How can one achieve the cheapest yet very reliable web application configuration?

Let's assume at least 2 application servers for $80/mo plus a DB server will support most people's applications for a while - and we just want to achieve good reliability (at least 3 nines).

One can rent a pretty good VPS solution for around $80/mo right now from any reasonable provider (Amazon EC2, Slicehost, etc.). However, none of these VPS hosting solutions are perfect, and I've experienced more than 99.9% downtime with each of them.

I'm not sure how best to configure these 3 machines - the best would be to put the two app servers on different providers (best with different network connections) and use HAProxy to keep an eye on each other. If one fails, it will update the DNS to remove it from the pool of IP addresses for your application. But what to do about the DB server? That's still a single point of failure.

I have had issues with DNS in the past, but this can be handled by an external dedicated provider like DNS Made Easy for very cheap ($15/year). This supports dynamically modifying DNS entries if necessary as well, which is handy if you're not planning on being able to update them manually during a crisis.

Backups should be done to an external source (S3 or FTP site) at least once per day - again a minimal each month. You need an automated deploy and restore script in order to get past 3 nines I think from your backup as well.

I don't quite feel like this is quite there due to the DB availability, but it'll cost you around $80x3 + disk space + DNS = $250/mo.

Can one do better?

  • "I've experienced more than 99.9% downtime" I hope it's only more than 0.01 downtime. ;-) – splattne Jul 1 '09 at 20:04
  • Three 9s (99.9%) is genereally what you get when a non-idiot runs a computer network. Basically: Without doing anything - a couple of routers, stock hardware ordered from a reputable vendor, that is what you have. Having less means incompetence. Baiscally you call for 0.365 DAYS (roughly 8 hours) of downtime per year. That is enough time to reload backups for small computers 3-4 times a year. – TomTom Aug 8 '13 at 11:18
  • If you get VPS let me tell yu that my inhouse VPS cloud has not had that ever since we started. I suggest just going with a decent host - because yours seem to be really bad. – TomTom Aug 8 '13 at 11:19

Cheap, Minimal and Robust? You've just hit the evil triangle of "Good / Fast / Cheap" - Pick 2, you can't have all 3. :)

Realistically you're doing pretty well if you can do a geographically-redundant application with failover for $250/mo.

The only alternate I can think of would be the VPS solutions you've already had or perhaps Microsoft Azure might fit the bill.

  • I just want good > 99.9% and as cheap as possible - that isn't too much to ask is it? :) – Chris Harris Jul 1 '09 at 17:04
  • I think the moment you said you want to mitigate a single-host failure that you went over $250/mo :) – Brandon Jul 1 '09 at 17:31
  • The name I have heard for this is the Chinese Laundry principle: Good/Fast/Cheap, choose any two. – eleven81 Jul 1 '09 at 17:48
  • +1 I love the evil triangle ! – Antoine Benkemoun Sep 25 '09 at 12:34

I agree that you are doing pretty well where you are.

Make sure that you have solid recovery processes, so that any failures that you do experience can be resolved quickly.

"Three Nines" availability equates to roughly 8 hours of downtime in a year. If you don't count maintenance windows, that isn't a particularly demanding standard for a mainstream hosting company. I am skeptical that you have seen worse on two well regarded hosting services.

Remember, as far as the host is concerned, once they give you a working server, you are back up. If you have a failure, and 90 minutes later they have given you another server, your downtime is 90 minutes. If it takes you three days to get it working, that additional three days is not their problem.

The database server is clearly a single point of failure .. you have to determine your tolerance for that risk and the cost for mitigating it. I don't know much about making a DB server redundant other than that it is really hard for small systems. If that worries you, your best bet is to contract for hosting of the database itself rather than using a VPS and managing your own database server. One example (from Google, not a recommendation) can be found here; it isn't particularly expensive.


http://www.linode.com/index.cfm has been very reliable for a few of my friends. They have multiple locations and you can scale pretty easily.

  • I don't think picking a "reliable" host is going to fix the problem - the solution should be robust to a single host failure, don't you think? – Chris Harris Jul 1 '09 at 17:02
  • They have multiple locations to host multiple VPS at separate facilities for 20 bucks a month it's a pretty cheap solution which meets most of his requirements. – egorgry Jul 1 '09 at 17:10

I think you are facing 2-3 Problems here:

Uptime definition

Three nines including announced maintenance windows or excluding? How far in advance will you announce the maintenance windows?

Hardware Uptime

That will only scale with redundancy. You never know when it will break and three nines means only about 9h downtime per year. So even a 4h contract will grant you only one downtime (2 if it is indeed fixed within 4hours - but that's not exactly cheap to get). IMHO you have to have 2 Servers for that

Software/Service Uptime

Are you going to define uptime by ping reachability, HTTP answers, how? - You can easily solve that by defining in advance what you consider "up" (or if that is easier what you consider "down" - but that would be badness enumeration)

As far as cheap goes, I don't know if by cheap you mean initial costs or running costs. You can get some cheap server nearly everywhere, but you probably need more of them to rule out hardware breakage, or get some decent hardware with HP, Dell, IBM with higher initial cost but a much lower probability of hardware breakage and thus will have a lower running cost. Just calculate over lifetime which one is going to be cheaper.

If you just want some URLs:

server4you was quite nice for me EC2 with georedundant locations (EU and US) could also be very interesting

EDIT: To be clear when I say 2 Machines I mean redundant Machines per Use case (redundant DB, redundant Web, redundant DNS, redundant $whatever

  • Yes, total uptime should exceed 99.9%, not "excluding maintenance." I'm defining uptime as full http and DB accessibility. – Chris Harris Jul 1 '09 at 17:12

Depending on your exact traffic and needs and requirements, I've hosted many smaller web apps over on 1&1 without much trouble at all.

Other than that... pretty much any popular hosting provider out there has decent (read: published) uptimes.

99.9% uptime means you're accepting a downtime of 8 hours, 45 minutes, 35 seconds within a one year period.

Most decent hosting companies hit that fairly easily.

  • I don't think that's true -- I know I have been hit several time by EC2 outages and network card failures or HD failures at other places as well. – Chris Harris Jul 1 '09 at 16:56
  • We use amazon and they make no guarantee of uptime that I'm aware of. – egorgry Jul 1 '09 at 17:13
  • Maybe I'm just lucky then? :-) I've hosted stuff on 1&1, dreamhost, and various other places and rarely get hit with down time. I have not done any serious EC2 hosting so I can't speak to that. In most cases, hosted facilities tend to get things up and running faster than I can on my own hosted stuff... so its a toss up I guess. Nobody likes being up at 3am... – KPWINC Jul 2 '09 at 17:02

I'd recomend discount ASP.net. I like the fact that you can get a free sandbox server, and I can't remember any downtime I've seen.

  • I don't think picking a "reliable" host is going to fix the problem - the solution should be robust to a single host failure, don't you think? – Chris Harris Jul 1 '09 at 17:03
  • How else are you going to pick, a Ouija board? You asked about hosting, so you start with the host. If it is unreliable, game over. If it is reliable, you still need to make your own setup reliable at the host. – tomjedrz Jul 1 '09 at 17:26
  • Since providers don't (and can't AFAIK) use a single host all are reliable as long as your application doesn't crash on all hosts in the pool. It's not your job to congifure their hosting pool (nor can you). If you are looking for hosting space providers that's (to me ) a different question. – Jim B Jul 1 '09 at 18:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.