Java One Final Day Thoughts

It's the final day of Java One and my session was in the morning so I didn't get to see any of the keynote or other morning sessions.

Harnessing the Power and Performance of GlassFish Tuning for High Availability
Rene van Wijk, Tom Lubinski

This was a very rambling talk covering standard JVM tuning and general application server tuning. The talk obviously had a knowledgeable speaker but he ranged around many topics without real reasons as what he was talking about didn't necessary  match up with the current slide.

His talk mentioned File Store optimisation for JMS, Last Logging Resounce, Data Grids, No SQL, Bean pooling, file batching. I'm not sure many people followed it.

The talk then split to a monitoring demo which was a little salesy demo of RTView and dived off into monitoring of WebLogic not GlassFish.

Powerful Analytic Apps fueled by Java for High Performance and Scalable Analytics
Amit Rustagi eBay

This talk was by a Big Data analytics specialist at eBay and looked at how they use Java for analytics. The first half was interesting but light on specifics but to summarise;

eBay has 100s millions of users, 100,000s of application servers. Analytics 101 there are 4 layers;

• Acquire - Store structured and unstructured data
• Organise - Map Reduce
• Analyse - Data access layers and workflows
• Decide - Reporting and Data Mining tools

The missing layer is instrumentation to generate data for the metrics.   eBay has a number of platforms that does this instrumentation.

Tracking - tracks users from all channels including main site, affiliates, paid search, social media, mobile, marketing emails. For example they need to track how much should be spent on paid search so all conversion from paid search needs to be tracked and analysed. Within a few ms of the request eBay tracks and personalises the site.

Experimentation in eBay, they analyse the user and decide whether they should serve the user one of the experimental sites i.e. a site under test before full rollout. All decisions are based from the customer's persistent cookie and analytics of their likes and preferences. The user experience is then logged for further analysis.

Personalisation is a further platform where they can identify a user across marketplaces, Pay Pal and shopping.com and stubhub. All information is collected in real time and the sytem adapts your page on the fly.

Their key challenges are the same as everybody, Real Time, Performance, Scalability, Cost effectiveness. For the analytics team as they are in the request - response flow to do the tracking, Experimentation, Personalisation their performance is critical.

The talk then went into a basic description of standard Java design patterns but no detail about eBay so no real insights from the second half.

Steve Millidge