We have tested the rawio and o_direct patches from Andrea Arcangeli
with great success, and wondered if they will be part of the next
kernel release, 2.4.4?
So far we have tested these patches on four different systems.
With the latest versions of these patches (rawio-6, o_direct-3)
all systems works perfectly ok. Both with and without O_DIRECT
enabled in big file operations.
The O_DIRECT support really helps performance a lot
for some types of applications. We happen to have such an
application, and we would like to see O_DIRECT
support in the kernel source in the near future.
This would lift Linux to a new level in terms of disk performance
for "self-caching" applications.
A performance test we did on our application shows a huge
improvement, upto 8 times better data read speed with direct-IO.
With buffered IO, the CPU is doing caching most of the time.
I attached a graph showing this performance win using O_DIRECT
versus buffered IO. Each thread reads at a random offset a random
amount of data from a big data set. The dataset is placed on
a software RAID 0 with many disks. With that configuration
we are able to fully utilize the 160 MB/s SCSI controllers.
-- Michael Susæg, M.Eng. Mail: Michael.Susag@fast.no Software Engineer Web: http://www.fast.no/ Fast Search & Transfer ASA Phone: +47 21 60 12 27 P.O. Box 1677 Vika Fax: +47 21 60 12 01 NO-0120 Oslo, NORWAY
Try FAST Search: http://www.alltheweb.com/
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to firstname.lastname@example.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Apr 30 2001 - 21:00:18 EST