[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: autoscaling



On Tue, 8 Jan 2013, Diego Spinola Castro wrote:

> Mike, is very clear to me but lets go deeper. Sorry for annoying you but i need to be ready for dummy questions !In a tradicional environment i would use larger servers
> because i know apache needs more memory for start more clients and to be able to handle more sessions. Lets talk about Openshift PaaS, why an app wrote for handle scale will
> run better splited in small gears than larger one since apache is the same?
>

The theory is it won't run better, it'll run the same.  But you'll get
significant cost savings when increasing and decreasing gears when
compared to a traditional environment where you typically have to plan for
highest capacity even though you may only hit it 5% of the time.

	-Mike

>
>
> 2013/1/8 Diego Spinola Castro <spinolacastro gmail com>
>       Much more!thanks
>
>       2013/1/8 Mike McGrath <mmcgrath redhat com>
>             On Tue, 8 Jan 2013, Diego Spinola Castro wrote:
>
> > I'm quite confuse, an apache gear with 512mb is able to handle the same number as 1gb gear? if is true, why wouldn't increase apache max_clients conf ? Just
> doing the devil's question. Wondering some
> > customer asking: Why wouldn't use larger gears?
> >
>
> Most people don't have access to larger gears (unless they've emailed us
> at openshift redhat com and requested them) and ultimately when the
> commercial offering is complete, they'll also cost more than the small
> gears.
>
> In terms of increasing max_clients, in traditional computing that's what
> you would do.  With our PaaS we're working with lots of small building
> blocks.  For example, instead of 1 1G gear, our architecture would
> recommend two 512M gears.  The goal is lots of small computing resources
> working together, not few large ones.
>
> We understand though that many applications aren't written with this new
> architecture in mind yet which is why we also offer the larger gears at
> the application tier.
>
> Databases are a whole other discussion but I think the need for larger
> gears there is more obvious.
>
> Does this make more sense to you?
>
>         -Mike
>
> >
> > 2013/1/8 Mike McGrath <mmcgrath redhat com>
> >       On Tue, 8 Jan 2013, Diego Spinola Castro wrote:
> >
> >       > Hi guys, i'm wondering how openshift handles cartridges sizes when autoscaling, looking into haproxy_ctld.rb code i found that autoscaling is based on
> current sessions and it's hardcoded by
> >       10. What
> >       > happens if i use 1gb cartridges or even 2gb? Does openshift knows that it's a larger cartridge ?
> >       >
> >
> > We've been looking for some more real world examples to get better
> > baselines for these numbers.  Keep in mind you really shouldn't be using
> > larger gears for more users.  There's some work to be done here to make it
> > great but you should think of larger gears as the same number of users
> > needing more memory per user.
> >
> >         -Mike
> >
> >
> >
> >
>
>
>
>
>

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]