Interesting! Benchmark Results for Game Optimization scripts(sort by name ascen)

[theantagonizer]

I did not see this mentioned as a possible issue so I thought I would throw it in. I read an interesting article on defragging which mentions the platters issue and provides a good graphical explanation of it. When MyD defrags and sorts by name it doesnt take into consideration the end of the first platter and start of the second and so on. Half of your game files could be at the end of the first platter and the rest could be on the start of the second. This would still be defraged however the needle still has to move from the end of platter one to the beginning of platter two in order to access what appears to be a contiguous folder structure.

I do not want to irk anyone by posting links to other sites so if you are interested in reading the article simply search google for "why doesnt linux need defragmenting"

The article is the first result with that exact title although it is mirrored at several sites.

[poutnik]

AFAIK logical sector numbering is not done plate by plate, but track by track across the plates.
The numbering by the former way would not make sense.

[jeroen]

I read an interesting article on defragging which mentions the platters issue

The article that you read is stupid and bogus. There is no platters issue, as Poutnik says "it does not make sense". Harddisks with multiple platters behave exactly the same as harddisks with a single platter, fast at the beginning and gradually getting slower towards the end. You can easily test this yourself with a speed measurement utility such as HD Tune. The harddisk heads never move from the end of the first platter to the beginning of the second platter. The harddisk controller automatically stripes the data onto all platters, it uses all the platters simultaneously, not one by one.

"why doesnt linux need defragmenting"

This is a myth. Linux disks get fragmented and need defragmentation just like Windows. There are lot's of articles floating on the internet with nice fancy theoretical explanations, and all are totally wrong. And by the way, disk optimization is far more important than defragmentation, a little fact that all the wise-ass authors of those articles fail to mention.

[theantagonizer]

The article that you read is stupid and bogus. There is no platters issue, as Poutnik says "it does not make sense". Harddisks with multiple platters behave exactly the same as harddisks with a single platter, fast at the beginning and gradually getting slower towards the end. You can easily test this yourself with a speed measurement utility such as HD Tune. The harddisk heads never move from the end of the first platter to the beginning of the second platter. The harddisk controller automatically stripes the data onto all platters, it uses all the platters simultaneously, not one by one.

Thank you I was not aware of this. For my own clarification what you are saying is that the logical system writes a block to the start of the first platter and then the next block (Of the same file) is placed at the start of the second platter and the next block is then placed at the start of the third platter so that is how it is using all platters simultaneously?

"why doesnt linux need defragmenting"
This is a myth. Linux disks get fragmented and need defragmentation just like Windows. There are lot's of articles floating on the internet with nice fancy theoretical explanations, and all are totally wrong. And by the way, disk optimization is far more important than defragmentation, a little fact that all the wise-ass authors of those articles fail to mention.

I thought about adding a note about this in my post but I decided (poorly) against it. I do not agree with the theory even a little. Attached is a screen shot of my hard drive after a copy of a few hundred pictures to it using the linux system. Those "speckles" are the pictures copied in the fashion that this article is advertising as better. The red files are very large archives I had to add to exclusions temporarily to save defragging time while i cleaned up the mess (with MyD) linux made. A more accurate article would have explained that the linux file system is only better when you do not have any files larger than a few dozen megs and most of the files you do have are changing size regularly (dynamic sized, IE databases). I posted that article to add to the discussion about what can affect performance not to advertise for anything being any better than anything else. If anything my screenshot just goes to show that we need a MyD for linux!!! 

[jeroen]

For my own clarification what you are saying is that the logical system writes a block to the start of the first platter and then the next block (Of the same file) is placed at the start of the second platter and the next block is then placed at the start of the third platter so that is how it is using all platters simultaneously?

Yes, like that. I'm not sure how big the blocks are, perhaps the striping is per sector, or perhaps even per bit.

we need a MyD for linux!!! 

Yes. It's a pity that the Microsoft defragmentation API is not available for Linux, on which MyDefrag is based, so MyDefrag cannot be ported.

[quanthero]

Linux filesystems do indeed fragment albeit at a slower rate than NTFS. However, this does not automatically imply that defragmentation of Linux FS is beneficial. There are reliable studies/experiments showing performance degradation from fragmentation on Windows, but there are no such studies for Linux (AFAIK). Also, the structure of Linux filesystems differs from that of NTFS, so defragmentation of Linux filesystems may give unexpected results such as a reduction in performance (due to e.g. the fact that related files are moved farther apart from each other during defrag).

However, I do agree that optimization is absolutely necessary on Linux filesystems. IMHO, something like 'Hot File Adaptive Clustering' (automatically moves most frequently used files to the beginning of a volume close to metadata) used in OS X will be very useful for Linux. MyDefrag's approach of having different 'zones' will be useful as well.

[jeroen]

Linux filesystems do indeed fragment albeit at a slower rate than NTFS.

No. I do not have hard statistics at hand but the Linux filesystems will fragment at a very similar rate as NTFS.

There are reliable studies/experiments showing performance degradation from fragmentation on Windows, but there are no such studies for Linux (AFAIK).

There are no studies because Linux does not have tools to measure and correct the fragmentation. With all respect, but the absence of studies does not prove that there is no problem.

Also, the structure of Linux filesystems differs from that of NTFS, so defragmentation of Linux filesystems may give unexpected results such as a reduction in performance (due to e.g. the fact that related files are moved farther apart from each other during defrag).

The theoretical degradation that you describe also applies to NTFS, there is no magical difference in this respect. Also, the effect depends on the defragmentation program. Not on the filesystem.

[quanthero]

Linux filesystems do indeed fragment albeit at a slower rate than NTFS.

No. I do not have hard statistics at hand but the Linux filesystems will fragment at a very similar rate as NTFS.

Depends on filesystem in question. ReiserFS fragments at crazy rate. Ext2, ext3 and ext4, on the other hand, fragment at somewhat slower rate than NTFS. How do I know? I did some tests (please see below). For example, my old laptop running Linux on ext3 filesystem has less than 1.05 average fragments per file.

There are reliable studies/experiments showing performance degradation from fragmentation on Windows, but there are no such studies for Linux (AFAIK).

There are no studies because Linux does not have tools to measure and correct the fragmentation. With all respect, but the absence of studies does not prove that there is no problem.[/quote]

There exists a tool that measures fragmentation on Linux. Please see the following thread in Gentoo forums for more detail: http://forums.gentoo.org/viewtopic-t-429915.html. But I do agree with you that the absence of studies does not prove the absence of problems. The truth is that one cannot say anything definite about performance degradation due to fragmentation in Linux, unless someone did some thorough tests on this topic.

Also, the structure of Linux filesystems differs from that of NTFS, so defragmentation of Linux filesystems may give unexpected results such as a reduction in performance (due to e.g. the fact that related files are moved farther apart from each other during defrag).

The theoretical degradation that you describe also applies to NTFS, there is no magical difference in this respect. Also, the effect depends on the defragmentation program. Not on the filesystem.
[/quote]

I agree that many things depend the defrag program (are you referring to various file placement optimizations built into defrag programs?). But, IMHO, filesystem plays some role, too. E.g. xfs deliberately 'fragments' data into several allocation groups, which may be very beneficial performance-wise when volume is filled by more than half.

[theantagonizer]

LOL I didnt mean to hijack this thread into a debate on NTFS or Ext. I think we can all agree that both have some drawbacks that should be fixed in the near future. Both are not really perfect when dealing with the difference of small files to large files and files that are used often being "intelligently" sorted. This is why we use MyD  Ext4 is offering a defrag tool so the developers agree sometimes it is needed. When the day comes that NTFS or Ext version X incorporates some of the logic behind MyD while operating the whole world will benefit! 

[Sauliavas]

well i found this topic interesting, and tehn read some readme info about sorting and found some trivial thigs about sorting by last acces time or name, so  i think right idea to sort a game dir is exclude tmp files, and sort by creation date ascending - meaning it will placed just like game creators wanted at instalation

[Darlis]

sort by creation date ascending - meaning it will placed just like game creators wanted at instalation

So the programmers know which file will be accessed in which order and implemented an installer that will copy the files in that order? But why is never mentioned when searching for how to optimize game performance?