Re: Uhh...Informix on block device?!

In article <01bc4f6c$8c9ea240$b416a5ce@primenet.primenet.com>, Richard <washer@primnet.com> wrote: >> I'm just taking up a sysadm/DBA on a new job. Just found that our >Informix >> is using raw chunks but these chunks point to the block device under >> /dev. (e.g., /dev/lv00 as opposed to /dev/rlv00). >> >> This is _extremely_ weird to me, as I know we should use /dev/rlv00, but >> Informix is still happy about it. I'm concerned about the /dev/lv00 is >> buffered I/O device and something worst could happen. On the other >> hand, a simple "dd" tests shows (not surprisingly) the block device are >> faster than raw devices...so... Of course a block disk device will be faster with dd, which is doing double buffering for sequential read is a trivial example. I suspect that the dd example would converge to the same speed as you increase the buffer size. But, a RDBMS that has N buffering for random read/write won't behave this way. You definitely want the character device which does no buffering.. Informix does it's own buffering, therefore the kernel disk cache is redundant. The double copy is also wasteful of CPU time. Whoever set it up this way made a mistake. -- Rudy Chukran | IBM RSCS: CHUKRAN at AUSTIN IBM RS/6000 Technical Support | internet: chukran@austin.ibm.com Mail stop 4103,11400 Burnet Rd. | Austin, Texas 78758 |