> > After having CVSupped to the latest 4.0-CURRENT tree > > (just now), I noticed the amazing speed of the new ATA driver. > > The amazing speed of the new ATA driver? Were you using 32 bit transfers > and multi-sector IO with the older driver? > > I assumed from the benchmarks posted that I had no reason to feel > alarmed about the fact that the new driver offered no tangible > performance increase. > > My understanding was that we'd "feel it" when DMA transfers were enabled > for those drives that support them. I also tried the new ata code and wasn't impressed by the speed: With the old driver I achieved 12 MB/s with almost no CPU load on my ST34321A drive. The new driver works flawless in my configuration (I only have a single IDE drive attached, the rest is SCSI. Chipset is Ali Aladdin), but I only get about 7MB/s from the drive while having 70% irq load.
(I have done the tests with dd from the raw device and a block size of 1MB) So I didn't notice any problems with the new code, but we'll have to wait until DMA is supported to get a really decent performance. Just to be complete. Here are the relevant portions of dmesg output for just another working configuration: ata-pci0: <Acer Aladdin IV/V IDE controller> rev 0xc1 int a irq 0 on pci0.15.0 ata0 at 0x01f0 irq 14 on ata-pci0 [...] ad0: <ST34321A/3.29> ATA-4 disk at ata0 as master ad0: 4103MB (8404830 sectors), 8894 cyls, 15 heads, 63 S/T, 512 B/S ad0: 16 secs/int, 0 depth queue Daniel To Unsubscribe: send mail to majord...@freebsd.org with "unsubscribe freebsd-current" in the body of the message
