Wednesday, I talked about the new free search engine that Yahoo! and IBM have made available for download, We on the product team are gratified at the rave reviews it has garnered, but we weren’t as clear about the hardware requirements as we should have been. Larry Dignan of ZDnet had an excellent post about how our free search engine still requires hardware, a point first made by Jason Berberich in a comment to a post on Nick Carr’s blog. So I asked a few questions about what kind of hardware we really need.
Normally, people wouldn’t spend a lot of time checking out the price of hardware for a piece of software, but this is a special situation, because our new search engine is inevitably being compared to Google Mini and Google Appliance. Those offerings are not free, but they do come bundled with a piece of hardware, so it’s legitimate to ask about the cost of hardware required by our free software.
And, in our passionate focus on simplicity, we didn’t provide the details on what hardware you needed for different size search needs, we just put in a minimum recommendation for testing and a recommended configuration for our maximum 500,000-document capacity. So it was perfectly reasonable for bloggers to conclude that we were recommending that configuration for any use of our search engine—we kinda said that.
But it’s kinda not true. We were being conservative. Our recommended configuration is what you need to run a robust production implementation for half a million documents. So we decided to provide more information (which we should have done in the first place) about what kind of configurations we recommend for fewer documents. We updated our product information page earlier this evening (click the learn more link to see all the details).
So what does that mean for our hardware costs? Let start at the low end.
Our minimum hardware, which can handle up to 20,000 documents, requires a single 1.5GHz processor with a gigabyte of memory, and 80 free gigabytes of disk space. I’m sure that you could find many servers around that met these requirements, but, hey, I work for IBM! So I looked for the most inexpensive IBM server that would do the job and configured an IBM xSeries 100, Model 8486E2U, which sports an Intel Celeron D processor 326 at 2.53GHz, one gigabyte of memory, and 160-gigabyte disk drive for $843.
Now it’s not fair to directly compare the Google Mini to our search engine, because they have different features that might be more appealing to some customers than others. Although we’ve been praised for how simple our product is to set up, some would argue that it’s easier to plug in a Mini. Other folks who have a server lying around would claim our solution really is free. Still others would remind you that their Web server is maintained by a hosting company. They can’t exactly stick a Mini into that hosted environment. So, different people have different opinions and different needs.
But for those people that have really have a choice, price is important. And Google Mini’s lowest price is $1,995 (although that buys you 50,000-document capacity, not our 20,000). So, you could legitimately ask if Google’s cost of ownership is lower than ours for 50,000 documents.
Luckily for us, it’s not. For 100,000 documents, our newly-minted recommendation is for a single dual core processor of at least 2GHz, with one gigabyte of memory and 100 gigabytes of free disk space. I priced an IBM server (xSeries 100 Model 8486E7U) configured with a Pentium D processor 820 at 2.80GHz, one gigabyte of memory, with a 160-gigabyte disk drive for $1,083, less than Google’s price for half the number of documents. Google’s price for a 100,000-document system is $2,995.
We also looked at the high end of our capacity, the 500,000-document configuration. What would it set you back to buy a 3GHz dual processor server with two gigabytes of memory and 250 gigabytes of free disk space? I was able to configure an IBM System x3400 Model 797312U with one Intel Dual Core Processor Model 5050 and one Intel Dual Core Processor Model 5110, with two gigabytes of memory, and a 500-gigabyte disk drive. It cost $2,866, compared to $30,000 for the low-end Google Search Appliance that handles 500,000 documents.
To be fair, Jason believes that the hardware you get with a Google Search Appliance is priced at $8,000, so spending less than $3,000 might not be a fair comparison to the power you’d get. Still if you want to, you can go buy the $8,000 server to use with our free software and get some change back from your $30,000.
Now those who know me understand that I am no expert on computer servers. (They are laughing at this point because they know how much I am not an expert—you know who you are.) So I apologize in advance if I have made an error here. If you want to have mirrored disks and maybe some other features on your servers (which are probably excellent ideas for many uses), these prices would probably be higher. But my suspicion is that even with these extras, our free software helps you come out ahead. Larry calculated that a server meeting our 500,000-document requirement with mirrored disks still cost less than $5,000.
So, to sum up, the bloggers making this point about our total cost of ownership are absolutely correct that we should talk about hardware costs, too. And their analyses were based on exactly what we wrote on our Web site. But as they say in the NFL, “Upon further review,” we were being overly pessimistic about what we require—in our zeal to be simple. Sometimes simple can be oversimplified, and we needed to correct that. Thanks to the bloggers who raised this point and gave us a chance to explain ourselves better—let’s keep the conversation going.
