Web Hosting & The Need for Speed
We are determined to ensure Hilenium customers use the best hosting technologies at the best prices for their businesses. Consequently, we are building test tools and blogging about the result. Hilenium Speedshop is our blog channel dedicated to lifting the veil on website performance.
For Hilenium Speedshop we plan to build testing tools and then document our research outcomes. We will use these tools to confirm our ideas on website optimisations. Performance theories are great, but they are just that, theories. If it isn’t tested we won’t endorse it. Hilenium is dedicated to a server-less business model that allows us to test new server and cloud vendors, and use the best for our clients. One of the low-hanging fruit of website speed optimisation is on-site asset optimisation and Content Delivery Networks. For this reason, our first tool is in Beta – The Image Load Tester.
Image Load Tester
The Hilenium Image Load Tester’s job is to test all the various configurations of image loading. It allows us to vary parameters including image size, quantity, SSL & CDN.
It wouldn’t be fair of us to ‘rabbit-on’ in our first Speedshop blog about testing and metrics without presenting a sneak peek into what we are doing.
As a quick example test, we ran a 200 image per page test. Images were 1920 x 1280 pixels with JPEG compression set at 70. The images were loaded over SSL. To keep things real, the assets also loaded from one of our web servers, and not the image location e.g. CDN:
- 4 x CSS Files – 24.7 KB in total
- 5 x Font Files – 57.2 KB in total
The base HTML document is 35.8KB and it’s served gzipped. The JS/CSS/Font asset files are neither minified nor gzipped.
As you can see from The Load Tester Results we evaluated several image loading sources:
- Server – The test web server’s image directory
- AWS CloudFront CDN (Amazon)
- Azure Edge CDN (Microsoft)
- Shotgun! We will leave this up to your imagination. It will feature in a future blog.
The test ‘no shows’ include:
- Google Cloud CDN – What is Google thinking? – they win unfriendliest CDN configuration award – we had a deadline.
- Alibaba Cloud CDN – A whole new level in verification in gaining a corporate account – two days later still not an active account capable of deploying a CDN.
- Fastly CDN – Looked good but we didn’t have the time to get our head around configuring their system.
- MaxCDN – They only have a 15-day trial then a monthly bill. We didn’t feel like signing up for something to evaluate that we may not use again but by accident keep getting billed.
We ran 13 tests across each image source and you can see the results below:
The most interesting result in our very limited test (see many caveats below) was the web server was the most consistent (i.e. least variation of the group). It was also the fastest. However, we don’t believe it would remain the fastest in a high-traffic event ie. many concurrent site visitors.
KeyCDN had some great response times and also some really slow ones too. If had not been for the last two or three tests it would be in the ‘ball-park’ of the hyperscale clouds. KeyCDN was the fastest to configure but does not include SSL cert generation for a vanity CDN domain.
Azure and AWS performed similarly, though my perception is that Azure was more consistent. Both AWS & Azure offer SSL certificate generation for vanity CDN domain names. Azure was the easiest from a UX point of view but took quite a few hours to be ready for use.
Shotgun! is a secret so you just have to use it a benchmark. I will give you a clue – it sits right in the middle of all the CDNs.
So who is the winner? No one. This test is not definitive enough to make a judgment – it’s a Sneak Peak! Stay tuned for our Hilenium Speedshop blogs for more in-depth analysis.
All the caveats on why not to read too much into this data:
- We ran the test from one location – the Hilenium office which, whilst having a dedicated, stable internet connection is not terribly fast – Bring on the NBN!
- The test only ran 13 times – a very low number.
- There were no compensation factors for other traffic that may have been on our network at the time.
- The test site was on a shared web server and ran sequentially – no compensation for its load factors over time.
- We ran the test from a single computer so the web server & CDN are only responding to one page-load at a time.
Hilenium Speedshop has big plans for Load Tester. We want to shed some light on the intricacies of asset loading with a particular focus on imagery. At the moment we are going to investigate in depth the following:
- Image Size
- Quantity of Imagery
- Image compression
- Whether to use a CDN
- Value propositions of different CDNs
- SSL and performance
- Using Vanity domain names with CDNs
Of course, if all of this seems a bit too much information there is an easy way to benefit from all our web hosting research, simply get business hosting with us.