block device driver blk_alloc_queue settings
 

I wrote a block device driver which concatenates multiple /dev/sd* disks and create a single disk.

I used blk_alloc_queue() to create a queue for my concatenated disk and registered a dispatch function using blk_queue_make_request(). In the dispatch function i forward the io to the sd disk based on the bio->bi_sector.

But, the performance(i/o per second) of my concatenated disk is just above 50% compared to sd disk.

I think i am not setting some parameters for the queue and that might be bringing down the performance of my concatenated disk. Here are the blk_queue_* functions which i have used.

blk_queue_make_request
blk_queue_bounce_limit
blk_queue_stack_limits
blk_queue_max_sectors

Can someone help what am i missing?

Do you set your logical block size using blk_queue_logical_block_size?

nini09,

I am using 2.6.18-128.el5 kernel. There is no blk_queue_logical_block_size function in this version. So, i tried using blk_queue_hardsect_size() which i believe is the equivalent of blk_queue_logical_block_size(). But, there is no change in the performance.

How do you test the performance of Linux sd disk and your new concatenates disk?

I am using IoMeter to test the Performance.

It look like your test target is disk device. Both disk device type is different, sd disk is SCSI disk and your disk is raw block disk.

nini09,
sorry for the late response. I figured out the problem, now i am getting good performance. The problem was that i did not provide a handler for flush function in the queue.