Messages in this thread | | | From | (Alan Cox) | Subject | Re: Porting Greg Lehey's rawio.c from FreeBSD to Linux... | Date | Tue, 4 May 1999 18:59:38 +0100 (BST) |
| |
> 1) You don't trash your buffer cache > > 2) You don't waste your time and that of the database when it > could avoid the filesystem redirection entirely? > > Or am I mistaken in believing that the block device interface is also > pumped through the vm/buffer cache system?
You are correct. It is also not co-incidental that one of the main things being tested with the O_DIRECT patches and doing DMA to/from user space is Oracle. The O_DIRECT flag as with SGI machines does (or can do depending on the card) direct DMA to and from user pages. This is precisely what things like a fault tolerant oracle setup on shared scsi or fc busses happens to need.
> system interfere with that testing? It would seem to me that you're > assuming that you folks know more about all possible uses of Linux and > how to best tune the system for all those uses, than anyone who could > possibly come along and want to use a Linux system for a particular purpose.
I don't think that is the case - all the VM is tuneable and you can already accurately measure through the page cache throughput (man hdparm for a simple tool that includes this). If you want to measure raw disk I/O rates without OS overhead then you can't do it without the patches for raw I/O
Alan
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |