All about online advertising

  • Increase font size
  • Default font size
  • Decrease font size

Look before you leap - Even for a second!

E-mail Print PDF

The old saying "look before you leap" is still true in today's world of high tech computing. If anything, yesterday showed us it applies as much to computers as it does to us humans. As you may have heard that the keepers of world time added an extra second to June 30th, 2012 to align it with Earth’s rotation time. This extra "leap" second brought down many servers to their knees all around the digital world. Before I get into the details of what happened, let’s talk about what was impacted.

What was impacted?

There were a lot of prominent websites that were impacted by this leap second that you can read all about on Techcrunch and other sites. Here is a list of systems that I am aware of that were impacted:

  1. Versions of Linux kernel less than 2.6.9-89.EL (RHEL4) or kernel-2.6.18-164.el5 in RHEL5
  2. Some versions of OpenManage from Dell on Linux systems. If you run primarily Dell servers this might be your issue
  3. Apparently any multi-threaded application that relied on Linux’s futex for synchronizing thread access
  4. Tomcat server
  5. JBoss
  6. Casandra
  7. Hadoop
  8. Potentially, MySQL
  9. Firefox, Thunderbird and Chrome

What happened?

Seems like there were a few different systems that were impacted by the leap second:

  1. Linux – On some older (or not patched) versions, Network Time Protocol (NTP) servers had a bug that would cause a livelock. The concept of leap second is not new to most modern servers. In fact, NTP has built in functionality to handle it. On the day of leap second shift, NTP informs the servers that they need to strike 23:59:59 two times. A newer version of Linux Kernel uses "hrtimer" to handle leap second shift, this is known to cause livelock in the systems under some circumstances. The new patch uses "second_overflow()" method that deals with the situations more gracefully. If you were not using the patched version of Kernel then you would be updating the leap second with hrtimer that could have caused issues on the server. This also seem to have impacted the functioning of futex which most of the multi threaded applications use. Because of this a plenty of java based servers - JBoss, Tomcat etc. were impacted.
  2. OpenManage – Some versions of OpenManage software from Dell that is used to manage Dell servers were impacted. This was because of a bug in OpenManage. Restarting the servers solved the problem.


Typical symptoms if you are hit by this leap second issue are the following:

  1. Total crash of Linux server
  2. Livelock within Java based systems that rely on futex. This will cause high CPU utilization, in the range of 150% - 250%, high load average on Linux
  3. High CPU utilization on Linux servers by system processes
  4. The systems will still continue to respond but might be sluggish


There are a few things that one can do to fix this issue:

  1. Shutdown NTP and manually set the time and start it again:
    sudo /etc/init.d/ntp stop
    sudo date -s "`date -u`" 
    sudp /etc/init.d/ntp start
  2. Reboot your system
  3. It is always good to patch the kernel with the latest code




Last Updated on Tuesday, 03 July 2012 15:45

Click Fraud – Definition, Detection and Mitigation

E-mail Print PDF


One of the biggest issues facing the online advertising industry today is click fraud. A very large portion of online campaigns rely on click through rate (CTR) as a measure of success or some variation thereof. Hence, if the clicks being registered are fraudulent then that will throws off the entire foundation of assessing the success online campaigns.

Some click frauds are malicious while others are innocent. Whatever the case maybe both types of click frauds are undesirable. A malicious click fraud could be performed by an advertiser or a publisher. In either case a bot (short for a robot) or a person is involved who intentionally clicking on ads. A publisher would do this to artificially make the performance of their site look good. An advertiser would do this to their competitor who has bought search terms on CPC (cost-per-click) basis to reduce their budget and hence advertising power.

An innocent click fraud could be mistakenly performed by a search engine crawler who crawls a page and every link on the page. Some of these links are clicks tracking links for the advertisement on the page. Most of the ads are delivered using Javascript these days. As a result, merely crawling pages with ads on them will not cause an impression or a click. This is so because to cause an impression to load or a click to happen the Javascript has to be executed and rendered by the crawlers. Since they don’t do that, chances of them causing lots on unwanted clicks and impressions is quite low. Older technology for delivering ads could suffer from this kind of click fraud. There are some web proxy software that try to cache pages of a sites to make them quickly accessible for their users. They might unintentionally causes fake impressions and clicks.

Last Updated on Wednesday, 26 August 2009 21:41

Yield Optimization

E-mail Print PDF


Yield optimization is a technique utilized by ad servers to improve the performance of a given advertiser creative. In this technique the ad server tries to identify publisher impressions which are working well as per campaign parameters from the impressions that are not. It then tries to place more and more creatives on the impressions that are working and less on the ones which are not. Eventual goal is to place all creatives on the impressions which are working well. Yield optimization could be as rudimentary as tracking CTR (click through rate) for a given site and optimizing creatives based on it. On the other hand it could be as sophisticated as feeding a host of campaign specific parameters, like time of the day, publisher, ad size, geographical location, channel, price etc, into a machine learning system and letting the machine make the decision based on all those parameters about creative placement.

What you need to know about yield optimization: There are a couple of things you need to know about yield optimization:

Last Updated on Wednesday, 26 August 2009 21:42

Retargeting – When is it effective?

E-mail Print PDF


What is the best way of targeting the users on the web? There are theories after theories about what the best way is and how other theories are incorrect. I would say - target the users who have the maximum propensity to buy your product and you are in business. The question is how do you find such users? One low hanging fruit to targeting users who might be interested in your product is to retarget them. This simply means that you target these users based on the fact that you have ‘seen’ them somewhere before and you can infer something about them  based on where and when you saw them before. For example, if a user was ‘seen’ on a business related site in the morning they might be interested in business related products and offerings.  Another example could be that a user was ‘seen’ on Apple’s online store configuring a MacBook, this user did not eventually complete the transaction and exited the site. Yet another example could be of a user who partially completed a credit card application and then exited the site. If you have ‘seen’ these users before, you can target them again on the web and you have a pool of users which are more likely to buy something from you then the rest of the users online. Anecdotally speaking, retargeting comprises of 80% of performance for most campaigns.

Last Updated on Wednesday, 26 August 2009 21:42

Related Articles