In an interview with MarketingProfs.com, former Excite founder Joe Kraus was asked if there was anything he wished he had done differently at Excite. Part of his answer really came as no surprise:
[We] built our whole company on SUN hardware and EMC disk drives. That meant that it was very expensive to make our search index bigger. If you look at successful architectures today they are all Linux and Intel-based. They are very, very cheap.
This didn’t apply only to search. When I worked on Excite Inbox, we used incredibly expensive hardware. Front end web servers, for instance, were Sun Ultra 5 workstations. They were incredibly expensive (I want to say something like $20,000 a pop) and packed a scrawny amount of computing power, even by 2000-2002 standards. In contrast, about that time companies like Dell were offering dual CPU Intel rackmount servers at about $3,000 a pop (in the configuration we wanted). That’s just on the frontend.
The backend was more insane. Piles of Sun E4500 Enterprise Servers hooked up to large EMC storage arrays. Several of those E4500’s were running Oracle, a (very expensive) requirement of the backend mail system we were running. I’m pretty sure that Excite was almost single-handedly funding Sun to survive the dot-com implosion.
You neglected to mention that we spent quite a bit of time saying “this is dumb” why are we buying Suns. Also, you neglect that we also were pouring money into Oracle’s hands to do the database half of the puzzle that would require about a 4 hour outage to rebuild the indexes at least once a quarter.
I did mention Oracle briefly, but yes…we paid a lot of money for a component of the architecture that was among the least stable.
As for why we were buying Suns, if I remember correctly it was because OpenWave pretty much required it. I believe the only other platforms InterMail would run on weren’t very attractive. We couldn’t even run the frontends on lower cost hardware because InterMail didn’t provide libraries for other platforms (i.e. Linux or FreeBSD) and their IMAP servers didn’t perform adequately enough to put our frontends on Linux and use the JavaMail IMAP provider.
Of course, we could have funded a team to build a new mail backend that didn’t have such silly requirements (Solaris and Oracle). But I seem to remember on at least one occasion a product manager was heard saying, “but they’re Java programmers.” Thanks for the faith in our skills, BD.
How much of that money spent on Oracle licenses could have been spent replacing software.com with an existing (or even from scratch infrastructure?). Something like
(N Sun 4500s) * (31 Procs/4500) * ($$$$$/proc oracle licensing) => really big number!!!
Isn’t it fun to talk about all that hardware in casual conversation?
In the scramble to create new features (spell check, virus scanning, whatever) the mail team didn’t do a good enough job of making the business case for a different architecture. It’s hard to stop on the path you’re on and move to another. It’s one thing to complain about how silly something is (engineers do this so often I think that it becomes noise). It’s still another to turn those complaints into a reasonable argument for change.
Another good Joe K. Bit where he talks about Excite:
http://bnoopy.typepad.com/bnoopy/2005/03/the_long_tail_o.html
A blog by bnoopy:
bnoop bloggy blog?
During “Big Squeeze” a couple of us actually had proposed different mail architectures (it was really too late at that point). I seem to remember Martijn mentioning setups at other large mail houses where they built around low-cost machines using open source.
There was a lot of pushback by ops, though. They didn’t want to manage a lot of little machines. They were happier running fewer huge machines.