LWN: Comments on "VM changes in 2.6.6"

order of I/O

corbet — 2004-04-17T00:17:17+00:00

Actually, the order is determined by the I/O scheduler. But the scheduler can only work with requests once they are handed to it. Changing that order can make a big difference in what the driver sees. And, apparently, in some situations, the performance difference can be significant - either better or worse.

order of I/O

giraffedata — 2004-04-16T22:53:52+00:00

As a result of the changes, the order in which dirty pages are written to disk has changed; writing always happens in file-offset order now.

I find that hard to believe. The order in which pages are written to disk is controlled by the block layer/device driver, and tends to be disk address order. I presume this means to say the order in which the I/Os to write dirty pages to disk are requested of the block layer always happens in file-offset order (as opposed to order in which they became dirty) now. It's hard to see how that makes a big difference in performance, considering I/Os to clean all the dirty pages are requested at about the same time.

Aphorisms in the Linux community

Duncan — 2004-04-16T10:14:52+00:00

> It's like juggling four bars of wet soap
> with your eyes shut while someone is
> whacking you with a baseball bat.

LOL! I predict I'll see that in a sig line sometime in the future!

One of the interesting things about open source is insight into the simple
humanity aspect of the kernel hackers on occasion. Linus's always
real-world applicable humor is another example, and one couldn't help but
draw their own conclusions at the contrast between the pix of Bill Gates
with cream-pie in his face, and Linus, sitting on that dunk tank board. I
know which guy's kernel *I'm* more likely to be comfortable with running
MY "mission critical applications" on, even if those "mission critical
applications" are no more than XMMS, my newsreader (PAN), and web browser
(Konqueror), all running at the same time.

Duncan