Six Tips to Prepare Online Retail Servers for Cyber Monday

Cyber Monday, the busiest day of the year for online sales, is fast approaching and with time running out, ForLinux, one of the UK’s leading providers of open source hosting and IT support, is reminding Internet retailers to perform essential server maintenance or potentially miss out on bumper sales.

Monday, Nov. 28th, is the date when the Internet industry predicts Internet transactions to reach their highest level ever, with sales predicted to top the £1028m* spent online last year.

Steve Nice, technical director at ForLinux, has been advising his clients to prepare for Cyber Monday for several months. “It’s crucial that online retailers prepare for the upsurge in traffic that happens on Cyber Monday. Time is running out to make sure servers and applications are ready to withstand the demand. With so little time left to prepare, we’re encouraging online retailers to follow our six last minute tips: check your logs, bandwidth, database, RAM, Apache configuration and reboot.”

Click through for six tips to ensure your servers are running at peak performance on Cyber Monday, as identified by ForLinux.

Check free disk space for logs. Increased traffic will see an increase in the log size, and this can quickly use up any remaining space. You can start by checking the amount of space your logs are taking up by running

du -sch /var/log/

and then check how big the partition is by running

df -h

“You will need to allocate additional disk space to your log partition, which can be changed back after the busy period,” explains Steve. “Assign a partition to the logs to avoid server freeze. If the partition fills, the application will continue to respond but logging will cease.”

If the server has been built using LVM, operators can see how much space is free to assign to partitions by running

vgs

Logs should also be rotated more frequently — weekly rather than monthly, for example. Review the logrotate settings by typing

less /etc/logrotate.conf

Online retailers should also check the logs for any warnings. Some default builds of services such as PHP have all levels of logging enabled by default. A quick way to check is by running the following command

grep warn /var/log/httpd/*

As well as checking for warnings, ForLinux recommends keeping several copies of the logs and making sure the rotate happens after the stats have been run, otherwise log entries will be lost.

Make sure the site’s current connectivity limit is burstable. The hosting provider will be able to provide this.

Find out how much bandwidth has been used so far in November. If the limit is close then it’s best to talk to the host in advance and buy additional bandwidth as it is often cheaper than the rate charged for excess bandwidth.

Tune my.cnf. With increased activity comes increased load, which could cause a site to reach pre-set defaults and the applications to refuse new connections. To avoid this, run MySQL tuning script against the MySQL database and increase limits where possible and in line with predicted traffic. Ensure that the database server has been on-line for at least 48 hours without being restarted to get the best test results.

Also, ensure that there are no corrupt tables, and that all tables have up-to-date indexes using the MYSQLCHECK command. You should also remove any unused databases from the service, which not only frees up disk space, but reduces the resources used for caching.

Using the command “free” will show the amount of RAM available on the server

free -m

When this memory is used up, SWAP memory will be utilised instead, resulting in significantly reduced performance as SWAP is disc-based. If possible, increase physical RAM and do not just add more SWAP as this may only cause more problems.

Review default settings for things like MaxClients, Timeout and the number of allowed child processes. The apache Timeout is usually set to 120 seconds, by default. It is a good idea to reduce this to 20 seconds or lower depending on the applications that are running on the server.

Also check the memory limits in the php.ini file, e.g. If the server has 2GB of RAM and your php.ini allows processes to use a maximum of 128MB, then setting the MaxClients directive in apache to 300 may cause problems with the amount of the RAM being used on the server.

If the system fails, a reboot could result in the server going into FSCK (Disk Check), which can take hours to complete, particularly if the server hasn’t been rebooted for a long time —  as this is dependent upon the size of the hard drive and bad blocks found. Schedule the reboot over a quiet period as soon as possible, as waiting for the disk check to finish means sales could be going elsewhere.