High Availability Question

[alexr186]

I know 3tera has a high availability option where you can mirror the apps (virtual servers) between grids? I was wondering how that works? Does it use a lot of bandwidth? How does the sync goes when one goes down and one is up? How would you setup the dns?

I was looking into setup a virtual server that is mirrored on both grids, so if one grid go does, the other one fills in.

[nobu]

It's the volume(s) of your app and not the app which is mirrored (in the grid).

You can copy the app, and export to another grid, too.
However, it's up to you to keep the data in sync, and switch between apps
via DNS change or other means.


As far as disk mirroring is concerned, it's done on the private network and doesn't use any bandwidth you may have to pay for.

If you construct multi grid app mirroring, it depends on the setup, type of app, etc..

For DNS, you can use a service like DnsMadeEasy to monitor your site and switch automatically.

[alexr186]

So to make sure I have this right, so you have a grid with an application that has one virtual os running on it. It is normally mirrored on two physical machines at all times, not you could not expand that mirroring to be 4 physical machines, two on one grid and two on another grid, with them all keep up with each other in real time?

Thanks

[PeterNic]

Alex,

I think you had the impression that AppLogic will live-replicate appliance RAM state on multiple physical servers. We are not doing this; it would not work -- Tandem used to have some very expensive machines that were doing this in hardware (now HP Non-Stop line)... doing it in software on commodity servers, with or without hardware virtualization, is currently not feasible due to network speed and latency. There is difference between fault-tolerance (http://en.wikipedia.org/wiki/Fault-tolerance , what Tandem provided) and high availability (http://en.wikipedia.org/wiki/High_availability, what AppLogic provides).

Grids provide universal high availability -- i.e., high availability that is provided to the application without having to code specifically for it. There are two key aspects (at least from application's point of view):
- all storage volumes are mirrored on at least 2 servers. If a server holding one of the mirrors dies, the data is still read/written to the other mirror. You can rebuild the mirror (before or after the downed server comes back online)
- appliances (in GSC's case, there is one appliance for each GSC application) will be restarted on a different server if the server on which they were running dies. There is a short downtime, though, and loss of transient state - your appliance sees this as unscheduled reboot.

The goal of universal high availability is not to provide application that will never experience downtime (although using grids and some existing DR methods you can build a fully redundant application across several grids); instead, it makes it tolerable to lose a physical server and (a) not lose data and (b) automatically restore your application to operation within a few minutes (instead of the hours waiting for a support person to build a new server and move your hard drives to it or restore data from backup -- all the while your customers are calling you and asking where their service is). Human intervention is still needed in some rare cases. Our goal is to let you sleep through at least 4 of 5 server failures and only learn that your service recovered automatically in the morning (I don't know why most failures happen in the middle of the night, but that has been our observation).

I hope this clarification helps.

Best regards,

-- Peter

[alexr186]

Thanks for the reply, reading some of the information on the different mediums with applogic, it was hard to tell, so it is great to clear it up.


Thanks for the great replies.

Alex