Bug 4377

Summary:	[Performance] The switch to turn off the generation of .sigdata file
Product:	[Build System, Metadata & Runtime] BitBake	Reporter:	Robert Yang <liezhi.yang>
Component:	bitbake	Assignee:	Robert Yang <liezhi.yang>
Status:	VERIFIED WONTFIX	QA Contact:
Severity:	enhancement
Priority:	Medium	CC:	alexandru.c.georgescu, mark.hatle, poky.bs.watcher, poky.watcher, richard.purdie, sgw
Version:	1.5
Target Milestone:	1.5 M3
Hardware:	x86
OS:	Multiple
Whiteboard:
OS type for building Yocto:	---	Type of Regression:	---
Verified:		Documentation change:	---

Description Robert Yang 2013-04-22 12:32:36 UTC

The .sigdata file isn't usually used by the end user, we can generate it by default, but it would be good if we have a switch for it, for example, BB_NO_SIGDATA, this would not save much disk space (only about 120M for a core-image-sato build), but it would save a mount of disk I/O, when many people (or multi builds) works on the same sever, the disk I/O is usually the bottleneck, the build would cost much time on I/O waiting, reduce the disk I/O would save time.

Comment 1 Richard Purdie 2013-04-22 12:45:41 UTC

The sigdata files are there to allow for debugging of the system. Whilst they may not be commonly used, they are invaluable in trying to debug what went into a given checksum and what might have changed.

I seriously doubt these have much impact on overall build time. I'd want to see statistics proving this was some kind of build bottleneck before any such patch was taken.

Please feel free to provide supportive evidence of this being a problem but right now I'm not convinced it is (quite the opposite) so I'm marking the bug accordingly.

Comment 2 Richard Purdie 2013-04-22 12:46:39 UTC

As an aside, if I/O is a worry, we could compress them. Again, I'd like to see numbers showing this is a real problem first though.

Comment 3 Robert Yang 2013-04-23 08:21:19 UTC

(In reply to comment #2)
> As an aside, if I/O is a worry, we could compress them. Again, I'd like to
> see numbers showing this is a real problem first though.

Hi RP, I've done several builds without the generation of .sigdata and python's temp/run.*, but it doesn't make much difference, only about 40 seconds saved for a core-image-sato build (the host's load is not high, the whole builds cost 156m53.460s), so drop this one is the right way.

Comment 4 Alexandru Georgescu 2013-05-24 13:19:16 UTC

Verified as per above comments.