High Availability Setups

[poetics5]

I'm currently with linode, who are great by the way, but would like to move to the vps cloud. I have a number of questions first that I haven't been able to find answers to:

1. Will private ip's be metered or unmetered?
2. How many network interfaces will be created if a machine has a private ip?
3. Is it possible to create and use my own dns / dhcp server for my vpses to use?
4. Will I be able to take over and ip [ heartbeat ] in the case of failover?
5. How is automated scaling in the event of a spike charged? just whats used or flat fee?
6. Is the scheduled scaling feature already integrated, and how is this billed?
7. Where are the U.S. servers located
8. Is it possible to setup a low level cdn with your service or will I be limited to one (two counting the uk) geo locations?
9. Will any additional security be provided for private ips? (I understand I'll need to secure my boxes either way)

Looking forward to your reply
Greg

[poetics5]

Also will it be possible to share a public ip across vpses (CLUSTERIP support)?

[Ditlev]

Thanks for your post Greg - good one, I've answered below, but some of the other guys might add more, and if you'd like more info feel free to post a followup!

Originally Posted by poetics5

Will private ip's be metered or unmetered?

Private IP's will be unmetered for now.

Originally Posted by poetics5

How many network interfaces will be created if a machine has a private ip?

Private ip's will reside on addt virtual/alias nics, such as eth0:1

Originally Posted by poetics5

Is it possible to create and use my own dns / dhcp server for my vpses to use?

Yes, you can setup your own DNS - we will actually provide images for it at some stage soon - but you can also utilize the DNS management tools in the backend system of VPS.net (soon to come)

Originally Posted by poetics5

Will I be able to take over and ip [ heartbeat ] in the case of failover?

Our setup is selfhealing. If a physical hypervisor fails, all the clients on that box will be distributed across the cloud's other hypervisors. Since no content is locally stored the migration would be less than a minute (tests show max 40 seconds) of downtime.

You can also use the HAproxy images (http://vps.net/wiki/load-balancer) that we provide as an image to do a heartbeat like failover between your individual servers - just like you could with a normal dedicated server setup.

Originally Posted by poetics5

How is automated scaling in the event of a spike charged? just whats used or flat fee?

It's being finalized these days, and I expect it to be an optional addon where clients can ask VPS.NET to add x nodes if overall utilization of current server is yy% for more than zz mins.
You'd be charged for the additional nodes added.

Originally Posted by poetics5

Is the scheduled scaling feature already integrated, and how is this billed?

Not implemented yet. It will be billed as an 'one-time-fee' per 24 hours - we have the technology to do billing per hour, but we really want to keep things simple and easy to overview.

Originally Posted by poetics5

Where are the U.S. servers located

The US cloud is in our DC in Salt Lake City. It's an excellent setup. It was #1 most reliable hosting provider according to Netcraft four times in 2008, including November and December.
The UK cloud is on the UK2 infrastructure - currently second rated in the world by Netcraft.

Originally Posted by poetics5

Is it possible to setup a low level cdn with your service or will I be limited to one (two counting the uk) geo locations?

We are strong believers in CDN's being fully integrated/mounted parts of the cloud infrastructure, and we are working on several models to make that happen. Expect something CDN'ish integrated within the next 60 days.

Originally Posted by poetics5

Will any additional security be provided for private ips? (I understand I'll need to secure my boxes either way)

As for security for private ip's, they will be confined to the vlans
of course anybody inside the vlan can try to attack you, so you must protect your vps. We will provide 'secured' images that you can launch knowing they are pretty well locked up.

Great questions Greg - and again, please feel free to post followup questions.

We also like Linode, heard a lot of good about them. Our approach is somewhat different though, at Linode you buy a virtual server at VPS.NET you buy a bunch of dedicated resources that you can fully allocate to one server, or split up into many smaller servers - and reallocate at any time.


D

[Ditlev]

Originally Posted by poetics5

Also will it be possible to share a public ip across vpses (CLUSTERIP support)?

Well, yes - but only via the use of a setup like HAproxy for now. We would be 6-12 months away from anything more advanced than that (it is on our roadmap though).


D

[poetics5]

Hi ditlev,

First off I'm blown away that both the ceo responds to forum posts and that you actually know the geek stuff :P

As far as sharing an ip, I'm referring to the CLUSTERIP feature in iptables. I guess the real question would be are ip addresses locked into one vps, or are they free to be used by any of the machines we create? I'm looking to load share my load balancers in a failover configuration. It allows me to fully utilize my load balancers in an active / active / active cluster allowing me to have multiple failures (a requirement for our new network)

I like HAProxy and it will be apart of network, but I also require failover support.

We are strong believers in CDN's being fully integrated/mounted parts of the cloud infrastructure, and we are working on several models to make that happen. Expect something CDN'ish integrated within the next 60 days.

Thats great to hear, can you provide any details? Will you be utilizing a third party service or establishing space in addition data centers just for low level cdn users to utilize for their networks?

Originally Posted by poetics5 View Post
Is the scheduled scaling feature already integrated, and how is this billed?

Not implemented yet. It will be billed as an 'one-time-fee' per 24 hours - we have the technology to do billing per hour, but we really want to keep things simple and easy to overview.

I'm definitely looking forward to this, as I'm kinda anti EC2 (I don't agree with how they handle blackouts and hate the concept of non persistent machines), this would allow me to handle our heavy lifting without leaving home so to speak

Our setup is selfhealing. If a physical hypervisor fails, all the clients on that box will be distributed across the cloud's other hypervisors. Since no content is locally stored the migration would be less than a minute (tests show max 40 seconds) of downtime.

How does this handle single instance failures. My primary goal is to eliminate single points of failures. Will the cloud detect a service going down on an instance, or if the xen instance crashes? I know you can set xen to automatically reboot on crash, but sometimes theres a bigger problem than requires a hand off. I guess my question is does your system support any method for preventing single points of failure?

Originally Posted by poetics5 View Post
Is it possible to create and use my own dns / dhcp server for my vpses to use?

Yes, you can setup your own DNS - we will actually provide images for it at some stage soon - but you can also utilize the DNS management tools in the backend system of VPS.net (soon to come)

I should clarify, I'm wondering if it's possible to create a dns / dhcp server in order to establish a private lan for my vpses to use, or will i be restricted to the private ips provided by vps.net?


Pardon me if I seem like I'm expecting the world, I've just been watching your project closely, and really want to present your cloud as our provider in our new network spec. I understand everything is still getting off the ground and your team has been seriously moving faster to improve based on feedback than I think anyone expected. I'm sure there are many others lurking and drooling as I have been at the idea of using your cloud once it's time to upgrade their infrastructure.

If there is someway to develop an infrastructure without single points of failure I can make the rest work. If not add me to the list of people drooling to use the service once CLUSTERIP and heartbeat support is added.

Greg

[poetics5]

Also if there is anything I can do to help test these features, or references needed please contact me directly. I'd be more than happy to help you, as I'm sure there are lots of people wanting these features

Greg

[nullmind]

Hello poetic5 .. let me assist you also with some of those questions

I guess the real question would be are ip addresses locked into one vps, or are they free to be used by any of the machines we create? I'm looking to load share my load balancers in a failover configuration. It allows me to fully utilize my load balancers in an active / active / active cluster allowing me to have multiple failures (a requirement for our new network)

At the moment we have a system in place that prevents a VM from using an IP assigned to another VM, this will stay in place for external ip's as prevents people from taking over somebody's else IP.

For shared IP's we intend to do something different, you will be able to request a range if internal IP's for your account (not VM specific) and we will unlock those to be usable on any of your VM's .. this also means 2 of your VM's could share an internal IP so you can utilize CLUSTERIP and other tools like it.

(RE: CDN) Thats great to hear, can you provide any details? Will you be utilizing a third party service or establishing space in addition data centers just for low level cdn users to utilize for their networks?

We will be utilizing the CDNLayer, who is nothing other than Internap CDN, it Supports Flash Media and Windows Media for streaming files. CDNLayer also supports a wide array of encoding formats for downloading or caching media—including Windows Media Player, Flash Media, DivX, H.264, Move Media Player, Microsoft Silverlight, QuickTime, MP3, RealSystem G2, RealPlayer, Real Networks, HTML, TXT, GIF, JPG, and PDF.

CDNLayer currently delivers content across 21 nodes across the globe: 15 nodes in the Americas, 4 in Asia, and 2 in Europe.

I know you can set xen to automatically reboot on crash, but sometimes theres a bigger problem than requires a hand off. I guess my question is does your system support any method for preventing single points of failure?

If XEN itself crashes, our system detects it in less than 5 seconds and starts booting the VM's elsewhere in the cloud, as for single VM's .. we have disabled the ability to reboot on crash, as tests shown the majority of times the problem persists and the VM just continued on reloading itself over and over, for that scenario it's the user's task to keep it monitored.

I should clarify, I'm wondering if it's possible to create a dns / dhcp server in order to establish a private lan for my vpses to use, or will i be restricted to the private ips provided by vps.net?

DHCP servers are notoriously hard to secure, in order for that to work you will have to use the private ip's we gave you as the only ones the DHCP server can relay, and more you will have to lock the permissions to only give a IP to your other VM's MAC addresses .. otherwise people can hijack one of your ip's .. it's up to the user to configure that properly (but we will look into having a DHCP image in the future also )

Pardon me if I seem like I'm expecting the world

Dont worry, we stride to provide the world and the moon as a bonus to our customers

[poetics5]

I guess the real question would be are ip addresses locked into one vps, or are they free to be used by any of the machines we create? I'm looking to load share my load balancers in a failover configuration. It allows me to fully utilize my load balancers in an active / active / active cluster allowing me to have multiple failures (a requirement for our new network)

At the moment we have a system in place that prevents a VM from using an IP assigned to another VM, this will stay in place for external ip's as prevents people from taking over somebody's else IP.

For shared IP's we intend to do something different, you will be able to request a range if internal IP's for your account (not VM specific) and we will unlock those to be usable on any of your VM's .. this also means 2 of your VM's could share an internal IP so you can utilize CLUSTERIP and other tools like it.

Thanks nullmind, you just made my week FYI to anyone trooling for answers, this would allow you to support heartbeat as well.

Will this be available in the 6 - 12 month time frame mentioned by devlib, or is it coming sooner.

Also once private ips are available is it possible to assign all public ips between one or two machines and use only private ips for my remaining vpses?

Pardon me if I seem like I'm expecting the world

Dont worry, we stride to provide the world and the moon as a bonus to our customers

And here I thought just the reality of a vps cloud was my xmas present.

Thanks,
Greg

[nullmind]

Thanks nullmind, you just made my week FYI to anyone trooling for answers, this would allow you to support heartbeat as well.

My pleasure

Will this be available in the 6 - 12 month time frame mentioned by devlib, or is it coming sooner.

Yes, the private ip's will be within 2-3 weeks (more complex setups such as automatic setup of CLUSTERIP, thats what will be a while)

Also once private ips are available is it possible to assign all public ips between one or two machines and use only private ips for my remaining vpses?

Unfortunately no, the primary public IP assigned to the VM will not work on other vm's (but you will be able to get additional public ip's)

[poetics5]

Yes, the private ip's will be within 2-3 weeks (more complex setups such as automatic setup of CLUSTERIP, thats what will be a while)

As long as I'm able to use either a private ip or public ip on any vps within my account I can setup CLUSTERIP and heartbeat.

you will be able to request a range if internal IP's for your account (not VM specific) and we will unlock those to be usable on any of your VM's

Is this the down the road features your refering to?

Unfortunately no, the primary public IP assigned to the VM will not work on other vm's (but you will be able to get additional public ip's)

To clarify: I can request additional public ips, that i could then move back and forth across machines?