search

Aborres / compcache

Automatically exported from code.google.com/p/compcache
1 stars 0 forks source link

Incorrect queuing and scheduling on zram #88

Open GoogleCodeExporter opened 8 years ago

GoogleCodeExporter commented 8 years ago 
I think this two should have better default values.

# cat /sys/block/zram0/queue/read_ahead_kb
128
# cat /sys/block/zram0/queue/rotational 
1
#

rotationl should be 0, obviously.

As of read_ahead_kb, linux kernel already is pretty good at detecing sequential 
and random IO, so it should be be issue, but smaller read-ahead will make 
memory usage lower, as well isn't nacassary, as IO for zram have very low 
latency compared to rotational devices. Maybe there can be some effect with 
code and data cache if read_ahead is enabled, but i really dubt it. Having it 
smaller could improve slightly read performance, but i do not benchmarked it 
yet.

Original issue reported on code.google.com by witold.b...@gmail.com on 18 May 2011 at 11:40

GoogleCodeExporter commented 8 years ago 
I tried dd_rescue /dev/zram0 /dev/null and get slightly better values with 
8-16KB rather than 128KB.. Not sure that it would reflect any real-world 
performance benefits.

Using an Athlon XP 2600+ (~1900MHz), Gentoo system and the staging version from 
kernel 3.1.5.

Original comment by ande...@gmail.com on 28 Dec 2011 at 4:10