Older blog entries for davem (starting at number 3)

Jens is interested in improving elevator now too :-) Really, the problem is the missed coalescing, the latency concept is not fundamentally broken. The algorithm can get a lot of both worlds by coalescing whenever possible. In this way it performs latency reduction on groups of requests instead of on a per-request basis.

Got distracted for two days working on a prototype fibre channel card, can't talk much about it, but I can say there'll be a Linux driver for this beast :-)

Sun released their IFB (aka. "Expert 3D") graphics card yesterday. It's a bit pricey, but I'm so itching to do a Linux driver for it I may just end up buying one.

The cscope release made Larry McVoy nearly wet his pants.

I think I know why Andrea's new elevator code is so much worse than 2.2.x's disk sorting. The general scheme of the elevator is to keep low the latency introduced to old requests by new requests being added. In doing this, it prevents merging with other requests to the same area of the disk. In actuality this "latency reduction" is really increasing latency, because a missed merge means it will now take two I/O operations to move data which could have been done by a single request. Also, since the elevator will move the I/O request towards the end of the I/O queue, it is almost certain that requests to other sectors will happen first, thus an unnecessary disk seek. I have no idea why the elevator make LVM/IDE dbench runs faster for him, as it seems quite wrong fundamentally.
Will study this more soon.

Just ignore rth, he's jelous or something I suppose. :-)

Cobbled together 2.2.15pre fixes for Alan, sent them off. The Netra-T1 irq probing bug and the sym896 issues on sparc64 should now be cured, as is the double poll_wait problem in sbus audio.

Watched the demo of the RazerCPL quake3 final match between nb.fatal1ty and D16-Makaveli. Fate has some insane new moves in q3dm13 I've never seen before, the best was the double RL jump out of the basement lava back safely to the second level. He tried it again later in the match but didn't have enough health to pull it off :-)

Head of TODO List: SKFP driver portability work.