Please enable JavaScript.
Coggle requires JavaScript to display documents.
bad writeback performance - Coggle Diagram
bad writeback performance
reports
kernel BZ 204253
prabably
the 16 sandisk udisk
write file is quite slow
investigate
high level idea
understand related traceevents
writeback theory
basic model
IO task
CPU
memory
page cache
writeback cache mode
data writtent to file is written to PG
eviction cache
when
how
is it really good to cache data?
most of data will re-read soon
exception
backup
just read & write once
IO device
all 3 kinds of resources are system-wide
reality
three resources
RAM becomes bigger & bigger
RAM is expensive than disk
CPU cores becomes more & more
disk becomes much quicker
ideas
HIPRI optimization
multi-page bvec follow up
bio_add_page() in iomap/fs
change fs/block_dev.c into iomap