Windows Partitioning

[TGS]

Partitioned Disk Performance

Text


Link for below (fixed)

Stolen from Microsoft:

Simple management of single disk partitions. Feedback from customers and system manufacturers indicates that users want single volumes because they do not want to manage multiple volumes and do not understand multiple-volume usage scenarios.


Disk Size/Partition Size Performance
Hard-disk performance benchmarks comparing Windows 2000 and Windows XP (after beta 2) show better performance under Windows XP for both large disks and large partitions.

For example, comparing performance under Windows 2000 for a 75 GB hard disk partitioned with multiple smaller volumes versus a single volume shows that disk performance drops by about five to ten percent for the single large volume.

However, when comparing disk performance for the same small versus single large volume configuration under Windows XP, performance drops by only one to two percent for the single large volume.

This improved performance results from the combination of performance improvements described earlier, plus the improved ability under Windows XP to manage large disks and large partitions.

DB partitioning Though mostly related to how data is segmented.


Link for below

Stolen from Microsoft:

Partition your disks so that your boot files and operating system files are on separate partitions. If you partition your disks in this way, you increase your recovery options. For example, you may be able to start to a command prompt, and then modify or replace any damaged boot files that may be preventing you from starting Microsoft Windows 2000.

Link for below

Stolen from Microsoft:

Consider Database Partitioning at Design Time
If your application uses a very large database and you anticipate an I/O bottleneck, ensure that you design for database partitioning up front. Moving to a partitioned database later usually results in a significant amount of costly rework and often a complete database redesign.

Partitioning provides several benefits:

The ability to restrict queries to a single partition, thereby limiting the resource usage to only a fraction of the data.
The ability to engage multiple partitions, thereby getting more parallelism and superior performance because you can have more disks working to retrieve your data.
Be aware that in some situations, multiple partitions may not be appropriate and could have a negative impact. For example, some operations that use multiple disks could be performed more efficiently with concentrated data. So, when you partition, consider the benefits together with alternate approaches.

From what I see partitioning a disk is the performance oriented way to go. Even the transition from Windows 2000 to XP still did not equalize the performance penalties from multiple small partitions and a single large partition.

If someone could find any more information on single large performance I would appreciate it.

(edited for link correction)

 

[gsellis]

You might want to reread some of that. When they say partition, they meant move the data across multiple disks, not partitions on a disk. The W2K example was allowing the OS to be installed twice on C: for recovery (this was before Window PE was available or an installable recovery console like XP and W2K3 allow). But, they do reference using disks for each major component and using multiple processors to pull it.

And although Mr. Tullough has interesting advise, it seems born from old habits. Remember that he is an enthusiast volunteer, and not one of the OS guys at MS. And as you quoted, XP improved NTFS performance, but there is missing data there to make any conclusions. This may only refer to read performance inside the volume depending on size and not read performance in total across all the partitions if the data is scattered.

 

[TGS]

That's why I asked if anyone had performance metrics on single large volumes versus, smaller partitioned volumes. I know that some of it mentions partitions across physical devices, that's why I mentioned it. Also mentioning that segmenting data in that fashion can lead to performance degradation, trying to give both sides of the coin. Though I'm having difficultly finding documentation on how utilizing a single volume would negatively or positively impact overall volume performance.

As to the methods of how they determined how much faster partitioned compared to single volumes was not giving. Though they clearly outline a percentage of performance drop from going with the latter. Mechanisms which were obviously improved by using XP over 2K.

Links from Tulloch again NTFS Hacks

Stolen From the above link:

4. Use More Partitions
In Windows 2000, when you partition a large disk (50GB or more, say) into several smaller NTFS volumes (10GB each), you can speed disk performance by up to 10 percent. NTFS on Windows XP has been improved to perform better overall, but you can still squeeze a percent or two of better performance out of a large disk by partitioning it into several smaller volumes.

At the bottom you can see the books he's written on Windows 2000 Administration in a Nutshell, Windows Server 2003 in a Nutshell, and Windows Server Hacks for O'Reilly so I would hardly call him an Ethusiast only source.

 

[KoolDrew]

In Windows 2000, when you partition a large disk (50GB or more, say) into several smaller NTFS volumes (10GB each), you can speed disk performance by up to 10 percent. NTFS on Windows XP has been improved to perform better overall, but you can still squeeze a percent or two of better performance out of a large disk by partitioning it into several smaller volumes.

This just plain wrong. Your OS, applications, and pagefile should be on the same partition. Seperating them will only worsen seek times. The only reason I see to make multiple partitions is when you have a lot of data that is acessed infrequently. For example, I have a lot of MP3's and by putting them in their own partition the heads stay in the other partition most of the time. This helps seek times within that partition. It also has other benefits such as if I do need to reinstall the OS I still have all my MP3's on my other partition. However, if I had another drive it would be even better to just put my Mp3's on the second drive and not use multiple partitions at all.

Here is a discussion at Ars about the benefits of partitioning:
http://episteme.arstechnica.com/groupee...ms/a/tpc/f/99609816/m/997008565731/p/1

 

[TGS]

Assuming your are there as brutaldrew, you give the same opinion as here. The only person to claim any performance metrics states that his friends partitioned system outperformed his single partitioned disk by 10-15%

Stolen from Ars's FatFreddy:

My coleague and I have identical systems, identical disk, processor, memory, OS and service packs.

My disk is a one partition mess. Yes, I actually prefer to use one partition, but I'll defend the rights of other people to partition as they see fit.

His system has a 10GB partition (C for Windows XP, a 20GB partition which he uses for archiving projects that are put on hold for a while (D, and a 30GB partition with his programs and data (E.

We spent most of yesterday running tests to see how the systems performed and why one should perform better than the other. The result, when treating large data files, his system was between 10% and 15% faster.

To see why, we analyzed the systems with DiskMon and FileMon from SystemInternals. The principle reason for the difference is simply that he has his Pagefile and Temp files on the partition with the Programs and Data on the E: Drive. As the program loads, you will see a flurry of activity on the C: Drive as the program and multiple DLLs are loaded. Once the program is loaded, and the data handling starts, the number of accesses to the C: drive is minimal, about 2% of the accesses. So, he has in effect limited 98% of the accesses to a region which is 50% of his disk. Lower head movements, and hence greater performance.

There is no one size fits all solution. The nature of your work, or play for that matter, has a bearing on how you might want to partition a disk.

For the vast majority of home users, seek times and performance are not the issue. You don't need blinding performance to type a letter to a friend, surf the web, or send a few e-mails.

The OP didn't say what he was using his system for. But the fact that he had built it himself does seem to imply that it is more likely to be a home PC than a work PC.

When it comes down to it, if performance isn't your major concern, then personal choice is a valid reason to partition as you wish. And for those who say that it is just plain wrong, what are your credentials? At least I quote from someone who has a long career in Windows, in pointing you at the article:

http://www.microsoft.com/windowsxp/using/setup/expert/tulloch_partition.mspx

 

[gsellis]

I actually have built systems that are 100% alike (lots of them). Performance can vary and that sample proves nothing (sample of 1). Just how the bits went down, how stuff is on the disks, etc., could explain that away. I have serious doubts that two guys, who each have the same components, have built machines that are exactly alike, service for service, size for size, etc.

The physics of how the drive seeks across multiple partitions, what is in cache, and what is sought, where, dictate how it responds. Since the system has to seek from 0 where the partition is, then seek the file tables, then seek the files as it jumps back and forth, mean that mulitple partitions is always going to take longer.

 

[TGS]

Originally posted by: gsellis
I actually have built systems that are 100% alike (lots of them). Performance can vary and that sample proves nothing (sample of 1). Just how the bits went down, how stuff is on the disks, etc., could explain that away. I have serious doubts that two guys, who each have the same components, have built machines that are exactly alike, service for service, size for size, etc.

The physics of how the drive seeks across multiple partitions, what is in cache, and what is sought, where, dictate how it responds. Since the system has to seek from 0 where the partition is, then seek the file tables, then seek the files as it jumps back and forth, mean that mulitple partitions is always going to take longer.

That's why I'm asking, but so far a Windows Expert and a forum tester have come to the same conclusion. With the workings behind the results. Localized seeks against smaller disk area, opposed to "wider" seeks against the entire disk. I understand the basis of what you are saying but it doesn't make sense as with a large volume you will always have more disk area to seek against due to the nature of how data is typical placed on the platters. There are no disk functions that require seeks against multiple partitions per alternating IO. If that was the case partitions would cause a significant performance drop. The case is that while you localized seeks in one partition seek to another parition, and have similiar performance due to the fact after the seek between partitions data sets will be very close as the partitions should cause the data to be laided closer than a large partition would. It's the arguement about moving like data sets near each other. My OS files on a small partition will never be mixed with rarely accessed vob files. On a large partition this is completely possible. With the increase of data sets across the volume, as most disk drives will do over time you are forcing more seek time due to the fact you have to seek across larger data sets. With the partitions, you force data to be located in fairly specific disk areas. I'm guessing the reason that XP nearly eliminates the penalties for using a single partitioned area, is that they now move data to a more logical location based on use. I'm guessing that the most accessed files are shifted to the outer edge of the disk while, everything else is slowly shifted towards the inner portions. In a fairly logical format, this is what you do with you partitions. It's more overhead to the admin, but in the end it's not a performance impact.

Edit: Also the relocating of files for XP only works for the C partition as well. So anything beyond the C partition in a single volume scenario may not even see the improvements made with moving from Windows 2000 to Windows XP.

 

[gsellis]

From the same article you quoted above (the fixed link)

Disk-Layout Optimizations
Microsoft implemented certain disk-layout optimizations in Windows XP. To perform this optimization, during idle time Windows XP moves pages used for booting the system and launching frequently used applications to ensure these files are laid out contiguously on the hard disk. The contiguous disk layout of these pages results in reduced disk seeks and improved disk I/O, contributing to improved boot time and application launch time.

Windows XP does not perform these optimizations across volumes. Therefore, for this optimization to be available to users, the hard disk must be partitioned as a single volume.

In this example, Windows XP reads ~200K of information at system boot time if E, F, and G are NTFS volumes; however, Windows XP reads ~19MB of information at boot if E, F, and G are FAT32 volumes. Implementation of an NTFS volume leaves more system memory available for user applications.

Don't use FAT32 (recommended by the server author)

Last line in the article

Call to Action

Microsoft strongly encourages system manufacturers to manufacture single NTFS volumes on all systems where a 32-bit version of Windows XP is preinstalled, using the tools described in this article

The whole summary of the document argues against your very point, that multiple partitions are better. MS is trying to convince you that the opposite is true.

Link from above

 

[TGS]

I've already addressed these issues.

Originally posted by: gsellis
From the same article you quoted above (the fixed link)

Disk-Layout Optimizations
Microsoft implemented certain disk-layout optimizations in Windows XP. To perform this optimization, during idle time Windows XP moves pages used for booting the system and launching frequently used applications to ensure these files are laid out contiguously on the hard disk. The contiguous disk layout of these pages results in reduced disk seeks and improved disk I/O, contributing to improved boot time and application launch time.

Windows XP does not perform these optimizations across volumes. Therefore, for this optimization to be available to users, the hard disk must be partitioned as a single volume.

They clearly state disk IO handling for XP systems, has improved. One of those improvements is the relocating of files on the C partition. Even so it will not perform like a partition disk will. To utilize the functions of the file relocation, based on use it requires the user to have a single partition. This is a limitation of the relocation mechanisms. Not of the disk partitioning.

Edit: Also it really doesn't specify on these "Disk-Layout Optimizations" if it's anything more than a defragmentation. If it's strictly for keeping files contiguous, than it's just an automatic defrag utility during idle time. For which would really only have practical use on home user systems. Typical high IO disk systems, would not have much idle time. Or they would do the neccesary optimization operations prior to the write destaging, or reads from cache. Though that could be an arguement carried over from high end disk systems, not exactly relevant for a single disk discussion.

In this example, Windows XP reads ~200K of information at system boot time if E, F, and G are NTFS volumes; however, Windows XP reads ~19MB of information at boot if E, F, and G are FAT32 volumes. Implementation of an NTFS volume leaves more system memory available for user applications.

Don't use FAT32 (recommended by the server author)

Last line in the article

What does this have to do with the partition discussion?

Call to Action

Microsoft strongly encourages system manufacturers to manufacture single NTFS volumes on all systems where a 32-bit version of Windows XP is preinstalled, using the tools described in this article

The whole summary of the document argues against your very point, that multiple partitions are better. MS is trying to convince you that the opposite is true.

Link from above

If you don't selectively quote, you can see the reason why they recommend this, which is hardly a "Technical" reason for doing so.

NTFS is the recommended file system for computers running the Microsoft Windows XP and Windows Server? 2003 operating systems. Microsoft strongly encourages system manufacturers to manufacture single NTFS volumes on all systems where a 32-bit version of Windows XP is preinstalled.

Benefits for end users. Preinstallation of NTFS offers many end-user benefits related to functionality, security, stability, availability, reliability, and performance:

? Support for large hard drives. Hard-drive vendors expect to deliver drive-size capacities in excess of 127 GB in the near future. Windows XP and Windows 2000 provide native support for NTFS volumes on such large sizes, while a FAT32 volume is supported only for sizes up to 32 GB.

? Support for large file sizes. Under Windows XP and Windows 2000, NTFS supports a maximum file size of up to the disk size, while FAT32 supports a maximum file size of only 4 GB. Windows XP delivers new features (such as support for acquiring and editing video files) that frequently result in creation of files that exceed 4 GB in size.

? Simple management of single disk partitions. Feedback from customers and system manufacturers indicates that users want single volumes because they do not want to manage multiple volumes and do not understand multiple-volume usage scenarios.

? Improved performance. Feedback from customers and manufacturers indicates that users want systems optimized for general performance and boot times. New file system features, boot time improvements, and other performance improvements in Windows XP have been implemented only for NTFS, as described later in this article.

Like I've said, I've gone over these issues and we are going in circles if we are going to selectively quote articles.

 

[gsellis]

Who are you going to believe? A guy who writes on server administration over the MS Hardware Design group? My bias is to the design group as that is where my experience is.

I too, have had to use that argument for folks that insisted on writing 'data' to another partition. The whole reason to have servers was a central location, that was centrally managed and backed up was far superior to a 'data' directory AND users more easily understand single partitions. Just because it is used as an argument does not invalidate any of the other information.

 

[TGS]

Again how does the end-user ease of use play a factor in disk performance? Down that article is states there is a performance penalty for using a single partition disk, in both Windows 2000 and XP.

I'm not saying that it's not easier for an end user to understand C is my data, D is my CDRom. I'm saying that there is a performance penalty by doing so. All those sources say the same thing. A performance penalty is caused when using single partitioned volumes. Irregardless of what the end-user experience is, I would like to discuss the performance aspect. The end-user can give two squats about disk performance as long as overall performance does not appear to be slowed by daily operations. From that article they are just putting it nicely for system integrators to KISS, as it will lead to less confusion on the end-users part. That is not a technical arguement.

NTFS can lead to decreased support calls. The temptation for the manufacturer may be to preinstall with FAT32 and let the end user choose to convert to NTFS. However, typical users do not understand or may be unaware of the issues for such decisions. Microsoft encourages manufacturers to deliver systems with NTFS volumes to ensure that users receive the most secure, reliable, and stable Windows installation possible. The expected outcome of preinstalling NTFS for manufacturers will be decreased support calls for both the OEM and Microsoft.

Just so we can stop beating the horse, I want to gather information on the performance aspects not on the end-user experience.

 

[gsellis]

And elsewhere in the article is says there are performance gains. They also say Post Beta2 (the article is posted after SP1 release). I think we have even had a NTFS update since then in XP.

If it is driving you crazy, go to Sysinternals and post a message to Mark Russinovich. Mark will let you know in detail.

 

[TGS]

No, it's stating the performance has gone up on Windows XP compared to the performance drop in 2000. There is still an outlined performance drop. I outlined that in the first post. It's not driving me crazy at all. I confident that the sources I've quoted are 100% correct. I'm just wondering why the use of generalities and selective quotes is being done to say otherwise? Using the same sources that I'm linking off no less.

Tulloch has that 2005 article on how the performance drop still exists in the current implementation of NTFS under XP. The one being linked against the most is a 2003 article. I believe all those sources, I also believe MS would tell system integrators to use single partitions to avoid customer confusion.

 

[gsellis]

Originally posted by: TGS
No, it's stating the performance has gone up on Windows XP compared to the performance drop in 2000. There is still an outlined performance drop. I outlined that in the first post. It's not driving me crazy at all. I confident that the sources I've quoted are 100% correct. I'm just wondering why the use of generalities and selective quotes is being done to say otherwise? Using the same sources that I'm linking off no less.

Tulloch has that 2005 article on how the performance drop still exists in the current implementation of NTFS under XP. The one being linked against the most is a 2003 article. I believe all those sources, I also believe MS would tell system integrators to use single partitions to avoid customer confusion.

And I am confident that your sources are mistaken. But hey, I have only been supporting Windows OS's since 1994, so I don't know anything. Maybe if I had written that book.

 

[Monkey muppet]

each partition would have their own file allocation table which would be smaller on partitions within the same drive compartivly to larger table. - wrong/right??

 

[TGS]

Text