Storagereview/7200rpm hard disks: why are NTFS scores higher than FAT32?

[Leo V]

I looked at www.storagereview.com's hard drive benchmarks, and noticed that all Windows NT (NTFS) scores were a notch above the equivalent Win9X (FAT32) scores. Does this really mean that using NTFS is faster than FAT32? I'm going to install Windows2000 Release Candidate 1, should I go with NTFS 5.0 instead of FAT32 for speed purposes? (This could also influence my choice of hard drive http://www.sysopt.com/forum/smile.gif

Thanks a lot,
--Leo V.

[Perce]

It is because of NT mainly, with NTFS, it has slightly faster I/O operations. But unless you using NT5/Win2000, NT4 does not have disk defrag. So after a while it runs like a fatman going uphill with plastic bag overhead!!! Win2000 has disk defrag, so no problem there. http://www.sysopt.com/forum/smile.gif

[Leo V]

Thanks Perce. In this case I will get the Quantum Katana ATA66 disk (it's faster than Western Digital in NTFS), Win2000 RC1 and NTFS5.

[Zonker]

NT is way faster in cached disk performance than 9x... much much faster http://www.sysopt.com/forum/smile.gif

You can Defrag hard drives in NT4 with Executive Software's Diskeeper, a wonderful program

http://www.executive.com

[Leo V]

I can easily grasp that Windows NT is faster than 9X at disk caching http://www.sysopt.com/forum/smile.gif What I was trying to establish is this: would NTFS be faster than FAT32, if both were used only under Windows NT.

BTW thanks for the link.

[This message has been edited by Leo V (edited 07-24-99).]

[Stan]

Well NT can NOT read FAT32. It can read NTFS and FAT16 only.
Maybe you meant FAT16 ???

Stan

[Zonker]

Here's a coverage of MS file systems by windows magazine

http://www.winmag.com/library/1999/0201/fea0056e.htm

In addition to better security and file recovery, NTFS provides significant performance gains. Because it uses a more sophisticated search algorithm than FAT, NTFS can access files on multigigabyte partitions and drives much faster.

[Leo V]

Zonker, that's perfect. Exactly what I wanted to know. Question closed. http://www.sysopt.com/forum/smile.gif

Stan, Windows NT4.0 on my machine reads FAT32 just fine after I installed FAT32 drivers from Winternals. The exception is that the boot drive still has to be FAT16 (I currently have WinNT4/Win98 in dual boot). However, this shouldn't matter anymore soon, since Win2000 has native FAT32 as well as NTFS support.
--Leo V.

[BBA]

Zonker, NT5 is faster with FAT32 than with NTFS5!

So, I guess W2K does pretty much work on FAT32 huh?


NTFS does use much more restrictive access rules than FAT, and as such, it does so much auditing of rights when every file is even touched that it is slower than fat, unless you use CVF's and raid. It is mostly NTFS that causes NT to shutdown so slowly as every service has to be stopped and secured including file rights for NTFS volumes.


Also, the cluster arrangement is totally different, NTFS is closer to drive space than to FAT clusters and therefore also has slower access times.

I think NTFS is great for 2 things only, security and large disk support.

BBA

[Leo V]

Now wait a second here!

Now FAT32 *is* faster than NTFS5 given you use Windows 2000? This is getting really confusing. BBA, I can see your reasoning (it makes sense), but what about Zonker's link to the article that appears to claim the contrary?

I *must* get the facts straight before I receive and format my new hard disk :~o

[This message has been edited by Leo V (edited 07-24-99).]

[BBA]

OK, I ran no performance benchmarks except for timing load times for various apps, including shutdown/restart for W2K itself. With NTFS, startup/shutdown/reset took around 3 minutes. With FAT32 it only took 1& 3/4 minutes. Apps load quicker with FAT32 and Explorer opened quicker. The only reason I timed it is I crashed a NTFS build, fdisked and formatted FAT32 and left it FAT32 and noticed a huge difference in app times.

I really can't tell you if Q2 is faster with NTFS or FAT32 because I haven't gotten that far yet. I can imagine I have enough RAM that disk access time will not affect these benchmarks as all program material will be in the memory by the time the bench runs.

Ok, so it might only get 8M/sec sustained data rate instead of 8.5M/sec....

All I can say is try it both ways and make your own mind up about it.

BBA

[Zonker]

I just timed my Win NT4 shutdown and startup times on an NTFS volume, here are the results...

shutdown = 34 seconds
startup = 66 seconds

NTFS gets crammed loading and shutting down due to the fact that not unlike FAT16/FAT32, the volume is really badly organized and fragmented... from the start.

If you run Executive Softwares Diskkeeper on a routine basis with NTFS, it is noticably faster, in my experience.

[BBA]

NT4 isn't nearly as slow as NT5 for shutdown/startup. It can be with lots of time/installs/abuse/fragmented files, but NT5 still takes the cake!

BBA

[ANTONIO E GUERRA]

The fact that Nt 4 or Nt5 are very slowly does not have anything to do with the file structure but with the network structure. During the start up process, the network operating has to check all the components of the network (especially the network card) and establish a hardware log; in the authentication process, the network operating system has to establish the user's password and id and verify with the token right aasigned to your computer. These operatings don't have anything to do with the file structure of your system.

[Leo V]

There seems to be no complete concensus regarding which file system is faster under Windows 2000 (both sides have convincing arguments AND results).

This means I'll have to conduct my own testing. I'll bet that file system "A" is faster; then I'll format with system "B" and run benchmarks. Afterwards, clean reformat/reinstall to file system "A", and rerun the benchmarks. If my bet was right, I won't have to reformat again. If wrong, that'll cost me yet another format and reinstall.

For benchmarks, I will take the clean install with all hardware drivers configured, and measure the following separately.
(I will get a Quantum KA ATA66 13.6GB 7200rpm 8.5ms hard drive, and an Abit BE6 ATA66 motherboard; PIII/450 shouldn't be a bottleneck).
1) Startup time
2) Shutdown time
3) Photoshop 5.0 load time
4) Time spent making a copy of a 320MB swapfile
5) Time spent making a copy of a folder with 2,048 files (32KB each, all copy of one original text file; this is easy to set up http://www.sysopt.com/forum/smile.gif)
I will post my results here once I run the benchmarks...however that won't be for at least 2-3 weeks. (Unless the real answer emerges before then). If you have other good/easy benchmarking ideas, please tell me.
--Leo V

[Zonker]

FYI.... WIN2K includes Executive Softwares Diskkeeper Lite... for just the reason I mnetioned

http://www.execsoft.com/press/19990412.htm

The Microsoft Press book "Running Microsoft Windows NT Server 4.0" describes Diskeeper as ‘The most commonly used disk defragmentation program for Windows NT.’ Diskeeper is certified as Microsoft BackOffice compatible and it speeds performance for the entire family of BackOffice server products.

[Leo V]

BTW I've just found that the official word from Microsoft is that NTFS 5.0 (in Windows 2000) should be both faster and more advanced than FAT/FAT32. But I'll run the tests anyway (I'll just do FAT32 first then).

[Mike H.]

Leo,

Check out IBM's new 7200 RPM deskstar HD's. I hear there the fastest IDE's, creaping ever closer to scsi.

[Leo V]

Not really. Check www.storagereview.com's comparison benchmarks at:
http://www.storagereview.com/cgi-bin/bench_compare.pl?button=Lets+Compare!&Drive_1=IBM+Deskstar+22GXP+(22.0+GB+ATA-66)&Drive_2=Quantum+Fireb all+Plus+KA+(18.2+GB+ATA-66)&Drive_3=Western+Digital+Expert+(18.0+GB+ATA-66)

(IMPORTANT: delete the space in "Fireb all" to make the link work; it somehow reappears in the message)

The Quantum is ahead in 1999 benchmarks, as well as in NTFS benchmarks (which in general produce faster scores, supposedly). Besides, its seek time slaughters everyone else (except serious SCSI drives). Quantum's seek time is advertised as 8.5ms but appears more like 7ms.

[This message has been edited by Leo V (edited 07-26-99).]

[BBA]

I still beg to differ on start up and shut down times of NTFS vs FAT. The whole reason being that NTFS is a secure file system, and individual file rights are checked each time the PC starts up or shuts down. FAT deos not have this occurring.

The test can only be accurate if you compare a NT5 install on a pure NTFS5 drive against a NT5 install on a pure FAT32 drive.

In the next week or so I will do some official type benchmarking of NTFS vs FAT32 operations and data rates.

I will post them.

BBA

[Leo V]

Thanks BBA. Like you said, the benchmarks I proposed would involve a fresh install of W2K/RC1 on a freshly formatted hard disk, for either file system (using a computer I will build in a couple weeks, unfortunately delayed for budget reasons). We should receive Release Candidate 1 any day now (ordered last Tuesday). BTW are you using W2K Beta 3? I wonder if the W2K version might have any effect on relative file system performance (considering MS's stand).
--Leo V.