Just as I was thinking of packing up for the day Google announced the launch of the Google 6.0 Appliance. The focus of the announcement was very much on the scalability of the new version, with the capability of searching billions of documents. Much is made of the small hardware footprint of the appliance, as if this is a primary consideration in the selection process. Indeed it seemed to me that the product data sheet was an exercise in how many times the marketing staff could include the word scale in a seven-page pdf. How many organisations have billions of documents they need to search? They may well have billions of documents, but searching them may not be a good idea, because the result sets are going to be massive given that, in enterprise search the diversity of search terms is no where near the scale of the web. At the end of the day the Google product data sheet makes for some entertaining reading. I note that the GSA can search with ease and accuracy. However accuracy, the fraction of classifications that are correct, is not a relevant metric for information retrieval because of the high number of nonrelevant documents in any collection. Read p143 of Manning's book on IR if you do not believe me.

Of course another issue with billion document collections is how much it is going to cost over (say) a five-year life-span. Noticeably absent from the launch information was how much all this is going to cost. I would be very surprised if it was significantly less than other enterprise search applications at the billion document level. I have no doubt that the speed of install for the GSA will be very high, even for very large document collections, but that is not the same as speed of implementation, where user needs are determined and accomodated in the tuning of the appliance. Certainly it seems that there is better support for tuning, but that would not be difficult to achieve based on what has been provided to date.

It is also interesting to see the web background of Google creeping in, with the ability of being able to search cached pages even when sites are down. Sites? In an enterprise with billions of documents on file servers? Yet I am sitting here writing this post on a Dell running Isys Desktop Search which has functionality in terms of entity extraction and the ability to step through documents to view keyword occurences that are not yet available on the GSA.

The only successful enterprise implementations are when there is a clear statement of requirements against which the offerings of vendors can be evaluated and tested, and there is an investment in a search team to maintain search quality as business needs and collection management require. If Google V6 meets those requirements then you should have a win-win situation. Take into account that as of today there are over 10,000 messages on the Google Group for the ESA/Mini products. Is that what you regard as adequate customer support for an enterprise search product you have just let loose on a billion or more documents?  Google is continuing its commitment to enterprise search but to me it still seems to be focused on a smart sales pitch to an unsuspecting CTO than on meeting the search challenges of enterprise users. The lucky beta user quoted in the press release is systems integrator MTCST, with $75M turnover and 300 employees who is using it for an un-named federal client. Is that the best Google could do for a major product launch? See also the assessments in CMSWatch and Beyond Search

Martin White