Alluxio randread with pool performance

[maobaolong]

Alluxio Version: Master branch

Describe the bug

Alluxio randread with pool performance

To Reproduce

[global]
direct=1
fallocate=none 
#ioengine=libaio 
group_reporting
time_based
size=10m 
nrfiles=50 
iodepth=1 
stonewall
runtime=600

[bs64k_job32_read]
filename=testfile/bs64k_job32_read
bs=64k
numjobs=32
rw=randread

Expected behavior

Not too bad against sequential read.

Urgency

Urgent, this is a blocker for push alluxio into some of our AI cases

Are you planning to fix it

Not sure, but will have a try.

Additional context NO

[maobaolong]

@beinan @HelloHorizon Would you like to help to solve this issue? Thanks.

[dbw9580]

the issue is more centered around the chunk in transmission, rather than block or page at the worker's side. The client caches the chunk as an internal buffer, but in case of a small random read, the chunk can still be too large, causing a read amplification.

[LuQQiu]

Check whether can set per file chunk size AlluxioFuse detects whether sequential -> pre-read or random -> read exact

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in two weeks if no further activity occurs. Thank you for your contributions.