Although there were a lot of announcements around cloud computing this week, I want to use this forum to clarify and expand a little on my comments last week about the marketability of grid computing. Besides, I don’t think cloud computing is going anywhere, so we’ll have plenty of time to examine this news in the weeks to come.
As for my comments, which were based upon a discussion with Forrester’s Frank Gillett, I received some feedback suggesting that the piece unfairly cast grid computing in a negative light. This was in no way my intent, however, nor do I believe it was Forrester’s intent in publishing Frank’s report. The point of the report — and the point I was trying to drive home — is that “grid computing” is not a term likely to sell many products in the enterprise IT space.
Grid computing clearly has many uses across many industries, and the reports from organizations that have adopted it have been overwhelmingly positive thus far, but the reality is that the term (and the technology) have been hyped, tweaked and personalized so much that many potential users no longer have any idea what it means. Those of us who eat, drink and breathe grid computing for a living understand just how complex and far-reaching its universe is, but we are definitely the minority. When IT decision-makers at most enterprises hear “grid computing,” they don’t think “mission-critical,” “transactional,” “service-level agreements” or “easy.” Unfortunately, many of them still think “compute-intensive,” “cycle scavenging” and “infrastructure overhaul.” Today’s grid technologies have advanced into areas rather distant from grid computing’s traditional embarrassingly parallel sweet spots, but it doesn’t appear to me that end-users have all caught on.
As Frank and I discussed, this isn’t news to providers of grid technologies, many of whom have at least attempted to change their marketing to focus on solutions rather than technology. This is where I think it really becomes clear that grid technologies are far from dead, they just need to be sold differently. I would reckon that if the same respondents to Frank’s survey were asked if certain business benefits were of interest to them, or whether solving particular datacenter problems would be appealing, these people would almost certainly express interest. Grid computing and its brethren can tackle a broad spectrum of business priorities, and the technologies will no doubt see even greater adoption among non-cutting-edge users if they are introduced business objective-first instead of technology-first.
I also stopped by the ServerSide Java Symposium on Friday, and while there was no “Oracle buys Tangosol” buzz circulating around this year’s event, I did take in an interesting presentation by eBay’s primary search architect, Randy Shoup. Shoup, a distinguished architect with the online auctioneer, ran through the now-standard list of eBay stats — 276 million users, 2 billion-plus photos, $2,000-plus per second in sales, 1 billion-plus page views per day, more 2 billion petabytes of data, rolling out more than 300 features per quarter, etc. — and told the room how eBay handles it all.
It boils down to four things, said Shoup: partition everything; async everything; automate everything; and remember that everything fails. When it comes to partitioning data, for example, he said eBay has 1,000-plus logical databases that each are split horizontally. The same holds true for the application tier, where eBay has partitioned its approximately 16,000 app servers into pools, which also are split horizontally. Doing asynchronous data updates, he said, allows eBay to deliver near-real-time updates where necessary while sacrificing other, less-important updates. Batch processing is still applicable and efficient for certain things, Shoup noted. Automation is critical, he said, because scaling is easier with machines than with humans, and machines are cheaper, as well. In addition, machines can consider more factors in making decisions, and they can adapt to changes faster. As for failure mitigation, Shoup explained that eBay doesn’t make any changes that can’t be undone, and it logs problem fixes to speed repair in future situations.
As for those cloud computing announcements I mentioned, they are: “IBM, Unis Combine Autonomic, Cloud Computing”; “Yahoo Ramps up Cloud Computing Research”; “Elastra Unveils Cloud Server”; “Amazon Web Services Launches ‘Elastic IPs’ for EC2”; “RightScale Releases Reliability Controls for Cloud Computing”; and “MuseWorx Launches Cloud-Based Marketing Solution.” Also, I really suggest reading our two executive Q&As, in which the CEOs of Tideway Systems and Digital Ribbon discuss datacenter optimization and large-scale utility computing, respectively.
—–
Comments about GRIDtoday are welcomed and encouraged. Write to me, Derrick Harris, at [email protected].