Re: [PATCH v2 4/4] migration/calc-dirty-rate: tool to predict migration time

[gudkov.andrei]

On Tue, May 30, 2023 at 11:21:38AM +0800, Wang, Lei wrote:
> On 4/27/2023 20:43, Andrei Gudkov via wrote:
> > +    deadline constraint.
> > +
> > +    :param model: `MemoryModel` object for a given VM
> > +    :param bandwidth: Bandwidth available for migration [bytes/s]
> > +    :param downtime: Max allowed downtime [milliseconds]
> > +    :param deadline: Max total time to migrate VM before timeout 
> > [milliseconds]
> > +    :return: Predicted migration time [milliseconds] or `None`
> > +             if migration process doesn't converge before given deadline
> > +    """
> > +
> > +    left_zero_pages = model.num_zero_pages()
> > +    left_normal_pages = model.num_total_pages() - model.num_zero_pages()
> > +    header_size = 8
> 
> In the cover letter: "Typical prediction error is 6-7%". I'm wondering if the
> 6-7% is less or more than the real migration time. I think 2 potential factors
> will lead to less estimation time:
> 
>   1. Network protocol stack's headers are not counted in, e.g., TCP's header 
> can
>      be 20 ~ 60 bytes.
> 
>   2. The bandwidth may not be saturated all the time.

Yes, I totally missed network headers in my model!

I do not see any noticeable assymetry in prediction error. Sometimes
prediction overestimates actual migration time, sometimes -- underestimates.
I think that prediction error is caused mainly by VM workload not being
perfectly stable over time.