You are currently browsing the monthly archive for May 2011.
Last week I attended the IBM Big Data Symposium at the Watson Research Center in Yorktown Heights, N.Y. The event was held in the auditorium where the recent Jeopardy shows featuring the computer called Watson took place and which still features the set used for the show – a fitting environment for IBM to put on another sort of “show” involving fast processing of lots of data. The same technology featured prominently in IBM’s big-data message, and the event was an orchestrated presentation more like a TV show than a news conference. Although it announced very little news at the event, IBM did make one very important statement: The company will not produce its own distribution of Hadoop, the open source distributed computing technology that enables organizations to process very large amounts of data quickly. Instead it will rely on and throw its weight behind the Apache Hadoop project – a stark contrast to EMC’s decision to do exactly that, announced earlier in the week. As an indication of IBM’s approach, Anant Jhingran, vice president and CTO for information management, commented, “We have got to avoid forking. It’s a death knell for emerging capabilities.”
The event brought together organizations presenting interesting and diverse use cases ranging from traditional big-data stories from Web businesses such as Yahoo to less well known scenarios such as informatics in life sciences and healthcare, by Illumina and the University of Ontario Institute of Technology (UOIT), respectively, low-latency financial services by eZly and customer demographic data by Axciom.
Eric Baldeschwieler, vice president of Hadoop development at Yahoo, shared some impressive statistics about its Hadoop usage, one of the largest in the world with over 40,000 servers. Yahoo manages 170 petabytes of data with Hadoop and runs more than 5 million Hadoop jobs every month. The models it uses to help prevent spam and others that do ad-targeting are in some cases retrained every five minutes to ensure they are based on up-to-date content. As a point of reference CPU utilization on Yahoo’s Hadoop computing resources averages greater than 30% and at its best is greater than 80%. It appears from these figures that the Hadoop clusters are configured with enough spare capacity to handle spikes in demand.
During the discussions, I detected a bit of a debate about who is the driving force behind Hadoop. According to Baldeschwieler, Yahoo has contributed 70% of the Apache Hadoop project code, but on April 12, Cloudera claimed in a press release, “Cloudera leads or is among the top three code contributors on the most important Apache Hadoop and Hadoop-related projects in the world, including Hadoop, HDFS, MapReduce, HBase, Zookeeper, Oozie, Hive, Sqoop, Flume, and Hue, among others.” Perhaps Yahoo wants to reestablish its credentials as it mulls whether to spin out its Hadoop software unit. If such a spinoff were to occur, it could further fracture the Hadoop market.
I found it interesting that the customers IBM brought to the event, while having interesting use cases, were not necessarily leveraging IBM products in their applications. This fact led me to the initial conclusion that the event was more of a show than a news conference. Reflecting further on IBM’s stated direction of supporting the Apache Hadoop distribution, I wondered what IBM Hadoop-related products they would use. IBM will be announcing version 1.1 of InfoSphere BigInsights in both a free basic edition and an enterprise edition. The product includes Big Sheets, which can integrate large amounts of unstructured Web data. InfoSphere Streams 2.0, announced in April, adds Netezza TwinFin, Microsoft SQLServer and MySQL support to other SQL sources already supported. But this event was not about those products. It was about IBM’s presence in and knowledge of the big-data marketplace. Executives did say that the IBM product portfolio will be extended “in all the places you would expect” to support big data but offered few specifics.
IBM emphasized the combination of streaming data, via InfoSphere Streams, and big data more than other big-data vendors do. The company painted a context of “three V’s” (volume, velocity and variety) of data, which attendees, Twitter followers and eventually the IBM presenters expanded to include a fourth V, validity. To illustrate the potential value of combining streaming data and big data, Dr. Carolyn McGregor, chair in health informatics at UOIT, shared how the institute is literally saving lives in neonatal intensive care units by monitoring and analyzing neonatal data in real time.
Rob Thomas, IBM vice president of business development for information management explained the role of partners in the IBM big data ecosystem. As stated above, IBM will rely on Apache Hadoop as the foundation of its work, but will partner with vendors further up the stack. Datameer, Digital Resaoning, and Karmasphere all participated in the event as examples of the types of partnerships IBM will seek.
IBM has already demonstrated, via Watson, that it knows how to deal with large-scale data and Hadoop, but to date, if you want those same capabilities from IBM, it will have to come mostly in the form of services. The event made it clear that IBM backs the Apache Hadoop effort but not in the form of new products. In effect, IBM used its bully pulpit (not to mention its size and presence in the market) to discourage others from fragmenting the market. The announcements may also have been intended to buy time for further product developments. I look for more definition from IBM on its product roadmap. If it wants to remain competitive in the big-data market, IBM needs to articulate how its products will interact with and support Hadoop. In my soon to be released Hadoop and Information Management benchmark research that I am completing will provide some facts on whether or not IBM is making the right bet on Hadoop.
David Menninger – VP & Research Director