a "silent fix-up".)

Quote:

Quote:

couldn't one just start with an unpartitioned 4K drive, use a Vista/7 install CD partitioning tool to create one's partition(s), and then install an OS like XP on that drive, for exactly the same optimum performance results?

In theory, that ought to work . . . except you need to allow that the installer program might have its own requirements, which need not necessarily be the same as the OS itself. That's why for my test I chose to restore from an image instead of installing from a CD--to make sure I was testing the OS's requirements instead of the installer's requirements. (That, and because it was also quicker.)

Understood. Well, what I was primarily pondering was whether a Vista/7 install CD's partitioning tool could be used as a substitute for ever having to use Western Digital's (or another manufacturer's) partition alignment tool if one has a "4K" drive. E.g., (a) buy new 4K HDD, (b) partition with Vista or Win7 install CD, and (c) install XP without re-partitioning (see below re: KB931760). Then, if one has Ghost 2003, they can avoid those 4K/megabyte-aligned partitions ever being de-aligned by simply always doing "Partition -> to/from (Image|Partition)" imaging (whether imaging one or many or all partitions at once) rather than "Disk -> to/from (Image|Disk)". (Of course there would then be the issue of using Disk Management to re-size partitions, which could de-align them from 4K/megabyte boundaries; but 4K/megabyte-alignment aware third-party partition resizing tools would be the solution there.)

Quote:

I fully expect a Win98 installation CD would complain about non-CHS aligned partitions. In fact, Microsoft KB931760 suggests the XP installer has a similar limitation. But I would emphasize that the installer and the OS are two different things, and I think my experiment proves that.

Right. Hmm, well, according to http://support.microsoft.com/kb/931760, this issue can be bypassed by temporarily making a BIOS change during the XP installation process. Failing that, Microsoft has a hotfix (a new version of Ntoskrnl.exe to slipstream into your XP installation CD), but incredibly -- according to http://ihaveablog.wordpress.com/2008/05/28/microsoft-deployment-and-hyper-v/ -- they've not integrated this hotfix into SP3 (and since Ntoskrnl.exe in KB931760 predates SP3, it would be real unwise to slipstream KB931760 into an SP3 installation CD).

Guess if one's BIOS doesn't allow working around the issue, the next easiest thing (according to my brain at 4:45 AM) would be: (a) buy new 4K HDD, (b) partition and install XP with XP install CD, (c) Ghost all partitions (Partition -> to Image) to .gho file, (d) use Vista/Win7 install CD to delete all partitions and re-create them 4K/megabyte-aligned, (e) restore .gho partition images.

Quote:

Quote:

this would make Ghost 2003 compatible with 4K drives provided one only dealt in partition imaging and only restored partition images to existing partitions created by 4K-aware partitioning tools.

I think that's probably true, but until I've had a chance to start tinkering with 4K drives I wouldn't want to say for sure.

(Setting aside, for the moment, any argument over the definition of "compatible". For instance, Ghost 2003 will image/restore a Win7 OS, but since it doesn't know about BCD it can't edit it. A BCD edit isn't always necessary, but because it might be in certain circumstances, would one then say Ghost 2003 isn't compatible with Win7? The literalness of that has been debated in other threads elsewhere on this forum.)

Quote:

If the new megabyte alignment scheme (starting partition 1 at LBA sector 2048) is based on upcoming 4K sector HDDs, why did they choose sector 2048 instead of 64? 512*64/4096=8 4K sectors exactly.

You need to see the bigger picture. The performance gains aren't when you're transferring a few sectors at a time, Microsoft wants a standard that holds up well when transferring huge gobs of data. Think in terms of 4MB disk caches, or RAID-5 arrays, for example. To quote Microsoft, "A starting offset of 2,048 sectors covers most stripe unit size scenarios."

Okay, makes sense now. Hadn't thought about RAID.

IP Logged

	Welcome, Guest. Please Login