The Importance of Good Hosting
For several years I was with the same website hosting provider. The hosting was okay but I often felt the loading times of this and some of my other websites were slow.
I know there are many things that can influence this including the nature of the content, for example the size and number of image files, flash content, affiliate adverts, the hosting server’s capabilities, the impact of other websites on the shared server performance, the local ISP broadband connection speed and general web traffic volumes.
I was unhappy with a number of aspects of the old UK based hosting service provider including costs and the customer payment/administration of the reseller features. These, frankly, were not fit for purpose. Be careful what you sign up to.
Anyway, in August this year I opened an account with pingdom.com that “pings” specified websites at regular intervals and logs the response time. This doesn’t directly indicate the speed of a website to open or download but does represent the speed of response of the server it is hosted on. In turn a faster server should result in faster website loading.
In October I signed up for a new hosting service provider based in the USA and loaded my own and customer websites on to the new servers. I switched over this site at approx. the start of November and it operated there for a couple of weeks. I was interested to note an immediate and noticeable drop in the average response times logged. This is good. My site should load faster.
Was this due to the new hosting or had something changed on my website? Nothing else had changed but the hosting.
Unfortunately I encountered insurmountable technical difficulties in transferring over several customer accounts. The old hosting and new hosting denied any responsibility, so I was “piggy in the middle” with technical issues and customer websites to support. Throughout no less than 6 technical support enquiries no one in either of the hosting companies technical support teams could suggest any route to a solution.
It had also been pointed out to me in the interim that although the US based hosting provider had a certified “Safe Harbour” agreement in place, I could be in violation of EU legislation relating to Data Protection by hosting outside the EU zone.
I decided to to pull the plug on the new hosting and get a refund within the 45 day guarantee period.
So I switched everything back to the original hosting as it was all still operational and intact while I found another new hosting service provider. (I switched the nameservers).
It was interesting to note that the ping response time on this site went up to the previous levels throughout this period back at the original hosting.
Once I found a hosting service provider inside the EU that could supply my needs at an affordable price, I set up an account.
It is interesting to note that having switched the hosting of this website to the latest hosting service provider the server response times logged are at the lowest ever.
I could be mistaken as to the reason but since then I have noticed an immediate improvement in the number of site visitors in the statistics. Could this indicate improved rankings with search engines. I have subsequently made a number of other changes to the site layout and features so it may not be entirely due to the hosting.
However the server ping average response time remains at the lowest ever and this can surely only be a good thing.
The chart represents the data since “records began” in August. The values on this chart represent the daily averages gathered from a minimum of 24, hourly pings. So the peaks represent sustained poor performance. The spike-trough includes a period of downtime which pulls the average down and is not representative.
Section 1 shows the original hosting with an initial “plateau” of around 3 seconds, with some higher peaks. It then enters a period of greater instability where the average “plateau” increases, the frequency and fatness of the peaks increases with a maximum spike at over 5 seconds. Given that the average person browsing the web will not wait much longer than 3 or 4 seconds for a site to load, that was not a good section of results. It does settle out again but at a higher plateau than initially and all over the 3 second level.
There was a dip when the site went down for a period during the migration to new hosting.
Section 2 shows the response times for the US based hosting. Most of this time period has response times of between 2 to 2.5 seconds.
Section 3 shows the return to the original hosting at around the 3 second level.
Section 4 shows the latest hosting service provider with average response times, initially just over 2 seconds and then dipping below the 2 second level. The last 7 days have seen times logged at a minimum of 1.47 seconds with a general average of 1.89 seconds.
The worst peaks of 4.5 and 5 seconds are recorded during the original service provider’s jurisdiction.
The best dips in the chart are recorded during the latest service provider.
It should be said that within the data all the hosting service providers have the occasional “flyer” data points, e.g. of 9 seconds. I think the worst one was 18 seconds but I consider these to be random and not representative of the overall service.
It is unclear why there are distinct “plateau levels” in the data within periods serviced by a single host. Is this due to changes on the website, the hosting servers, the internet or in some aspect of the sampling method?
There are some spikes in the data on occasions that can be identified as having large volumes of data being transferred via ftp to or from the website. But the relatively short duration of these file transfers doesn’t explain the prolonged results and some of the best average results have also been obtained during high ftp traffic times.
It will be no surprise that I will be continuing to monitor the progress of the latest service provider but I have great expectations.
Update 16/12/11
Just updating this article with a graphic of the last 7 days.
My further contemplation of the differing “plateau levels” is that they are affected by changing the sample rate. By increasing the sample rate it is effectively making the data recording process more sensitive to variations. In a process where standard deviations are normal, this should not necessarily affect the “mean” average. In this case the nature of what is being recorded is not standard as all the deviations tend to be in one direction (the bell curve is skewed) so will make the underlying average response times over the duration larger.
The last week of data shows an average daily average of 1.775 seconds with a maximum of 1.869 and a minimum of 1.699. That’s almost twice a fast as my original hosting.
