Posts Tagged servers
The Results Are In…
And as it turns out, Linode is far, far faster as well as being more consistent for the hosting of this site.
Take a look at the graphs, the first was captured from this site in the last couple of weeks for a 24 hour period, hosted on dreamhost:

Yeah you’re reading that right – the label to the left is 10,000ms (10 seconds!), with the average response being around 3000ms. This was a fairly indicative day, there have been better and worse, but on average the response time under “good conditions” would be around 1000-3000ms. However the good conditions would only be < 90% of the time, the other 10% is dreadful.
Here’s the linode graph for the last 24 hours -

Quite a difference huh? The average response time has decreased to around 0.5 seconds with the slowest still being better than the best time dreamhost ever managed!
Of course, dreamhost and linode are appealing to different markets – rolling your own server configuration and maintenance requires a certain degree of Linux geek-ery which the average person is not about to tackle. However, clearly the quality of service provided by the mass-hosting providers (and dreamhost is one of the better ones), is not what one might expect.
Part of the problem is that the competition for hosting dollars has resulted in the proliferation of the “all-you-can-eat” hosting plan. Practically every aspect of these shared hosting plans is “unlimited”, and offered at sub-$10 per month rates. How can that be possible? The old adage of “if it looks too good to be true, it probably is” definitely applies here… it’s possible only so long as none of those sites with “unlimited hosting” is very popular.
While the superficial limits of the hosting might be “unlimited”, you’re still sharing a server with god-knows how many other people. Each one of those people has probably set up hosting for several sites, and is doing the “all-you-can-eat” thing too. The whole thing only works so long as none of the sites on the shared hosting gets very popular. You’ll find sites on this sort of host that suddenly attract a large amount of traffic have a tendency of killing the server for everyone (and often the accounts end up temporarily suspended).
My prediction is that these sort of hosting plans will not last in the long term. With the rise of virtualisation and cloud services such as Google Apps or Amazon EC2 there will be a shift towards self-managed or abstracted hosting, depending on the level of capability of the hostee. It will take years for this shift to occur though, due to the incumbent popular apps, such as Wordpress which are tied to the current architecture.
Anyhow, I can recommend Linode as an inexpensive, high quality service which works exactly how you’d expect… but you should expect to get your hands dirty at the $ prompt
Move to Linode Complete!
The blog is now happily running on my linode instance!
I had previously been hosting the blog on Dreamhost, and had been pretty happy with the service. Recently though, I signed up for a free Pingdom account.
Pingdom is a service which monitors your site for uptime and response time from different locations around the world, and gives you a real-time graph of those statistics. It will also email or SMS you when your site goes down, and then when it comes up again.
It was from this I found out that the quality of Dreamhost’s hosting my site varied wildly over the course of a day. Under “good” conditions the hosting was very good, with the response of the blog’s main page coming back in around 1 second (1000ms). However, this would spike reasonably regularly (at least once a day on average) to between 3000-6000ms (3-6 seconds) to respond.
Short outtages of 5-15 minutes were also fairly regular, probably once every day or two. I guess this is mostly dependant on which server you are hosted on, and how many people are running sites or scripts which kill the server on a regular basis. In any case, the quality of service was a bit up and down (pun intended).
I’ve had a linode account for a few months, mainly for Java hosting and other bits and pieces, so I decided to move the blog over here (LAMP) instead. The performance doesn’t seem to be much better (though requests are occasionally 800ms or so now), but the uptime should be better, and the max response time should be much, much better.
I’ll post some graphs of the results / differences once I’ve collected a good amount of data.

