[phylum]<CTamarin>Doc>Memo>Tamarin-criteria.bravo!1

The AISBU needs a spectrum of hardware offerings for its business. There is an immeadiate need for a high end machine. Over the long term there will be a need for additional hardware. Different choices of hardware have different attributes - different performanaces, different magitudes of effort. The memo attempts to describe some of the relevant criteria and to evaluate four classes of hardware with respect to this criteria.

Potential hardware can be divided into four major classes: Opcode compatible, standard microprocessor, microcodable with opcode changes, and Tamarin.

"Opcode compatible" hardware means using a processor that has user-writable microcode. This microcode would be written so that the machine would emulate the existing Interlisp opcode set.

"Opcode compatible with changes" hardware means using a commercial processor that has user writable microcode. Because of limitations in the microcode, certain Lisp opcodes would not be implementable efficiently. This would require some changes in the Interlisp opcode set.

"Standard microprocessor" hardware means using a processor that has a predefined and unmodifiable opcode set. The 68020, 80386, and MIPS processor are prime examples. The kinds of opcodes used in these machines are totally different than the Interlisp opcode set. To implement Interlisp on such hardware would require large changes to the compiler and internal Lisp data structures. The required changes would have to be examined carefully to ensure they would provide adaquate performance.

The criteria to evalute these classes of hardware are divided into two catagories: effort to implement and quality of result.

"Hardware Design effort" refers to the amount of time that is required to design and integrate the hardware.

"Software Design Effort" refers of amount of time required to design the changes to the Lisp system. This has to be performed by a Lisp wizard in the early parts of the projects. Design means the task of specifying the changes to the system. Evaluation means the task of determining if Interlisp can be successfully implemented on a potential piece of hardware.

"Software Implementation Effort" refers to the amount of time to implement all the changes to the Lisp system. This would be performed by the regular AISBU staff supervised by a Lisp wizard.

"Cost-Performance" indicates the cost of the machine for a given performance level. The numbers for a given machine are relative to the other machines in the table.

"Probably of Existence" means the chance that a suitable machine exists and that a Lisp implementation is possible on the machine.

There are several conclusions worth nothing in the chart. The short term choice would be an opcode compatible machine. However, such a machine is unlikely to exist. When one then considers trying to use "Opcode compatible with changes" machine, it turns out that the amount of changes would extend the schedule so it no longer had many advantages.

Using "Standard Microprocessor" hardware would be very attractive IF a high-performance version of InterLisp could run on it. Most of the hardware that Lisp has run used hasn’t been tailored to Lisp and there have always been archiles heels that have limited performance. I would expect such archiles heels to exist for standard microprocessors.

Tamarin has many advantages with good performance and provided a spectrum of hardware over a period of time. It is also more risky than the other alternatives due to its dependance on custom silicon.