Re: [MirageOS-devel] Tracing and profiling blog post

[Richard Mortier <Richard.Mortier@xxxxxxxxxxxxxxxx>]

On 30 Oct 2014, at 15:55, Anil Madhavapeddy <anil@xxxxxxxxxx> wrote:

> On 30 Oct 2014, at 09:36, Thomas Leonard <talex5@xxxxxxxxx> wrote:
>> 
>> On 28 October 2014 22:36, Daniel Bünzli <daniel.buenzli@xxxxxxxxxxxx> wrote:
>>> 
>>> Le mardi, 28 octobre 2014 à 23:21, Anil Madhavapeddy a écrit :
>>> 
>>>> Assemblage is still undergoing 'rapid iteration', so I'd suggest
>>>> switching to OASIS until we have a more stable version. ThomasG or
>>>> Daniel Buenzli can correct me if that's not accurate.
>>> 
>>> That's accurate. If we agree with Thomas on the PR I'm going to make next 
>>> week, the current API completely changes and assemble.ml files will need 
>>> rewriting. We need time to design a great system and it's better if we 
>>> don't encumber ourselves with users at the moment.
>> 
>> OK, done (thanks to Thomas for some OPAM help!). I think the main
>> thing now is to agree the API for libraries to record tracing events.
>> 
>> Here's what I have so far:
>> 
>> https://github.com/talex5/mirage-profile/blob/new-api/lib/trace_stubs.mli
>> https://github.com/talex5/mirage-profile/blob/new-api/lib/counter.mli
> 
> Does note_suspend/note_resume here refer to dropping in the scheduler (the
> equivalent of blocking the domain)?  It could easily be conflated with
> VM suspend/resume (which is a higher-level protocol).  The actual state
> transition is "waiting for I/O", so perhaps we should name it after that?

would've thought having a single parameterised function is better than being 
too specific?

> To echo Mort's comment, istr in Xenstore that there was a problem with many
> duplicate strings for labels, so it has a module somewhere that creates
> a Symbol.t that dedups similar strings, and also has the benefit that
> comparison can be done via fast equality rather than byte-by-byte checking.
> It also allows errors to be flagged up if (for example) duplicate symbol
> names are used in different programs.

yes; another point on similar lines is that it's good to keep things like 
labels as unique as possible as long as possible -- let the event consumer 
express any reconciliation between events of different types/in different 
subsystems that they want. (this was what allowed us in magpie to use identical 
sqlserver and iis instrumentation to produce performance models for both TPC-W 
and TPC-C workloads.)


-- 
Cheers,

R.

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail

_______________________________________________
MirageOS-devel mailing list
MirageOS-devel@xxxxxxxxxxxxxxxxxxxx
http://lists.xenproject.org/cgi-bin/mailman/listinfo/mirageos-devel