I have a site that is getting about 30k hits a day, nothing too special. After cleaning up some mysql calls I think I am having it run a little smoother during high traffic times, but during the course of trying to clean up the site's processes the host suggested I look into load balancing. I'm already paying $1k a month for a server (8GB RAM, unlimited bandwidth). I'm wondering "when" is the time someone with a website goes into the load balancing option for a site. Is there some rules or rubric to go by which tells a Sys Admin that it's that time to do so?

Firstly, what your delivering must be important IMHO :)

Secondly.. if your servers CPU capacity is above 40% its time to scale out a little bit..

Consider 2 webservers and a dedicated SQL backend as an example..

The rule of thumb is..

if you have 2 servers running at 40% capacity and one fails.. You will have one server running at 80%

You can see how bad this can go when you are running at 50%+

These same rules apply to many features of applications and hardware.

My best suggestion is, ask yourself is it worth the money?? And why are you spending so much??? :P

  • when you say 'capacity', what does that entail exactly? – HollerTrain Nov 8 '10 at 15:58
  • Usually im refering to CPU usage and Network throughput.. Depending on what your delivering, its often hard to maximise your network throughput.. I would use the CPU as the main measure... Hope that helps – Arenstar Nov 8 '10 at 15:59
  • can you figure that out via "top" in ssh? – HollerTrain Nov 8 '10 at 16:02
  • Don't forget the IO budget. It's often harder to scale than the CPU or Bandwidth, without moving to multiple systems. – pehrs Nov 8 '10 at 16:16
  • @HollerTrain.. Top only shows you a point in time.. i like using rrd's to graph the data, however you could watch top under peak loads to get a feeling of consistent server usage @Pehrs.. Your totally correct.. Its not really an indepth point i made, but by moving the sql backend away from servers you can focus on trouble areas by modulation of "services on servers" – Arenstar Nov 8 '10 at 16:22

Generally there are two triggers for load balancing. When the site grows beyond the capabilities of one server or when you want eh redundancy and reliability that may come from having two servers in case one may fail. Also, sometimes it can be cheaper to buy/rent two or more low end servers and load balance, instead of one large server.

30k hits a day is less than one request per second, so unless your traffic has a significant peak, or you need the redundancy, it is probably pretty early to be considering it.


Another way to determine when, is by thinking about the cost of an outage vs the cost of having a second server.

It is already costing you $1,000 a month to host. Can you afford £2,000 a month?

At what point will "downtime" cost you that extra $1,000 in lost revenue? How quickly can you reasonably fix a broken server. Would you customers really be adversly affected.

From a commercial point of view, you want to squeeze all you can out of the server until the pip's squeek otherwise your just throwing money away.

We all want 100% uptime, but it comes at a price.

I'm already paying $1k a month for a server (8GB RAM, unlimited bandwidth).

Run. Now.

That is VERY expensive - we talk of abut 700 EUR here per month. I can see such a server for about 150 USD from my local ISP. Unlimited bandwidth may add some to that, but 1000 is simply said... ...paying for a gold bar and getting lead.

Now he possibly thinks of milking you for more?

Start self-hosting on your own hardware.

Get a decent server (dual opteron, possibly 6-12 cores, up to easiyl 64gb RAM). Put it into a decent data center.


Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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