performance metric

[JN]

I don't know about the other buyers out there, but I find the number of CPUs shown in the market offers to be very misleading.  I have a CPU-intensive job and have tried lots of different hardware offered by different providers with anything from 1 to 8 CPUs.  But some of the 8 CPU systems are just dog-slow, while some of the 1 CPU systems zip right along.  Its hard to evaluate which offer is a better price without starting one up and testing it out.

Is is possible to get some sort of performance rating instead of just # CPUs?  Maybe FLOPS or a standard benchmark?

Buyers, how do you feel about this?  Has it been an issue for you too?

Spotcloud, please consider this a feature request!

Jonathan

[Adam Boduch]

Hi,

Thanks a lot for the observation.  Did you happen to notice how the SpotCloud provider rating correlated to the performance of your application?  If so, this would be useful information for us.

About the CPUs.  The offerings you see in the market are advertising virtual CPUs allocated to your virtual machine.  This means that anything running on your instance will detect, say 8 CPUs.

As to whether this is a selling point for the hardware offering is really context dependent.  If your application isn't taking advantage of several CPUs, there is no real advantage.  If your application will utilize all CPUs found in the system - I'm assuming yours will as you mentioned it is CPU-intensive - then might be something to gain from hardware with > 1 CPU.

I say "might" because the number of virtual CPUs has no bearing on how much of the host's CPU time is given to the virtual instance.  This has more of an impact on performance than any other factor I'd say.

Having said that, I think there is merit to your feature request.  Obviously providers actually outperform others - with the metrics (rating) to prove it - our valuable to buyers such as yourself.  We'll certainly put this on our radar, and thanks again for pointing it out.

Regards,

Adam

[JN]

Thanks for the reply.

No, the provider rating did not seem to have any relationship.  (Although I thought that the rating was just a measure of how often a pending instance did or did not result in a running instance.  Is this not the case?)

Jonathan

[Adam Boduch]

You're correct, failed instances do impact the provider rating.  Negatively.

However, there are a few other factors that comprise the rating such as available resources and how stable their API is.