Optimising an eCommerce Site

Our development work this week has included a lot of server optimisation for an eCommerce website and we dramatically improved response times, lowered overall CPU use and memory use by making changes to the server configuration.

The benefit is a better experience for the Visitors, Customers and Google (page load speed is a factor in Page Rank) and it also enables the server to cope with higher volumes of  visitors without becoming overloaded.

All of the changes were made to the configuration of the server itself, no changes were made to the website to gain these benefits.

Memory

We started by improving memory use, as this was frequently consumed and caused the site to become unresponsive and on rare occasions it become unusable.

The main changes came from changes to the Apache, MySQL and PHP configurations to reduce peak memory use to something the server could cope with.

MySQL

The first thing to do was analyse the MySQL peak use, which is a complex affair but our tools enabled us to quickly discover what the peak memory requirement is and enables us to identify areas we can safely change.

There were multiple changes made to the MySQL configuration that brought the peak memory requirement down from 18Gb to 6Gb, which ensured that MySQL remained in RAM and didn’t get swapped out to disk all of the time and that is very bad.

Apache & PHP

Apache was tweaked to cope with the regular demands by modifying the number of servers it keeps running and the memory they each consume. The changes themselves were small but they are multiplied by the number of servers that are created so the peak memory usage drops significantly.

PHP was quite an easy one to fix, this was a simple case of dropping the memory used per instance and there was one of these per Apache server so even small changes get multiplied to save lots of memory.

CPU Use

Having monitored the server for some time there were peaks in the load that corresponded with some MySQL queries being ran, we investigated further and  identified some particularly slow, and CPU intensive, queries by setting up and monitoring the slow_queries log.

MySQL

The CPU optimisation came by setting up the correct MySQL indexes to match the queries being ran, in a number of cases this dropped queries that were taking above 40s (yes! 40s) down to 0s (instant). There was a single query that was taking anywhere between 85s and 240s to complete that was painful to watch but thankfully this was just from the Admin interface and was ultimately brought down to 0s with correct indexes.

XCache - PHP caching

To simplify things, PHP is a text script that is interpreted every time it is loaded, executed to generate the page and then discarded. This is repeated every time the page is requested and this all takes time.

With this in mind, the way to improve response times is to use a PHP opcode cache, in our instance called XCache. This is an Opcode Cache that stores the interpreted PHP scripts and a way that is very fast to execute and generate pages. This in turn reduces CPU load and also improves page response times as the pages are served much quicker.

We installed, configured and monitor the installation of XCache and at the time of writing the cache is serving over 80% of pages from its cache.

Page Caching - Not sending content

Page Caching is the means that your browser stores files locally and doesn’t download files from the server every time it wants it.

The dilemma is: long expire times mean fewer hits on the site *but* this isn’t great if you’ve got frequent content updates as you want the new content to be sent to the visitor.

The improvement here was to get a correctly configure the expiration times for numerous different file types. E.g., dynamic webpages expire within 1hr, the product images expires after 1yr. Once an item expires, the browser will refetch the original.

Bad Robots - stopping leeches

There are lots of computers out there looking to steal your content, hogging your bandwidth, crawling through your site and generating page requests that don’t bring you any business at all.

There is a ‘nice’ way to request they don’t crawl through your content by using a file called ROBOTS.TXT but it’s not a requirement that servers use it and most malicious ones simply ignore it.

There is a way to block the robots from every connecting with your site that’s quite simple to do and we have configurations that block over 75 known bad computers and prioritise genuine visitors and good search engines like Google, Bing and Yahoo.

Summary

Overall, the changes highlighted above along with some other smaller tweaks enabled the server to cope with a higher volume of visitors and also improve the response times for each customer.

These are general principles for well configured servers that can be applied to pretty much any website without the need to change the website itself.

Contact us if you’d like us to improve your visitors experience or maybe you think your existing server just isn’t up to scratch.

Filed under: