And why did it cause a huge internet drop-out recently?
This past week the buzz-word floating about internet related conversations has been the drop out of a huge chunk of the internet related to an outage from the CDN provider Fastly. A good number of websites went out world-wide, and high traffic sites experienced either total outage or parts of their networks unable to be reached. It felt like a digital apocalypse for many. For some of our clients there was glee as their competition were taken offline by this outage. In the end, it was only for an hour, and late in the evening New Zealand time, but it still caused panic.
So how did an outage at a company no-one in the general public has really heard of before, cause such a ruckus? Well to get to the bottom of that we need to get a better understanding of how the internet functions, and some of the tips and tricks that webmasters employ to get their content in front of their users as quickly as possible so as not to lose users.
When someone goes to a website on the internet there is a flurry of communication between their device and various internet services to then serve the web page. Here is a rough pictorial guide to what happens:
Once the user has told their web browser what website they are wanting to view, requests are fired to Domain Name Service (DNS) servers in order to translate the address entered into an address that computers understand (an Internet Protocol (IP) address). That information is then used to talk to the appropriate server (or load balancer if the website is big enough, which then directs traffic to an available web server) to return the web page you have requested. That page may have a number of images and fonts and scripts linked to that all need downloaded in order to display the website you have requested on the device you are requesting from.
That’s a bit of the background behind how the internet works for websites. But where do CDNs fit into this mix?
Ever called someone overseas and noticed the delay between what you say, and their response? This effect is called latency. It’s the delay between your initial request, and you getting a response. Even with a global network using fibre connections, which are as fast as the speed of light, if I request a website on my device here in New Zealand, and it is hosted in the UK, every request to the web server is going to take at least half a second just to get from my device to the server and back, and that does not factor for any processing time on the web server slowing things down as well. If a web page has 30+ media assets, which is very common now-a-days, the website will feel almost unusable. The further away a server is from its users, the slower it will be able to respond to user requests.
This is where CDNs come in. A global Content Delivery Network is a network of computers located around the world. These computers are set up as a cache for the websites you are visiting. Website owners tell their domain names to resolve to the servers of the CDN instead of the origin servers, and then the CDN is configured to know how to get teh requested content from an origin server where the content is hosted. So, the first time you visit the website, the CDN server which is geographically closes to you, fetches your content from the origin host. It also keeps a copy of the content that the origin server has served, so if anyone else needs that content, it can return it directly instead of needing to route the request to the other end of the globe. This has the end effect of the website appearing to be served from the location of the CDN’s server that is closest to you. So each request to the web server now takes 50ms instead of 500ms+ The more ‘edge’ locations the CDN has, the better the chances of them having a server as close to you as possible.
The other advantage of CDNs is that you now have a pool of servers serving your website traffic, so if one edge location drops into an error state, other servers can take up the slack, without the need for a huge amount of traffic back to the origin server, adding load.
CDNs also get around a bit of a flaw in the way that internet browsers load media assets from web servers. Most web browsers will load content in a ‘blocking’ way, meaning they only open up a maximum of 10 connections (typically its only 4-6 connections without tweaking) to a remote web server / domain simultaneously. This means you have to wait for one asset to complete download before you can fetch the next one. Using a CDN, all assets can be downloaded simultaneously in a ‘non-blocking’ fashion, so page load speeds are vastly improved here too.
Due to all of these advantages, it makes a lot of sense for websites being served to a global audience to use a CDN to make their websites quicker for their end users wherever they are in the world. And there are a number of providers that offer this service to website owners. Some you may have heard of, like Cloudflare, Akamai, and Amazon’s Cloudfront. Fastly is another provider in this space that has a huge number of servers scattered around the globe, and boasts very impressive latency figures worldwide, which is how it has become popular with a number of larger websites around the globe.
Knowing what we know about CDNs now, it becomes easier to understand how half the worlds websites dropped out. The official line from Fastly is that a configuration error caused ALL of their CDN servers to refuse to serve any website content. It took an hour to resolve. If this had have been one or two servers then the CDN would have healed itself nicely and no-one would be the wiser – sites may be a little slower for some locations, but generally it’d be fine. But if you push out a global configuration that wipes out the function of all your servers, there is no saving that until you push out a revised configuration that undoes the breaking change. The more clients you have, the more websites are effected. From this outage, its easy to see that Fastly have a large client base around the world, and no doubt they are now contemplating their options for reliable CDN providers.
If you need help getting your websites working at optimal speed in front of a global audience, using trusted CDN partners, get in touch with Webmad and we’ll help you plan and implement solutions for optimal performance.