Search Done Right

Google just announced some improvements to its search appliance (“New Google Search Appliance Connectors“, “Compare enterprise search relevance using Side-by-Side in Enterprise Labs“). As always with the appliance announcements it’s very interesting to read between the lines. The experimental side-by-side feature looks cool but what really caught my attention is the mention that their SharePoint connector now supports “bulk authorization“. Now, wait, bulk authorization? Didn’t Google just announce support for early binding?

Bulk authorization is a late binding method where multiple URLs are checked at once. This may reduce the latency a little but still has all the limitations of late binding. Why would Google not use its newly touted early binding capability with SharePoint when SharePoint is the prototypical example of where it is needed. In SharePoint most users can only see a very small part of the content which makes late binding extremely inefficient.

A little more research leads to the Google technical documentation on access control. Unlike the marketing documentation, the technical documentation never mentions early binding and the only mention of ACLs relates to the new Policy ACLs feature (see the Google group or Adriaan Bloem’s post discussing early binding)

Policy ACLs check each URL against a set of URL patterns/ACLs associations passed by the administrator. Notwithstanding the fact that this is totally unpractical to manage and does not scale to a large number of patterns, it is not early binding but just a more efficient version of late binding, i.e., post-query result by result checking. In particular it still makes it impossible to support an exact result count or parametric search in a reasonable amount of time.

Talking about parametric search, in this same announcement Google’s product manager totally dismisses it, claiming that users only care about the top results.

Hmm… odd. Didn’t the Google appliance also tout parametric search for its 5.0 launch? Once again it seems impossible to find any documentation for it and this time, not even in the marketing datasheet. The only reference to it parametric search is an experimental module whose documentation does not inspire much confidence: “Google does not warranty or in any way guarantee the reliability of this code. Use it at your own risk.”

Which is not too surprising given what its actually doing: relaunching a search in Ajax for 100 results and calculating the facets in JavaScript…. now, that’s a hack!

Enterprise search is a hard problem. Security, connectivity, user experience are difficult aspects that require careful consideration and solid technology, not hacks. Parametric search and early binding are two well defined critical aspects of an enterprise search deployment, each used by more than 80% of our customers. Google bashfully announced their support in v5 and v6 while clearly not supporting them. This kind of irresponsible marketing is not only creating a lot of confusion in the market (the intended goal) but ultimately making customers unsuccessful and delaying the adoption of search in the enterprise. It’s time for customers, analysts, the trade press and even vendors to call the bluff out and bring back the enterprise search market to reality!

This entry was posted on Thursday, October 1st, 2009 at 10:15 am and is filed under Uncategorized. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.