How read/write requests are dispatched to the ramdisk driver

Hi,
I am trying to measure the performance of ramdisk (size is 2GB) on a Nehalem based system (with CentOS5.4) with 4GRam. It looks like I get better throughput (~4GB/sec) only when I try to read the data of 512K. If I read less or more than 512K, then the throughput is less.
If ramdik driver is using the request queue then it makes sense. But ramdisk driver is not using request queue and it has its own make_request function i.e brd_make_request(...). Can anybody tell me what could be the reason here in getting better throughput for a transfer of 512K.

Thanks & Regards,
Ramu

When you read more or less than 512K, what size are you using? If you are using values that are not multiples of the system page size (probably 4k) then the slowdown is understandable, especially if they are also not a multiple of a system word size (either 4 or 8 bytes, depending upon whether the OS is 32-bit or 64-bit).

I was using the transfer sizes 64K, 256K, 512K, 1M, 2M etc.

Regards,
Ramu

Well, those should be OK, although the nominal sector size for discs (including I believe ramdiscs) is 512bytes, and file system chunk size is 4k (4096) bytes. So, why there is a serious degradation if you select something other than 512KB to read I don't know enough to answer right now.