Skip to main content Link Menu Expand (external link) Document Search Copy Copied

“What are Processing Units?”

Even a web application that’s tuned for optimum delivery performance has a limit for the number of concurrent requests it can handle running on freistilbox. This limit is determined by the total number of Processing Units available on your freistilbox plan. Since freistilbox is a distributed system with services running on many separate servers fine-tuned for their purpose, listing raw hardware specifications like conventional hosting providers doesn’t make any sense with freistilbox. Instead, we chose PU as the main performance metric distinguishing our managed hosting plans.

PU overload

Each incoming HTTP request is assigned to a free Processing Unit of one of our application boxes. This PU will respond with the requested content and will then be freed up for the next request. But if HTTP requests are coming in through the Content Cache faster than your freistilbox setup can handle them, this can eventually result in all the Processing Units of your freistilbox cluster being busy. When there is no free worker process left to handle new requests, freistilbox will deliver a “Request Delivery Failed” error.

This happens when a) requests are coming in too fast, or b) your application responds too slow, or c) both. Let’s take a look at each of these scenarios.

Too many concurrent requests

So what can you do if there are too many concurring requests to your application server? Assuming a perfectly optimised web application, your only option is to upgrade your freistilbox plan to one with more processing units. But since most web applications are not yet perfectly optimised, there often are still ways to reduce the number of requests your application needs to handle.

For example, you could stop promoting your website… Just kidding!

The best way to reduce the number of HTTP requests going through to your application boxes is to optimise your content for caching. Every request that can be answered with content from a cache before it reaches your boxes is one request less that they’ll have to handle. There are several stages at which cached content will enable a faster response and less load on your freistilbox cluster:

  1. Your website users’ web browsers will cache static assets like images, CSS or JavaScript files so they don’t have to fetch them again with each new page they visit.
  2. The freistilbox Content Cache can cache static assets as well as generated content such as the pages of your website.

A well-behaving cache will only store and deliver content as your web application allows it in its HTTP headers. So, in order to optimise the caching efficiency of your website, you’ll have to define appropriate caching rules for each content object of your website. See “How to optimise your application’s caching efficiency” for details.

Slow request handling

A reason for all your Processing Units being busy could also be that it’s taking your application longer than necessary to respond to requests. Your application might do something computational-heavy. Or it might actually wait for one of the services it depends on to deliver a result. freistilbox already provides you with a number of these services itself, for example MySQL, memcached, Redis or Solr.

Long-running database queries, for example, are a common cause of busy PU. There are many reasons queries might take many milliseconds or even several seconds:

  • complex queries like multi-table JOIN’s that force the database to store intermediate results in disk-based temporary tables
  • huge result sets that need to be transferred over the network
  • missing indices, resulting in full table scans over the whole data set

In the worst of cases, it’s actually a combination of all of the above!

Your application might also depend on a third-party service or an external API. If that service is responding slowly or the connection between it and your application is congested, this will also cause your Processing Units to stay busy for far longer than you expected.

Tools like New Relic or the Drupal Devel module will help you determine where your application spends its processing time. If you experience unexpectedly high response times, use these tools to detect bottle necks and tuning opportunities.

“Can you help me find performance bottlenecks?”

Yes, we’d be happy to! Simply send us a support request describing in detail where you see poor performance with your web application. We’ll provide you with additional insight from our internal performance monitoring and with helpful tips from years of experience in running high-performance websites.

Please note that in-depth application performance analysis and optimisation is not covered by our free hosting support and will require an individual consulting process.