More about GPUs (Part 1 in a Series)
(mt) Media Temple’s GPU innovation is proving to be an ideal system to eliminate the arbitrary “you used too much” problem of the legacy shared server model. We’ve had some very interesting results with our new system and thought it was time to share some cool factoids and also to provide some further insight into how our system works.INTERESTING FACTOIDS:• Launch of the GRID: 2.5 months ago• Number of domains hosted on the GRiD: 40,000+• Percentage of sites under their GPU allocation: 99.6%• Number of sites currently exceeding their GPU’s: 36• Highest GPU overage recorded: 71,070• Highest GPU bill: $7,107.00• Average overage of those exceeding GPUs: $54.00HOW ARE GPUS CALCULATED?This information is actually contained in our FAQ section but we thought it was worthy of extraction for this post.
Each (gs) Grid-Server service plan is given 1000 GPUs monthly, which is equivalent to continuously using 10% of a processor core for one month. This breaks down to 1.4 GPUs per hour. A little arithmetic shows that this is about 40 customers per server compared to the industry standard density of 500 to 3,000.GPU is a measurement that derives from CPU time required by every hit/request made to your (gs) Grid-Server service. This includes your web server activity, programs, scripts and all associated parent and child processes across all nodes within our system (100’s of processor cores). The GRID system keeps track of all the measurements taken from your server and adds them to a running total which is calculated every 24 hours.
HOW TO FIND THE SOURCE OF GPU CAUSING RESOURCES?Although the number of things that can cause GPU usage is complex, experience has shown that major GPUs drivers are usually caused from server-side scripts. By request, (mt) Media Temple is currently developing a GPU analysis tool that will help our users diagnose what specific resources are causing GPUs to be consumed. This tool will be available in the AccountCenter along with the current GPU reports control. In the meantime, there is a report within your Urchin stats which can be used in finding which specific scripts are driving up your GPU count. Go to “Pages & Files”, then go to “Page Query Terms” and then look at the top scripts in the list. More then likely the scripts appearing here are an excellent giveaway to see what is driving GPU usage. This information can serve as insight as to how you might consider reducing the amount you are using.
We hope this information is useful to you. In the future, we will post more in this series of “More about GPUs”. Our goal is to help users discover methods to gaining better insight into their websites so that customers are more informed and can make better business decisions with their online planning.
