Advantage of having multiple linodes

Well, yes, there are both pros and cons. Off the top of my head:

Pros:

• Redundancy (if you build it right)

• Fault tolerance (if you build it right)

Cons:

• You have to build it right

• Administer two machines instead of one

• Many apps perform better with more RAM… fewer take advantage of multiple instances

So, it's a trade off. Building soft-fail apps that don't also have cascading failures is hard. I saw the guy who ran CNN.com give a talk on what happened to them on 9/11… and having load balancing actually caused them to take longer to recover, because the load was so high that, once one server failed, the increased load on others caused them to fail pretty readily, as well. That's an extreme situation (and, arguably, a not-conservative-enough disaster response was employed), but it does take more work to administer two machines than one. And that's basically what you're going to be up against. If you get two, let us know how it goes!

Hi,

We plan on using multiple linodes to manage our set-up.

The main reason is that we want to seperate out things like apache and mysql. This way both can run at the same time.

Also the seperating out of apps can make life easier, but can also make things harder.

Adam

adamg:

Are you giving apache and mysql as an example for myself being as I'm limited in RAM, or why would you have problems running apache and mysql together?

There where just an example.

Although if you have an intensive web application that needs mysql, you could hit RAM problems.

It is also nice to be able to run those two at the same time.

In the end it depends on what you want to run.

Adam

@bpendleton:

I saw the guy who ran CNN.com give a talk on what happened to them on 9/11… and having load balancing actually caused them to take longer to recover, because the load was so high that, once one server failed, the increased load on others caused them to fail pretty readily, as well.

I am one of the guys who was in the trenches at CNN that day. We had millions of hits per minute then, so it's hard to compare to a two node setup. We average over a 100k hits per minute most days. (yes that's hits per minute!).

Generally, unless you have a seperate load balancer or do some DNS voodoo (similar to akamai). A single larger node would probably be better.

@bakins:

I am one of the guys who was in the trenches at CNN that day. We had millions of hits per minute then, so it's hard to compare to a two node setup. We average over a 100k hits per minute most days. (yes that's hits per minute!).

Generally, unless you have a seperate load balancer or do some DNS voodoo (similar to akamai). A single larger node would probably be better.

If memory serves me properly, I believe you guys had ended your account with Akamai days before 9/11 and resigned that very day, heh

Bill Clinton

@bpendleton:

I saw the guy who ran CNN.com give a talk on what happened to them on 9/11…

Looks like a copy of this presentation (in note-form anyway) is here http://www.tcsa.org/lisa2001/cnn.txt

@Bill Clinton:

If memory serves me properly, I believe you guys had ended your account with Akamai days before 9/11 and resigned that very day, heh

Yep…

Of course our infrastructure has changed dramatically since then. Hopefully one day, I can saher details…

@bpendleton:

So, it's a trade off. Building soft-fail apps that don't also have cascading failures is hard. I saw the guy who ran CNN.com give a talk on what happened to them on 9/11… and having load balancing actually caused them to take longer to recover, because the load was so high that, once one server failed, the increased load on others caused them to fail pretty readily, as well. That's an extreme situation (and, arguably, a not-conservative-enough disaster response was employed), but it does take more work to administer two machines than one. And that's basically what you're going to be up against. If you get two, let us know how it goes!

Ahh, the war stories that I can't tell because they'd fall on the wrong side of corporate confidentiality. :/

Suffice it to say, try crashing all of your servers with a cascade crash. And then realizing that bringing up said network is not as easy as it sounds. :/