[LSF/MM TOPIC][ATTEND] cleancache extension and memory checkpoint/restore

[LSF/MM TOPIC][ATTEND] cleancache extension and memory checkpoint/restore

[Pavel Emelyanov]

Hi.

I'd like to attend the event to discuss the following topics:


1. cleancache extension

In containerized systems, when containers are more or less equal to each
other, we can save RAM and (!) disk IOPS if we share equal files between
containers. We've been using a unionfs-like approach and faced several 
disadvantages of it (I can describe them in details if required).

Now we're working on extending the cleancache subsystem to achieve this 
sharing. The cost of fully isolated filesystems is very high, I can provide 
numbers of various performance experiments, thus this is required badly for
containers.


2. memory checkpoint/restore

Yet another project I'm working on is CRIU -- checkpoint-restore in userspace.
One of the problems we've met is -- it's impossible to determine a task's 
current working set from the usepspace.

Another problem of the nearest future is the ability to create the memory snapshot
of a running task for more efficient migration and incremental checkpoint.


Thanks,
Pavel

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [LSF/MM TOPIC][ATTEND] cleancache extension and memory checkpoint/restore

[Dan Magenheimer]

Hi Pavel --

> From: Pavel Emelyanov <xemul () parallels ! com>
> 1. cleancache extension

> In containerized systems, when containers are more or less equal to each
> other, we can save RAM and (!) disk IOPS if we share equal files between
> containers. We've been using a unionfs-like approach and faced several 
> disadvantages of it (I can describe them in details if required).

> Now we're working on extending the cleancache subsystem to achieve this 
> sharing. The cost of fully isolated filesystems is very high, I can provide 
> numbers of various performance experiments, thus this is required badly for
> containers.

I agree this is a great use of cleancache!

FYI, the Xen implementation of transcendent memory optionally
does deduplication.  I think you are probably doing something
very similar so you may be able to leverage some code.  No changes
to cleancache are needed... you just need to register a different
"backend" (e.g. like zcache or RAMster or the Xen tmem stubs).

IIRC, the Xen implementation is much simpler than KSM because
the candidate pages are more closely managed.

Hopefully we can talk about it at LSF/MM!

Dan

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>