Bug 6940

Summary: microLinux: Characterize and reduce dynamic kernel memory usage
Product: [Build System, Metadata & Runtime] Meta-yocto Reporter: Tom Zanussi <tom.zanussi>
Component: meta-yoctoAssignee: Tom Zanussi <tom.zanussi>
Status: RESOLVED OBSOLETE QA Contact:
Severity: normal    
Priority: Undecided CC: jessica.zhang, poky.bs.watcher, poky.watcher
Version: 1.8   
Target Milestone: 1.8 M3   
Hardware: x86   
OS: Multiple   
Whiteboard:
OS type for building Yocto: --- Type of Regression: ---
Verified: Documentation change: Don't know

Description Tom Zanussi 2014-11-07 20:52:54 UTC
In the initial passes of kernel memory usage reduction, various dynamic memory tools were used to get an idea of the major contributors to kernel dynamic memory usage, and where cuts could be made to reduce it.

As part of this analysis, a tracing tool based on ftrace and the kernel trace event subsystem, Linux hash triggers, was created:

http://lwn.net/Articles/592311/

It provided some initial data and direction that led to some additional size reduction features such as PROC_MIN.

There was also an 'early hashtriggers' addition to that patchset that allowed very early contributors to dynamic memory usage to be recorded in the same way.  The results of that analysis can be found in these slides:

http://elinux.org/images/5/54/Tom.zanussi-elc2014.pdf

Basically, the results of the early hashtrigger analysis showed that there are a lot of control paths that contribute to dynamic memory usage during early boot, but the total memory usage of all that is small compared to, for instance, the network stack.

So we should revive at least the normal hashtriggers patchset and use it along with other tools to drill down into memory usage of a normally running system using the native network stack, since that's probably our most important use case in these IoT days.
Comment 1 Jessica 2015-02-09 23:59:05 UTC
MicroLinux is moved out of team