Extreme multicore licensing
Azul is taking the current trend towards multicore chips to an extreme,
but this brings its own problems — one of them being
potentially outlandish software licensing costs. In the world of
general-purpose processors, software makers such as Oracle have argued
they are due higher fees if customers use chips with multiple cores,
since the cores are having a direct impact on the benefit customers are
getting from the software.
Ditto for application server vendors. Because the Azul system is running applications through a JVM proxy, both the general-purpose server and the Azul appliance have to be licensed for the application server. If each Vega core were viewed as a separate chip for licensing purposes, licensing costs for that 384-core Azul system would quickly consign it to the scrapheap.
No set way has really been arrived at to deal with this dilemma yet, but Azul says it is working on it, and has arrived at a bizarre but perhaps workable arrangement with BEA. The Azul system itself doesn't incur licence costs, but the server tapping into the pool is licensed at three times its normal fee. "Even though servers that become authorised to use the Azul compute pool can tap into virtually unlimited power, they only incur a per-CPU charge of 3x the traditional amount," according to a joint statement from the companies.
This is justified by using a "power station" analogy, with the Azul compute pool in the role of power station. "There is no utility metering at the power station level — the metering is done where the decision to use the power is made, at a house for example," the companies say.
"This simplifies things a lot for BEA," says Azul's Khan. "Otherwise they would have to worry about how to charge for a 384-core system."
Illuminata analyst Gordon Haff said the analogy doesn't completely make sense, but the arrangement could work. "If it's a bit unusual, it's obviously a scheme that both companies felt they could live with and which would be acceptable to their customers," Haff says in a recent research note. "And BEA avoided truly coming to grips with multicore licensing — which they may have seen as the biggest win of all."
