On Wednesday, 29 March 2017 at 5:47 PM, Martin Kletzander wrote:

On Wed, Mar 29, 2017 at 05:31:42PM +0800, Eli Qiao wrote:


--
Best regards
Eli

天涯无处不重逢
a leaf duckweed belongs to the sea, where not to meet in life


OK, please do something with your client. Having the footer here on top
for every reply is *sooooo* bothersome when you are replying inline
(that part is fine).

Sorry, I removed footer, better now?
 
On Wednesday, 29 March 2017 at 5:19 PM, Martin Kletzander wrote:

On Wed, Mar 29, 2017 at 04:22:16PM +0800, Eli Qiao wrote:


--
Best regards
Eli

天涯无处不重逢
a leaf duckweed belongs to the sea, where not to meet in life



On Wednesday, 29 March 2017 at 3:45 PM, Martin Kletzander wrote:

On Tue, Mar 28, 2017 at 03:22:34PM +0800, Eli Qiao wrote:
hi Martin

(cc libvir-list)

I am a little confused about cat support.

I am currently rebasing my code on top of pre-cat branch from your private github repo, today when I check it you have removed it and create a cat branch and there are some related code pushed[1], can I know what ’s your plan for my patch set for CAT support ? should I continue my rebasing work? your though?

So we can work together on that. Since the rework of the sysfs
functions, some patches are easier to write from scratch then rewrite,
but I'm now just trying to setup the test suite, so that we have
something to test on, at least some of the code. So where are you in
the rebase right now? Do you think anything from the virsysfs.c code
could be enhanced?




Not so fast, only the first patch [1], I found that nodeinfo.c is removed :(

I think we need to extend virResCtrlGetInfoStr and virResCtrlGetInfoUint to virsysfs.c

thought ?

Yeah, we should wrap around /sys/fs/resctrl as we do with
/sys/devices/system so that it can be easily tested.
Sure, working on it, and done, will push it for review.

Also I will push some fake data for resctrl testing..



Also I got another idea about keeping the resource info. There is no
need for any global data to be stored as you are re-reading almost all
of it. The only info that stays the same is caches (that is already
saved in capabilities) and what caches are available for resource
control (that will be there as well). So I don't think we need yet
another global data storage.

Do you mean, we re-create all struct (reading them from /sys/fs/resctrl) when we create/destroy instance?
also, for get free cache ?

You have to update that for every request anyway, so what's the point of
keeping the data when they immediately become old?

I was thought that may reduce the time costing, not all of the content be refreshed, anyway, I will try to avoid global files in my later version. 

LoL lots of rebasing  :( 

Thanks for your suggestion.
This is what I did in my early PoC, that will much easier… but please keep in mind that only one thread can read/write to /sys/fs/resctrl at one time.

Yeah, that's what we have locks for.

the neck bottle is /sys/fs/resctrl

Sure you mean bottleneck, right? :)
yes, bottleneck,