Web Hosting Talk : Web Hosting Main Forums : Cloud Hosting : Any Cloud Provider that does Vertical Dynamic Scaling?

[polarisie]

Hi

I would like to know if there are providers out that there that offers cloud platforms that are capable of doing vertial autoscaling that does not require reboots.

This means that instead of scaling out horizontally (creating new instances) is it possible to scale the resources like CPU, RAM, network, etc without rebooting the server ?

Thanks

[nwmcsween]

I don't know of any, you would need to distribute everything or use some sort of single system image system like openssi or openmosix. As for the distribute everything you could do this but as of right now Linux can't, it can sort of but not really with commodity hardware. Someone may prove me wrong though.

[polarisie]

I recall reading somewhere that Citrix Xenserver does vertical scaling?

[FastServ]

Originally Posted by polarisie

I recall reading somewhere that Citrix Xenserver does vertical scaling?

It will up to the limits of the host node... I don't know of any solution that dynamically scales a single instance beyond that.

[chennaihomie]

You can do this on OnApp based clouds for sure. But it would be limited to max the node's resources.

[polarisie]

Originally Posted by chennaihomie

You can do this on OnApp based clouds for sure. But it would be limited to max the node's resources.

Do you know if OnApp does an inplace/same server scale up or does it do via a live Migration to an alternate hardware with processing power? And when the resources utilization drops is it able to vertically scale down?

Cheers

[CRego3D]

Originally Posted by polarisie

Do you know if OnApp does an inplace/same server scale up or does it do via a live Migration to an alternate hardware with processing power? And when the resources utilization drops is it able to vertically scale down?

Cheers

It will do the dynamic scaling on the same node if it detects enough resources for it, if it does not, then it looks for a node that does, initiates a live migration of the VM to the said node and dynamically scales it then.

And yes, scale down is also possible, in that case the Vm will be kept on the current node.

[brentpresley]

Why not do an SLA LAMP stack? You can scale one of those well beyond the limits of one node, and it only depends on traffic from the load balancer (which can also be made redundant).

In AppLogic, you can set your application to turn on a new instance of the server when your first server(s) hit a certain pre-set capacity (say 80% of optimal load), and then scale back when resources are no longer in use (say when load is less than 20% of optimal on each web server). Works beautifully.

[polarisie]

Hi Brent

What you described sounds like classic Horizontal scaling. In order to take advantage of this the application would have to be architected to support this.

However, with vertical scaling you could technically move any traditional app to the cloud without changes in code and have that scaled up and down (but ultimately limited by the underlying hardware)

Cheers

[polarisie]

Originally Posted by CRego3D

It will do the dynamic scaling on the same node if it detects enough resources for it, if it does not, then it looks for a node that does, initiates a live migration of the VM to the said node and dynamically scales it then.

And yes, scale down is also possible, in that case the Vm will be kept on the current node.

Carlos

As a cloud provider can this option be configured and set as a billable action ? Meaning can we bill the client if he/she would like to scale upwards (vertically) ? Are there any specific metrics for this?

Cheers

[brentpresley]

Yes, technically it is horizontal scaling. Vertical scaling beyond one node is just not possible until you get node interconnect speeds much much higher (even prototype 80Gb/s Infinniband would not be enough because the latencies would be too high to allow for RAM reads/writes between the nodes in a reasonable time).

This is a technical problem, and one that I forsee not going away at least in the next 5-10 years).

[lifewithcause]

Originally Posted by polarisie

Hi

I would like to know if there are providers out that there that offers cloud platforms that are capable of doing vertial autoscaling that does not require reboots.

This means that instead of scaling out horizontally (creating new instances) is it possible to scale the resources like CPU, RAM, network, etc without rebooting the server ?

Thanks

Cloud is NOT a real cloud if it does not "vertical scale". With Microsoft's DDTK + System Center + Hyer-V technology combination you can build cloud infrastructure that can scale horizontal or vertical without the need of reboot or downtime. I suggest you search for Microsoft partners for such a cloud.

[eming]

Originally Posted by polarisie

Carlos

As a cloud provider can this option be configured and set as a billable action ? Meaning can we bill the client if he/she would like to scale upwards (vertically) ? Are there any specific metrics for this?

Cheers

yup, OnApp takes care of billing of the scaled resources automatically.

Originally Posted by brentpresley

In AppLogic, you can set your application to turn on a new instance of the server when your first server(s) hit a certain pre-set capacity (say 80% of optimal load), and then scale back when resources are no longer in use (say when load is less than 20% of optimal on each web server). Works beautifully.

ya, OnApp does that as well. We work with two types of autoscaling.

1) Scale UP
Basically adds resources to the single as needed. OnApp keeps a tally of the usage and ensures your clients are charged for the extra resourced utilised during scaling. Scaling is typically without reboots (windows does not like this).

2) Scale OUT
When scaling out is enabled OnApp will keep an eye on your VM and grow it as much as possible within the confinement of a single physical server. And then add another physical server based on a clone of the first one or a defined template. OnApp's integrated Loadbalancer will take care of the distribution of traffic.

Originally Posted by polarisie

What you described sounds like classic Horizontal scaling. In order to take advantage of this the application would have to be architected to support this.

However, with vertical scaling you could technically move any traditional app to the cloud without changes in code and have that scaled up and down (but ultimately limited by the underlying hardware)

Cheers

Well, typically, in scale OUT (horizontal scaling) clients would have a setup with a DB cluster and they are aiming to add webservers as needed using the autoscaling features.
I guess this could be the same setup in AppLogic.


D

[polarisie]

Originally Posted by lifewithcause

Cloud is NOT a real cloud if it does not "vertical scale". With Microsoft's DDTK + System Center + Hyer-V technology combination you can build cloud infrastructure that can scale horizontal or vertical without the need of reboot or downtime. I suggest you search for Microsoft partners for such a cloud.

Thank you for that bit of information. Will look into it laer.

However in practical terms, I am not quite so sure as to how the MS combination would handle vertical scaling from an IaaS perspective (i.e Amazon EC2 type of platform) but I can understand how this can be done from a PaaS perspective which is already available in Azure. Even their latest(beta) addition of VMRole is actually a Paas on a IaaS.

In addition from what i understand the underlying Microsoft operating system only has support for features like hot CPU/RAM add with no option of HOT remove without requiring a reboot which would make vertical scaling down difficult. So technically i don't see how they would scale in place. The only option would be to live migrate the instance to another node and vice versa for scaling down.

Ps: Diagonal scaling would be a nice option (scale up and then out)

Cheers

[aodat2]

If I'm not mistaken, I think that Xen is able to scale vertically as well. Not too sure about the rebooting part of it but I'll check.