Romex Software Forum

Hi

It's seems that 4KB random read (QD 32) is not good enough compared to (Windows without F.C.)..

FancyCache enabled :-
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]
Random Read 4KB (QD=32) : 0.866 MB/s [ 211.4 IOPS]
Random Write 4KB (QD=32) : 2.869 MB/s [ 700.4 IOPS]

Test : 500 MB [C: 71.2% (10.4/14.6 GB)] (x2)

FancyCache disabled :-
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]
Random Read 4KB (QD=32) : 1.214 MB/s [ 296.4 IOPS]
Random Write 4KB (QD=32) : 1.286 MB/s [ 314.1 IOPS]

Test : 500 MB [C: 71.2% (10.4/14.6 GB)] (x2)

Tested on WD HD SATA-2 8MB Buffer. Windows XP SP3. FancyCache size = 244 MB (F.C.Disk v0.6.1 - Block Size = 32KB), http://crystalmark.info/

Because the cache size was 244MB, which is less than half of the Test Size (500MB). The read cache hit rate shall be almost 0% while doing a testing with some bench tools. So in theory the read performance shall be almost same as that of the original disk. However because of the implementation, the best peformance will be achieved when the block size is same as the cluster size of the file system.
please refer to viewtopic.php?f=25&t=706

Thanks for your reply!

Wow! 4KB (Block Size) is 3X faster than 32KB here..

----------------------------------------------------------------------
32KB Block size :-

Random Read 4KB (QD=32) : 0.866 MB/s [ 211.4 IOPS]
Random Write 4KB (QD=32) : 2.869 MB/s [ 700.4 IOPS]

4KB Block size :-

Random Read 4KB (QD=32) : 2.536 MB/s [ 619.2 IOPS]
Random Write 4KB (QD=32) : 2.553 MB/s [ 623.2 IOPS]

Test : 500 MB [C: 71.2% (10.4/14.6 GB)] (x2)
Date : 2011/04/01 8:22:44

FancyCache (disabled) :-

Random Read 4KB (QD=32) : 1.214 MB/s [ 296.4 IOPS]
Random Write 4KB (QD=32) : 1.286 MB/s [ 314.1 IOPS]
-----------------------------------------------------------------------

Thanks again! : )

P.S.
I have deliberately used (244 MB cache size) with (500 MB test file), because It's highly recommended to use a test file that is larger than the cache size of the disk being benchmarked, especially when using a block-level disk cache (2:1/3:1).