MortySnerd

Partitioning For Speed?

126 posts in this topic

Second, and more importantly, really, Windows keeps system files that are used often in its cache.

Not just system files, but any files.

I was trying to avoid the topic of kernel files...

Share this post


Link to post
Share on other sites

Greets.

Well, that's Executive's theory of disk defragging, and it's not all that bad of one. It goes like this:

Not that it matters, but according to emails I traded with Executive tech support they Don't move the location of files around on the disk for reasons of performance. Thye feel this is a waste of time (which is one reason their defragger is so fast). The only time they "move" files for placement on the disk is during the "boot-time" defragging when they provide the option to place all your folders at the beginning of the disk.

There is never a time when your disk will be any faster because of having multiple partitions, so why do it?

It WILL be faster for the times when my heads don't have to seek over/past rarely-used files (such as Ghost images and other back-ups) to retrieve files I use/access more frequentl (such as system/program/data files). With a single 160G partition, I can't control which files end up where. But if I PARTITION my disk into multiple segments however, I can prevent my heads from having to seek to the slow end of the disk to retrieve these more-frequently-accessed files because I confine them to the faster part of my disk.

Not sure why you're having such a problem with this. Disk performance factors aren't rocket science. The beginning is faster, the end is slower. If you confine your more-highly-used files to the fast end, your disk will be able to retrieve them more quickly. I'm talking about normal system usage.

The only time the disk will take more time is when I'm reading from or writing to the slow part of the disk, which (in my case) rarely. I make a back-up Ghost image once a month, which takes 5 mintes. For the other 23 hours and 55 minutes of that day, and the other 29 days that month, the heads have no reason to seek to the slow end of the drive for those Ghost images.

If you only have a single large partition, your heads will have to seek over rarely used files to retrieve more-frequently used files because you have no control over where they wind up on the disk.

Method A:

Frequently-used files (O/S, apps) > fast part of disk.

Rarely-used files (images, back-ups, etc) > slow part of disk.

Method B:

All files mixed together. You can't prevent some rarely-used files from ending up at the fast end of the disk, which will cause your heads to seek OVER them to the slow end of the disk to retrieve more frequently used.

So for the times the heads don't seek to the slow end of the disk (in the case of Method A, this will be the vast majority of the time), Method A will outperform Method B.

Share this post


Link to post
Share on other sites

Hi guys.. Kinda OT here, but how about would you go if you had more than one HD?

In this case I take it that there is less need for partitioning, but how about would you maximise performance (and maybe in the process organise things nicely).

Basically I am not too sure what how to best use a 74GB Raptor, 120GB WD1200JB and 160GB 7k250 for various applications (OS/App/Games/Downloads/Archive/Project work).

Share this post


Link to post
Share on other sites

One benefit I have seen to using a separate small partition for your swapfile and your internet cache files as well as your user and system environment variable data is that it it greatly reduces disk fragmentation. Just the small partition containing that data can be defragmented very quickl with a hugely reduced need for your other partitions to be defragmented.

Share this post


Link to post
Share on other sites
I think you mean "multiple partitions" because you can't use a hard drive without partitioning it.

The voice of somebody who has never used a dynamic volume...

Share this post


Link to post
Share on other sites
I think you mean "multiple partitions" because you can't use a hard drive without partitioning it.

The voice of somebody who has never used a dynamic volume...

(FWIW, a "dynamic volume" under NT is a type of partition, just not a "basic"/"MBR" partition. In fact, using a 3rd-party utility like Partition Magic, "basic" partitions can be just as "dynamic", in terms of sizing, as NT-native "dynamic" volumes/partions can be.)

To address the primary topic of this thread:

I am personally of the opinion that partitioning a HD can in fact improve real-world performance. rfarris objected to this, but I (correct me if I'm wrong here) am going to guess that he was only speaking of "raw" performance. Of course, adding logical dividers between your data isn't going to change any of the intrinsic performance properties of the device itself, such as access time and STR.

But there is a very valuable property that partitioning can have, at least under DOS/Windows' OSes. Since these OSes lack the ability to impose preferred filesystem-location policies for files within the filesystem, a workaround must be used. Creating partitions is that workaround. Considering how easily both FAT32 and NTFS fragment (FAT especially), and that there are real-world performance reductions that can be observed due to file fragmentation, then it stands to logically reason that by mitigating the effects of fragmentation, that one can also obtain a real-world performance increase, relative to the alternate case.

In fact, *nix best-practices actually explicitly recognizes this fact, by dividing up several key areas of the traditional *nix filesystem layout, which are generally symlinked to seperate storage volumes, based primarily on their likelyhood or tendency towards fragmentation, and therefore mitigating their otherwise detrimental effects on day-to-day real-world performance.

(Specifically, directories like /, /var, /temp, /log, /home, /etc, /usr, etc. I'm not the biggest *nix guru so I can't say exactly which are supposed to go where, but I know that most systems recommend at least three seperate filesystem volumes, one for the OS/boot/root, one for /home, and one for other easily-fragmentable data like /var and /temp.)

So can anyone tell me why *nix best-practices, are not useful nor applicable to Windows' best-practices? I think that explaining that fully, is key to defending the "one partition" viewpoint. Considering the long timeline that these best-practices have evolved along, I think that any explaination would be quite interesting to hear.

Share this post


Link to post
Share on other sites
But there is a very valuable property that partitioning can have, at least under DOS/Windows' OSes. Since these OSes lack the ability to impose preferred filesystem-location policies for files within the filesystem, a workaround must be used. Creating partitions is that workaround. Considering how easily both FAT32 and NTFS fragment (FAT especially), and that there are real-world performance reductions that can be observed due to file fragmentation, then it stands to logically reason that by mitigating the effects of fragmentation, that one can also obtain a real-world performance increase, relative to the alternate case.

If the only problem being solved by partitioning is fragmentation, then a far better solution would just be to get a good third-party defragmenter, such as Diskeeper, which defragments automatically, fast, regularly, and well. That is a far better solution to fragmentation than an inflexible division of a hard drive's free space.

Share this post


Link to post
Share on other sites
(Specifically, directories like /, /var, /temp, /log, /home, /etc, /usr, etc. I'm not the biggest *nix guru so I can't say exactly which are supposed to go where, but I know that most systems recommend at least three seperate filesystem volumes, one for the OS/boot/root, one for /home, and one for other easily-fragmentable data like /var and /temp.)

So can anyone tell me why *nix best-practices, are not useful nor applicable to Windows' best-practices? I think that explaining that fully, is key to defending the "one partition" viewpoint. Considering the long timeline that these best-practices have evolved along, I think that any explaination would be quite interesting to hear.

The quantic move in windows is when you separate OS/Apps from data. You can create more partitions, of course, but it will look more like micro-optimisations. The gain is not as important as this first separation.

Windows doesn't have to be partitionned as *nix is as long as it is not used as *nix. If you use your *nix box like a workstation, it may be smarter to partition it the same way you would with a Windows box (one partition for OS/Apps, another for data).

What really matters is how you use your box. You don't want to partition a workstation with only one user (at least at the same time) the same way you'd partition a server granting access to dozens (or more) of users simultaneously. The OS running is not far from being irrelevent, as all modern OSes relie on (mostly) the same technologies inside.

Share this post


Link to post
Share on other sites
I think you mean "multiple partitions" because you can't use a hard drive without partitioning it.

(What some people are having trouble distinguishing is the difference between the use of the word "partition" meaning to "divide into parts" as opposed to the use meaning "run a utility program to create a 'partition'." Assuming there are no idiots here, the usage should be obvious by context.)

So can anyone tell me why *nix best-practices, are not useful nor applicable to Windows' best-practices? I think that explaining that fully, is key to defending the "one partition" viewpoint. Considering the long timeline that these best-practices have evolved along, I think that any explaination would be quite interesting to hear.

Nobody ever said that -- I'm with you 100%, as long as those partitions are on separate spindles! The *nix guys (and I used to be one) were the original bunch-o-disks guys. Plus, as fragile as the Unix FS' used to be, even if there was only one disk, it was popular to create separate partitions for the OS, the site-installed program files, the user tree, etc, so that if you lost a file system you had a boundary on the amount of work you had to do to get the machine up again. But separate partitions on one disk was never done for performance reasons. For performance you create separate partitions on separate disks. Personally I have a Raptor named "Fragment" with nothing on it but a pagefile, the system temp directory, and Documents & Settings.

Back to the fragmentation discussion: If I thought (which I don't) that frequently accessed system files were slowing down my system because they were migrating outward on the disk, I'd acquire a disk defragger that let me have great control over where files were placed.

-- Rick

Share this post


Link to post
Share on other sites

seems the disagreement lies with the concept that effectively 'short-stroking' the drive by creating small partition (i use 12G) at the beginning of the disk, where both seeks & str's are fastest, will perform better using a drive with a single 160G partition.

i'm talking about the vast majority of system-usuage time. for example, on the rest of my disk, i have mp3s i've never listened to, photos i've never looked at. i guess you could call this mass-storage, and of course there are back-up Ghost images.

but for the vast majority of the time i'm using my system, my heads never have to seek past the first 12G-partition. You don't feel this config is faster than if they had to seek to the opposite end of the disk for O/S and prgm files that wind up there?

My 12G system/apps partition contains ~8G of data, so the drive is roughly 2/3rds full. I use Diskeeper as my defragger of choice. It features a "drive map" that allows me to see which files are located where. On my disk, the files are everywhere, including the end of the partition. In other words, they are not limited to the first 2/3rds of the partition.

Now if this partition were 160G, with all my other files included, there's no telling where system and prgms files would wind up as I update both on an on-going basis.

I am curious. Is the data on your drive limited to the beginning of the drive, or is is spread out through-out the drive? If your drive were 2/3rds full, would the last 1/3rd of your disk be empty?

As a side note here, I know the title of this thread asks about "speed" (aka "performance"). But I don't think anyone partitions their drive for reasons of performance. (Although I *do* feel that a small partition at the beginning of the drive containing o/s & apps WILL performa better.) Rather I think the real reason most people partition their drive with a small partition at the beginning is so they can wipe/reformat the system partition if Windows cr*ps out, without having to lose everything else.

I guess you could back-up all 160Gs of data onto DVDs, but there's no need for this with a small system partition.

Personally, I like the small system partition (with apps) so my Ghost images stay small. And this config allows me to restore an image if something goes wrong with my system.

But I do find it interesting that some people don't feel that a smal system partitionat the beginning of the disk will run a little zippier. One poster mentioned ealier that his PC seemed zippier after partitioning the disk with a small system partition at the beginning.

You agree that the beginning of the drive is the fastest part. Do you simply diagree that the heads cannot be confined to the system partition during normal system usage?

I'm trying to understand where exactly we disagree.

Share this post


Link to post
Share on other sites
Because OS and program files don't fragment, they will remain in just as tight an area of the disk as if you had partitioned them there even if you don't.

The first seek is certainly not a significant factor in the equation, given the lengths of application load times, relative to the length of even a full stroke seek on todays slowest drives.

In reference to OS and program files not fragmenting, you must not update your system. Everytime a system files is modified it is moved. If you are using windows this happens quite a bit. And defraggers generally won't touch these files.

I'm going to say I don't care about boot time optimization here to ignore the optimizations XP does to give you a faster boot time.

In reference to the first seek not being significant, I think you'll find that it is the MOST significant. In a workstation access patern data tends to be localized. So it is the first seek and subsequent settling that adds the most time to data access. After the first seek is done, most of the data will be read sequentially. Seeks after this are minimized away by regular defragmentation, firmware optimizations, and the localized nature of the data.

However is it faster to seek to the outer track or the inner track or the middle tracks? How does this affect settling time? Almost certainly it is easier to find the sector on the outer tracks. Thus in addition to transfer rate improvement you have first seek improvements.

Is the performance difference significant? I don't know.

But I would say that the organizational benefits should be the main reason for partitioning. In this at least we'll agree.

Share this post


Link to post
Share on other sites
There is a concensus among people who know what they're talking about.

In reality I find this to be the opposite.

People who don't know what they are talking about seem to agree. People who do know what they are talking about seem to disagree and argue to bring about a better understanding.

Anyway so how do we go about testing our differing opinions?

Share this post


Link to post
Share on other sites
One poster mentioned ealier that his PC seemed zippier after partitioning the disk with a small system partition at the beginning.

That was me and I also said that it would be hard to convince me that the size of the system partition has no impact on performance. Anyway, some of You guys do Your best to convince me but ...... :rolleyes: ...... not yet!

I too, have seen system files end up at the far end of the drive and I still believe that a smaller system partition, with the far end "closer" must be a better option. I'm not claiming that the difference is huge but nonetheless noticeable.

Keep the discussion going, this thread is interesting!

Christer

Share this post


Link to post
Share on other sites

I am surprised this is continuing. All the information that follows has already been discussed, and, apparently ignored.

So far, the most compelling arguments for partitioning for performance reasons have been delivered by Will. Alanx, if I am reading his recent posts correctly, seems to have taken up his argument as well. You both are asserting that the presence of less used files between your commonly accessed files is reducing the performance of the system if multiple partition are not used to seperate these files. I believe that you have not adequately considered your assertions.

1. Crucial to this argument is that the first seek is "the most important" as Will put it.

However, a superficial examination of commonly available data compels one to infer otherwise. As I noted earlier, even a full stroke seek, on the slowest of contemporary hard drives, is insignificant as regards the load time of small applications. Full stroke seeks on slower 7200RPM units take ~20ms. Loading an application seems to take many times longer than that on most computers I have used. This argument is essential to the thesis of those advocating partitioning to improve performance and the consequences of its inadequacy are far reaching. Because of localization this constituted the position's only genuinely valid argument. See 2 for a discussion of the importance of localization.

The average seek is a more useful metric for comparison, since it measures the time the head takes to move from a random position to another random position. I direct you to SR's performance database for accurate average access times. Deduct the appropriate rotational latency from those measurements to observe the seek time.

2. The overall argument for partitioning for performance revolves around the premise that introducing less commonly accessed data between more commonly accessed data introduces performance-impairing seeks. This argument is not compelling either.

The fundamental flaw with this argument is that it presumes that commonly accessed files all need to be close together. Upon consideration this presumption is almost laughably superficial. Only files that have an association need to be close together. While this should be inherently obvious, I will illustrate this by way of example, largely for emphasis. If 160GB of mp3s seperate Microsoft Office from UT2003, the positional requirements for neither the loading of Microsoft Office, nor Unreal Tournament, are penalized in this arrangement.

As I have noted, files that have an association with each other will be installed together, and be, necessarily, already close together. Partitioning does nothing to improve this.

3. The argument was made earlier that seperating application files from Windows system files will introduce seeks. This is poorly researched. Windows has cached system files for some time. They also, are not paged, so don't worry about that either.

It was asserted that Windows Update fragments system files, and that, since defragmenters apparently hesitate to move these files, not partitioning one's system would result in a degradation of performance due to unnecessarily high seeks between system files. Obviously, since system files are loaded into memory on boot, and stay there, this argument is also invalid.

To continue the rebuttal of the supposed performance advantages of partitioning, I would like to point out a negative performance consquence of the system advocated by Alanx and Will Rickards. The 'infrequently used files' that they refer to can be generally described as data files. They differ from what are considered 'program files' in that they do not contain, in and of themselves, executable code. It is a well known fact that data files consequently require an application to access them usefully. By artifically inflating the seek distance between applications and their respective data one reduces performance on data accesses when the application is not in memory. Opening a video file in WMP requires concurrent accesses to the video file and to the application localization. Obviously, artificially enhancing this seperation will reduce performance under these most common of tasks. Double-clicking a Word file will incur the same penalties under the system advocated by Alanx and Will.

Continuing in this vein I would like to point out a significant discrepancy in the reasoning used to advocate partitioning for performance. While those asserting the utility of such an endevour declared the single initial seek to be a significant performance problem, they ignored the fact that their advice exacerbated multiple long seeks. Requests greater than 64K are divided into 64K pieces. The head will not visit the artifically distant data file, once and quickly read it into memory, but will, instead, seek back and forth between the application and data localizations, greatly exacerbating any seek penalty introduced by excessive partitioning.

I do not intend to demonstrate that this penalty is a tremendous negative, only that it is many times more important than the seeks that drove the individuals, who are advocating partitioning for performance, to partition their data in the first place. Ignoring a more important factor and stressing the relatively less relevant is a well known fallacy. The conclusion that partitioning was a solution to seeking problems was derived from an incomplete analysis.

As for the concensus I implied that has been attracting so much spurious attention I cite the posts by sechs and rfarris in this thread. The posts of both individuals consistently evince that their originators are both knowledgeable as well as thoughtful. They constitute the only other regulars at SR who have submitted an opinion, excepting Will who has also been around for some time. While Will is also represents one of the sources of the wealth of information available on this board, I do believe he is wrong in this particular case. Three out of four, when most regulars would avoid a thread like this like the plague is a pretty solid concensus IMHO. A search of the archives will demonstrate that several other knowledgeable posters advocate partitioning only for non-performance-related reasons. In fact, this whole thread should have been deprecated by the discovery of the search button ;). Oh well.

I believe I have covered the relevant arguments thoroughly. I also believe that anything I missed will be quickly pointed out to me.

Share this post


Link to post
Share on other sites
All the information that follows has already been discussed, and, apparently ignored.

Not ignored. Simply disagreed with. :) We feel the same.

This argument is essential to the thesis of those advocating partitioning to improve performance and the consequences of its inadequacy are far reaching. Because of localization this constituted the position's only genuinely valid argument.

Your language is difficult to understand. I do the same when I try to blow smoke up my boss's arse, especially when I'm trying to defend a weak position. (Baffle him with BS.) So, are you ceding validity to our position/argument?

As I have noted, files that have an association with each other will be installed together, and be, necessarily, already close together.

Here is where you make one of your biggest errors. As PROGRAMS ARE CONTINUALLY UPDATEFD, these files will be SPREAD OUT ACROSS THE ENTIRE 160G DISK, and you have NO WAY to put them all together. You continue to ignore this point. Is this because you never update your program files?

Windows has cached system files for some time.

Some. Not ALL of 'em. How many depends on much RAM you have. Memory space is limited and precious.

The last part of your post discusses at length the accessing of data/storage files (such as a Windows media file) and how bad this is for performance. But we've already stated many times that data/storage files aren't accessed during normal system usage. So your dissertation on this topic is moot.

I have more comments to make, but need to run. I'll be back! (Arnold voice). :)

Share this post


Link to post
Share on other sites

Responses to alanx's post initial response are listed in reverse order.

1. Every critical part of the Windows OS goes into memory and stays there. So, specifically, what files are you talking about?

2. Adding extra partitions does not prevent the fragmentation of updated application files either. No matter how many times you press caps lock (incidentally, the forum offers several less intrusive methods of emphasizing yourself).

As for mitigating the length of these seeks. This is an argument which is without merit without benchmarks considering that you must, necessarily, lengthen other seeks.

3. My language is precise and conforms to the rules of English grammar. My arguments stand on their own regardless. Importantly, it is rude to give lessons on grammar or diction in a public board frequented by individuals from many nations. I will not argue that style of expression is unimportant when presenting an argument, but a criticism of an argument must be based on its merits. Frankly, anything else is BS Alanx, so let's argue like adults.

This:

Your language is difficult to understand. I do the same when I try to blow smoke up my boss's arse, especially when I'm trying to defend a weak position. (Baffle him with BS.) So, are you ceding validity to our position/argument?  So, are you ceding validity to our position/argument?

Has no place in an argument among adults. Most importantly it is a fallacy of distraction (commonly called the Red Herring) and so has no place in an argument at all.

Share this post


Link to post
Share on other sites

One more quickie.

Double-clicking a Word file will incur the same penalties under the system advocated by Alanx and Will.

My "documents" folder is located on my system partition. When I save a new/used document, it will be located somewhere within the first 12Gs (not far from MS Word itself, along with the SPs UPDATES I've made to MS Word over the months)

One a disk with a single 160G partition, you have no control over where this file with end up. You can't prevent your heads from having to seek to the far end of the disk to have to read/retrieve this file, or do pick up parts of the MS Word app that wind up there as you updated Word over the months.

Share this post


Link to post
Share on other sites

The first line of my last post should read:

"Response's to Alanx's initial response are listed in reverse order."

I would also like to say that Alanx seems to be retreating to arguments based around a small case of situations of doubtful relevance, and that the majority of my arguments that declare partitioning will not improve performance go unaddressed by his last post.

Share this post


Link to post
Share on other sites
Adding extra partitions does not prevent the fragmentation of updated application files either.

I never said anything about fragmentation. I merely said that they won't end up at the far/slow end of the disk.

Share this post


Link to post
Share on other sites
I would also like to say that Alanx seems to be retreating to arguments based around a small case of situations

I contend that program updates comprise no small case or situation. Rather they are frequent and on-going. They never cease.

Share this post


Link to post
Share on other sites
My language is precise

You language is unclear. For example, I still don't know if you were ceding validity to our position/argument?

Were you? (A simple yes or no will suffice.)

Share this post


Link to post
Share on other sites

Why don't you quote the entire passage. If it still isn't clear as daylight to you, I don't think you have any business carrying on an argument with someone over the age of 12.

3.  My language is precise and conforms to the rules of English grammar.  My arguments stand on their own regardless.  Importantly, it is rude to give lessons on grammar or diction in a public board frequented by individuals from many nations.  I will not argue that style of expression is unimportant when presenting an argument, but a criticism of an argument must be based on its merits.  Frankly, anything else is BS Alanx, so let's argue like adults.

This:

Your language is difficult to understand. I do the same when I try to blow smoke up my boss's arse, especially when I'm trying to defend a weak position. (Baffle him with BS.) So, are you ceding validity to our position/argument?  So, are you ceding validity to our position/argument?

Has no place in an argument among adults. Most importantly it is a fallacy of distraction (commonly called the Red Herring) and so has no place in an argument at all.

Share this post


Link to post
Share on other sites

I have come to the conclusion that you have been trolling me this entire thread. Are you related to Buddylite by any chance? Your posting style is virtually indistinguishable from his.

Enjoy the "noticeable" performance improvements of your partitioned hard drive.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!


Register a new account

Sign in

Already have an account? Sign in here.


Sign In Now