Poor performance NTFS large cluster sizes

[davetest]

Hello, I added a second harddrive, 30GB, NTFS, dynamic
and 32KB cluster sizes.

I created it this way because I primarily use it for large
media files only.

There are certain operations on this drive that are incredibly slow.
I took a 200mb file and *copied* it from one folder to another
and it took 50 seconds. Drive is an IBM gxp75 udma 100.
I can hear the drive thrashing.

If I try this operation on the C: drive, fat32 and smaller clusters
it takes 15 seconds. Drive is a maxtor DiamondMax Plus 45 Ultra ATA
100.

Any thoughts why this should be? Using HDtach analyzer,
the "slower* drive actually receives a higher score than the
"faster " one !

Dave

 

[R. McCarty]

From your problem description the only thing that stands out is
that the new drive is partitioned as a "Dynamic" drive.
How is the hard drives configured on your IDE channels ?

Also, using large cluster size it is important that the drive is
defragmented.
Are both of these drives defragmented ?

There is a tool on the internet called Disk Speed 32 that can test
drives for Access times, Burst and min/max transfer rates. You
should test each drive to see its baseline performance.
There shouldn't be a significant performance difference between
"Basic" vs. "Dynamic" and the native format of the drive.

I use a dedicated SCSI drive (64K Clusters) for all my music files.
However, it is a "Basic" disk, and handles the ~4-6 Meg files just
fine.

 

[davetest]

From your problem description the only thing that stands out is
that the new drive is partitioned as a "Dynamic" drive.
How is the hard drives configured on your IDE channels ?

Also, using large cluster size it is important that the drive is
defragmented.
Are both of these drives defragmented ?

There is a tool on the internet called Disk Speed 32 that can test
drives for Access times, Burst and min/max transfer rates. You
should test each drive to see its baseline performance.
There shouldn't be a significant performance difference between
"Basic" vs. "Dynamic" and the native format of the drive.

I use a dedicated SCSI drive (64K Clusters) for all my music files.
However, it is a "Basic" disk, and handles the ~4-6 Meg files just
fine.

I read up on some the differences between dynamic and basic and
nowhere have I seen mentioned any performance implications.
It's an interesting thought and I could certainly change it back
to basic to see if it changes anything. (I'll have to empty it first)
Regarding defragmentation, well there is a little but because many of
the files are in the 500-700mb range, it's hard to defgragment.
22GB in use, but only about 300 files in total.

The fat32 drive is basic and primary master, while the dynamic drive
is primary slave. Performance hit because of slave?

Copying files back and forth between the two is really fast -
30 seconds for a 500MB avi file.


Dave

 

[Alex Nichol]

Hello, I added a second harddrive, 30GB, NTFS, dynamic
and 32KB cluster sizes.

I created it this way because I primarily use it for large
media files only.

There are certain operations on this drive that are incredibly slow.
I took a 200mb file and *copied* it from one folder to another
and it took 50 seconds. Drive is an IBM gxp75 udma 100.
I can hear the drive thrashing.

Cluster sizes other than 4 K are less than optimal, because they do not
match the internal CPU page size of an Intel architecture processor, for
mapping virtual to physical memory. So any operation going through
memory (and almost anything does - certainly copying) has to be buffered
instead of being handled by DMA direct between device and RAM

 

[davetest]

Cluster sizes other than 4 K are less than optimal, because they do not
match the internal CPU page size of an Intel architecture processor, for
mapping virtual to physical memory. So any operation going through
memory (and almost anything does - certainly copying) has to be buffered
instead of being handled by DMA direct between device and RAM

Hi Alex,
I gave erroneous info in my original post.
I checked, and instead of 32KB clusters they are actually 8KB.
Hdtach shows this drive to be faster than the C: drive, while
the results of diskbench32 are inconclusive.
I ran the file tests on Sisoft Sandra and the results from that
are closer to matching my real-world feeling that the I: drive
is slower when copy/moving file around. Here's the result:
Buffered read c: 51mbps i: 68
seq read c: 30 mbps i: 23
Random read c: 7 mbps i: 5
Buffered write c: 40 mbps i: 41
Seq write c: 30 mbps i: *15*
Random write c: 9 mbps i: *5*

When I get a chance I'll set it to 4kb cluster and see what happens.
Dave