Last Friday Kanye’s new album leaked, 5 days ahead of its scheduled release date.  As Twitter exploded in a frenzy of activity, reports came that a number of websites – including forums, blogs, and other sites that typically post hip-hop music – were inaccessible.  We had a hunch that there might be a lot of traffic going to KanyeToThe.com, the most popular dedicated Kanye forum on the web.  We thought it’d be fun to check out the performance of the site as it weathered the storm of traffic brought on by a major event in its sphere.

We set up a monitor for the Kanyetothe.com/forum.  We set a real browser (Chrome) to request the page every few minutes from locations all around the world.  From our offices in Boston, we were only able to access the site roughly half the time, so we predicted that other locations would show similar issues.

Boy, we were right.

All Falls Down

For some context, here’s what a trending graph looks like for an average site across a number of global locations.

There are a few issues, but most of the performance variation falls within a fairly tight range from 5-10 seconds.

Now let’s look at KTT over the past week.

Ah! Avert your eyes! Yes, it’s a dense forest of poor performance. And yes, many of those spikes reach up over 50 seconds.

Along with the sky-high spikes in load time, average performance was bad — around 12 seconds Time to Interact — and the site experienced uptime of only 91 percent. (That percentage could be considered generous: all of the 30, 40, 50+ second samples counted as “”Available”” because they never timed out, but common sense tells us that nobody waits that long for a page to load. So in those cases it was effectively “”down.””)

Some of these samples we can see displayed nothing until almost the end of their load process:

Others actually displayed content fairly quickly, but the load process was dragged out by a few particular assets that likely prevented the user from interacting with the page, even though they could see most of the content. That’s super annoying, FYI.

A few more points we found interesting: 

1.) The performance issues were sustained throughout the week, not just on the days on which the album leaked and the album was officially released. We don’t have any baseline data on the site, but we have to assume that performance is not this awful on the regular, since it’s a long-standing and successful site.

2.) There was a conspicuous absence of bad performance issues on the actual release date, before they resumed and continued throughout the week. What happened? Was everybody who reads KTT in a Yeezus-induced blackout? Or did KTT provision extra infrastructure to handle the expected load on that date, only to be upstaged by the leak?  Who knows.

3.) Bad performance was spread across the globe evenly — see below for performance averages by continent for Europe, North America, and Asia.  This is interesting to us because many small to medium-sized websites we see show much worse performance in regions far away from where they’re hosted (KTT is hosted in California).

From looking at the response headers of certain assets on the page, we know that KTT uses a content delivery network (CDN) to deliver static content.  That would explain why performance was uniform across the world: one of the main benefits of a CDN is that page content is stored in nodes across a global network, so users everywhere receive “”local”” content.

But CDNs are supposed also supposed to prevent bad performance, not just spread it evenly around the world. Especially when a site is hammered by traffic, the CDN takes on a significant portion of the load and helps the site maintain baseline performance.

Looking into a few samples, we found that the worst performing assets were actually ones served from the CDN. That means something was very wrong. Usually static content served from a CDN performs far better than content served by the origin server or by a third-party provider. We can only assume, then, that the CDN was incorrectly configured for the site, or that the CDN experienced major problems that coincided with the time frame that we monitored KTT.

Blame Game?

This case proves that no matter what kind of site you operate, being prepared for heavy traffic should be a priority.  The nature of the web dictates that an immense traffic load can come at any time.  An article can go viral, bringing down a media site. A marketing plan can garner more success than expected. A product can feature on the Today Show.  A major artist’s album can leak. No matter what the subject and business goals of your site, prepare for the worst, so when the traffic comes you can capitalize on it, rather than get burned.