Open Stack

> -----Original Message-----
> From: Julien Danjou [mailto:julien at danjou.info]
> Sent: Montag, 28. August 2017 16:03
> To: Bedyk, Witold <witold.bedyk at est.fujitsu.com>
> Cc: OpenStack Development Mailing List (not for usage questions)
> <openstack-dev at lists.openstack.org>
> Subject: Re: [openstack-dev] [Monasca] GridDB repository support
> 
> On Mon, Aug 28 2017, witold.bedyk at est.fujitsu.com wrote:
> 
> > Can you give some reference to Gnocchi's performance measurements?
> > I have seen you presentation from the last Summit [1] but I cannot
> > find information about Gnocchi's write throughput. In the issues slide
> > you note the slow post API. The API responsiveness test shows that the
> > API can handle correctly up to 2500 instances. You collect around 60
> > metrics per instance and use polling interval of 10 minutes. That would
> translate to:
> >
> > 2500 instances * 60 metrics/instance / 600 s = 250 metrics/s
> >
> > Are these assumptions correct?
> 
> It really depends on the number of measures you include in each of your API
> requests. Ceilometer does very little batching.
> 
> If you batch those measurements, you could achieve up to 120k measures/s
> in Gnocchi already 2 years ago¹. Considering the various improvement we did
> over that timeframe, you can likely go higher.
> 
> And since it scales horizontally, you can double the request handling by
> adding another node. :)
> 
> I'll probably try to do some benchmark one of these days, but I don't have
> proper hardware to do so for now. :)
> 
> ¹  https://julien.danjou.info/blog/2015/gnocchi-benchmarks


The number looks good but the test you have referenced was performed with the file storage driver which is not really scalable. It would be great if you could provide numbers for the Ceph backend.
Please keep us updated.

Cheers
Witek