Update: Since I wrote this post, I’ve read many stories similar to mine of troubles with CloudFlare. They seem to be going for the bottom end of the market, and their offering leaves a lot to be desired if you need a serious CDN that doesn’t cause weird latency, annoying landing pages and other issues. I wouldn’t bother with this company again.
From the comments left on this post below I get the impression that nothing much has changed since I wrote this.
A commenter on Hacker News had this to say about my post:
I’m not sure why you’re surprised.
CloudFlare is amateur at best, compared to existing DSA/CDN offerings from Akamai, Cotendo, EdgeCast etc.
But those cost a lot of money (at least 50-100x the cost of CloudFlare) if you’re small/medium sized. You get what you pay for with $20/month.
I can only agree with his point. You get what you pay for and for $20 a month you’re not going to get a lot.
CloudFlare is a website performance and security product that launched publicly at TechCrunch Disrupt in September 2010. It’s a reverse proxy, firewall, and global content delivery network that requires very minimal setup. Their main product is free, but they also offer a pro version that provides a few extra optimisations and features available for $20 a month.
Other than email address obfuscation, I was not interested in the security features CloudFlare offered, one of the reasons being that if CloudFlare detects what it thinks is a suspicious user it will prevent them from accessing your site, display a CloudFlare branded page and make them enter a captcha to continue. I hate the idea of this – a false positive that triggers this will make your site look horribly broken and I also don’t want the user to ever see CloudFlare branding or anything on my site that I didn’t design myself.
I think this is one of the things CloudFlare gets very wrong about their whole offering, they should be aiming to be completely transparent no matter what happens. I don’t ever want to see their logo on my site or hear about how many ‘threats’ they prevented. That stuff needs to happen right out of my customers sight.
If your web server goes down and they are able to serve up a static version, at the top of the page they inject a ‘This page is offline, but here’s our cached version’ thing, along with the CloudFlare logo. I might be less annoyed about this if I hadn’t already shelled out $20/mo for the pro version of CloudFlare but they are essentially using the fact that my server is down as an opportunity to do some free marketing for themselves, despite the fact that I already pay them money. I don’t like this either.
Anyway, I switched off all of the security features with the goal of using CloudFlare purely to improve the performance of my site. Some of the features CloudFlare offer are still in beta. To begin with I left all beta features disabled except for JS/HTML/CSS minification.
The site I was trying to improve the performance of – http://frenzyapp.com is pretty simple and made up of just a few images and static html pages.
Things seemed to work quite well for a while, but sometimes I noticed my site was taking longer than usual (>3seconds) to load now that CloudFlare was in the picture.
I started doing some tests using the site performance testing tool at http://www.webpagetest.org which allowed me to test the page load times from multiple locations around the globe. WebPagetest showed that a single file – jquery.fancybox-1.3.4.css was taking over 3 seconds to load from certain locations and also showed some other issues with requests being cancelled. This was causing the whole site not to be fully loaded for over 6 seconds. Not good at all.
I ran a lot more tests and found I couldn’t get the problem to occur consistently from all locations but it would always happen when loaded from New York. I opened a support ticket with CloudFlare describing the problem and attached screenshots of a number of tests. They responded quickly and told me to try disabling the JS/HTML/CSS minification feature (which was after all, in beta) and see if that improved things. I did this and re-ran the tests 10 times from New York, again using WebPagetest to see if the problem persisted, unfortunately it did. Loading this file was taking over 3 seconds to load bringing the overall page load time to around 6 seconds. Still no good.
I tried disabling some of the other CloudFlare features such as the website preloader and ran more tests. Still the problem continued. I updated the support ticket with my findings and then waited patiently for 3 days. This time CloudFlare did not respond. I posted to their support system again to ask if a resolution was coming and tweeted them:
Almost immediately, I got a tweet back from Matthew Prince, the CEO and co-founder of CloudFlare. He suggested that I try completely deactivating CloudFlare and re-testing. We had a rather lengthy, friendly exchange which ended with this tweet from him:
This is encouraging. I am hopeful that the problem will now be resolved or that I’ll at least get an update soon now that I have a promise from the CEO.
I also took his advice and tried fully deactivating CloudFlare from the control panel, this makes the DNS A record to point directly at my server, taking CloudFlare out of the loop. I then re-ran WebPagetest 10 times from New York and found the problem had disappeared. The page is now fully loading in an average of 3 seconds or less compared with the 6 seconds taken with CloudFlare enabled. This provides further evidence that CloudFlare is causing the issue and not my server configuration. I update the support ticket with my findings and wait patiently for another 4 days.
I don’t hear a thing from CloudFlare, no more tweets, no update to the support ticket, nothing. I guess they must have changed their mind about resolving my issue ‘in a few hours’
I decide to re-test everything anyway, and also run tests from other locations to get an overall indication of the performance of my site with CloudFlare enabled versus CloudFlare disabled.
My findings are displayed below:
The times given are seconds to fully loaded. In New York – the problem location where the CSS file is taking greater than 3 seconds the total load time is, on average, 2.7 seconds greater with CloudFlare activated.
The site actually loads faster when direct from my server in every tested location except for two.
On average, across all tested locations, the site is loading 0.8 seconds faster with CloudFlare deactivated. This certainly isn’t what I would have expected from a product that is supposed to ‘supercharge’ your website.
One factor to consider is that these tests were done while my site was under almost zero load. If my server became overwhelmed then it would slow right down in which case CloudFlare may actually be faster. It also remains to be seen whether other sites using CloudFlare exhibit similar load times or if my site is particularly troublesome for some reason. Much more data would have to be collected to conclusively prove anything. I have uploaded a zip archive that contains screenshots of all the WebPagetest results with more information on each test in case you are interested or want to see the original source for those numbers.
Personally, I’ve decided to move on from CloudFlare. Despite what seems to be a promising service, right now they are not providing me with a faster website. They also don’t seem to be fixing my problems or keeping me informed. It’s a shame because I really think they are onto something, there’s so much that goes into keeping a website speedy and available these days. If I can change two name servers and have everything taken care of for me, it leaves me more time to put into building great products.
I do think CloudFlare will iron out these issues eventually. At the moment the company still seems to be in early/beta stages. They may be worth another look at some point in the future. They also need to do some work on improving their customer service and make sure they follow up with customers who have problems.