The fork in the road Lustre upstream client EMC started the work spring 2013 Based on Lustre commit faefbfc03 24 version No new features added Most patches address style issues Removed wrappers ID: 830315
Download The PPT/PDF document "Lustre + Linux Putting the house in or..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Lustre + Linux
Putting the house in order
Slide2The fork in the road
Lustre
upstream client
EMC started the work spring 2013
Based on
Lustre
commit faefbfc03 (~ 2.4 version)
No new features added
Most patches address style issues
Removed wrappers
Gripes about architectural design
Slide3Trip down Intel drive : Syncing with upstream client
Participants
:
Historically
Peng
Tao from EMC
James Simmons from ORNL
John Hammond and Dmitry
Eremin
from Intel
Frank
Zago
from
Cray
Goals:
Remove technical debt
Meet kernel coding styles
Remove wrappers
Support newer kernels
Slide4Master renovations so far
Support up to 3.15 kernels
Port of
proc
fs
handling to
seq_file
All
cfs
_* wrappers are gone
Liblustre
removal
Kernel coding styles enforced with new code
Slide5What about the other side of town?
All code cleanup for clients was applied to servers
LU-20 : No more patching the server side
LU-3406 : raid5
mmp
unplug patch.
LU-684 : Use
dm
flakey to test fail over
Remove the rest of the patches.
Only need to patch servers with LU-3406 for testing
LU-6141: Push most
ldiskfs
patches upstream
Slide6What needs to be done to be up to code?
Needs to be in style.
LU-5396 : kernel sparse cleanup
LU-5478 :
typdef
removal
LU-5577 : unsigned/signed fixes
LU-5710 : Grammar and spelling fixes
LU-5829 : Remove unneeded exports
LU-6142 : Kernel indenting of code
a.k.a
tabathon
.
Slide7With barn!!!!
Some technical debt is still left
LU-2675 : Remove
userland
libcfs
and
lnet
code
LU-5712 :
libcfs
train wreck.
Userland
and kernel header entanglement
Userland
and kernel source code entanglement
Abstractions still left.
Slide8Code violations
Upstream reviews dislike some code choices
Job ID handling
Random seed handling
General dislike of
procfs
Others not named
After syncing we need to address these issues
Slide9Conclusion
Work left
Style issues should be ready by end of 2.8
Libcfs
needs a lot of work
Kernel code style updates will be big update.
Need to work as a community to push
work upstream.
We can’t forget support
for latest
kernels.