ntfs performance compare 32k or 4k

W

WiReDPmP

Weaksauce
Joined
Nov 10, 2001
Messages
108
I've installed windows 3 times in the past three days due to conflicts, new drivers, etc.
At first I couldn't change my cluster size, then did *something* after I got a 1618 error relating to compression and not enough files? and boom, it changed. Now I can't get it to go back (xp format) to 32k. So I was wondering if there is a write-up out there that I cannot find that would help explain all this. btw, sandra scores 59 on 32k and 62 on 4k, marginal, maybe not even credible.
 
the default cluster size is best for everyday use.
If you have lots of small files then a small cluster size is best.
If you have lots of large files then a large cluster size is best.

If you use a large cluster size and have many small files it is very ineffeciant at storing that file. A 3k file (there are lots of them in a Windows install for example) would take up 32k of room if you have 32k clusters.
 
:D I crashed my dad's computer changing to 64k. So then it was ghosted and changed to 2k. SLOW WWW as a slug. It's a seagate dma w/great read speeds, but wtf is wrong w/ it? 2k clusters! yeah. 4k makes a huge difference. And gaming wise, I think that there are a lot of small files when it's loading between levels, so 4k seems to be working out fine.
But I haven't tested it.
 
WiReDPmP said:
And gaming wise, I think that there are a lot of small files when it's loading between levels, so 4k seems to be working out fine.
But I haven't tested it.
Click to expand...

Some games keep their files in huge files like quake pak files others keep thier files serpate (I just happen to know that GTA does) so it really depends on the game too.
Basically, if you don't know any better then don't screw with it. :p
 
gigabyte1024 said:
Some games keep their files in huge files like quake pak files others keep thier files serpate (I just happen to know that GTA does) so it really depends on the game too.
Basically, if you don't know any better then don't screw with it. :p
Click to expand...
Quake .pak files are zip files; they're not read sequentially. Just because they're one big file doesn't mean t
 
Zip files don't give you any guaranteed performance benefit because the game doesn't necessarily access the files inside that zip in a sequential manner. Though I would imagine some games would be optimized to do that... Even then with all the prefetch algorithms and other fancy optimizations in modern hard drives I don't think there will be any tangible difference.

Not that many of the people here would care but Windows memory pages are 4k too so if you use virtual memory, a 4k cluster size is optimal for the disk that has the page file.
 
gigabyte1024 said:
Are they not stored on the disk as one large file, therefore more efficient (theoritically) on larger cluster sizes?
Click to expand...

They'd benefit from being stored more compactly, and taking up less space overall... but in terms of reading them, no. The compression will likely have the impact of reducing the number of files which are contained entirely within a single block, which means more often you have to wait for two blocks to be read. Since fragmentation can happen, that can lead to quite a bit of waiting.

I wonder what I was doing that made me stop typing in the middle of my last reply and hit submit...:confused:
 
perfectdisk is a good def.
And no, HL2 files are extracted.
Also, to gain some performance, make sure your controller has NCQ on. Your hard-drive has cache enabled in the driver, and turn off drive indexing.
 
WiReDPmP said:
perfectdisk is a good def.
And no, HL2 files are extracted.
Also, to gain some performance, make sure your controller has NCQ on. Your hard-drive has cache enabled in the driver, and turn off drive indexing.
Click to expand...

I have heard that NCQ is often detrimental to single-user (i.e. desktop) performance.
 
what about ahci? is that native to ncq? +I got better results on the jmicron on ahci than sata on the intel8
 
WiReDPmP said:
what about ahci? is that native to ncq?
Click to expand...
AHCI is a standard that Intel is pushing (has developed?) for SATA controllers. I really do not exactly understand what the ``Advanced Host Controller Interface'' does, but I could imagine that it is a prerequisite for a single driver that will work across all SATA controllers.
WiReDPmP said:
+I got better results on the jmicron on ahci than sata on the intel8
Click to expand...

Better results in what?
 
Xeth said:
Not that many of the people here would care but Windows memory pages are 4k too so if you use virtual memory, a 4k cluster size is optimal for the disk that has the page file.
Click to expand...

Cluster size doesn't affect paging efficiency. The physical read happens at the sector level.
 
WiReDPmP said:
hell, all loads, including start-up
Click to expand...

Measured with a stopwatch? Can you post your before and after numbers?

mikeblas said:
Cluster size doesn't affect paging efficiency. The physical read happens at the sector level.
Click to expand...

Stupid question that I could probably google: What is influenced by cluster size? I assume that the amount of ``wasted space'' increases with large cluster sizes, given that the file size distribution stays constant.
 
Allocation granularity. If you want to store something, you're allocating it (or freeing it) in chunks that are the cluster size. You'd like your cluster size to be the median of all your file sizes, for this reason.

Allocation management. If you allocate a gig, you need to find n clusters which satisfy that gig. If the cluster size is lower, n is higher; if the cluster size is higher, n is lower. As such, some operations involving finding free space, marking free space, and so on, will be longer or shorter depending on the cluster size. This can be an issue for drives with lots (and lots! and lots!) of files, very very frequently being read to and written from. You know: like a big-time, heavily-used, shared file server. Even then, I'm not sure if file size is ever predictable enough to change or tune this setting.
 
I'd love to do benchies, but I never benched it. Board gave out beforehand. The controller I do think from experience had ncq on, and it took awhile to boot. ACHI on 9 passes, off 11 1/2 passes in start-up
edits:: I also wanted to point out that I had an old program called Wintune, and a seagate 200gb w/some of the best read performance in the ata-133 industry was getting around 365mb in cached read vs. my ibm deskstar that I got from a friend @ 625mb/s. Now this is very interesting! b/c that's exactly how the computer feels, like thick molasis! Maybe JUST MAYBE ntfs would show a lil different (slower) resluts because wintune doesn't know what it is, but what's interesting is uncached disk read about 7mb/s on the ibm, and the seagate was at an astounding 250 or so. To my knowledge this whole edit I jsut posted is of utter unimportance and should be thrown out :( 200ntfs/ibm60fat32
 
wait wait, everyboday just hold on. I was just asking this b/c I didn't wanna wait so long on my boot, my old 2100+ xp would boot in less than two passes. So now I think I got the just "jist" of it all, don't screw w/it, modify it instead to suit your needs, and for petes sake, go through your entire mobo cd and use your software oc'ing instead. Lets keep it all ok k


I can't even change my clusters anyway. All this boot cd/pm8 cluster size change errors are just too much, 4k all the way ye?
 
mikeblas said:
Allocation granularity. If you want to store something, you're allocating it (or freeing it) in chunks that are the cluster size. You'd like your cluster size to be the median of all your file sizes, for this reason.

Allocation management. If you allocate a gig, you need to find n clusters which satisfy that gig. If the cluster size is lower, n is higher; if the cluster size is higher, n is lower. As such, some operations involving finding free space, marking free space, and so on, will be longer or shorter depending on the cluster size. This can be an issue for drives with lots (and lots! and lots!) of files, very very frequently being read to and written from. You know: like a big-time, heavily-used, shared file server. Even then, I'm not sure if file size is ever predictable enough to change or tune this setting.
Click to expand...

Thanks for the explanation. I did not consider that larger cluster could help reduce CPU and I/O load when space for a large file needs to be allocated. Am I correct in assuming that -provided that files are large- a larger cluster size would also reduce the time required to defragment.

Regardless of this, I interpret your answer to imply that trying to optimize cluster size is most likely wasted effort for a home system.
 
You must log in or register to reply here.
Back
Top