| OCR Text |
Show reproduced, allowing all of the ATQ to hardcopied on just four letter-size pages. However, at 1.5 hours per quarter-view, or 6 hours total, this is still a ridiculously slow way to get hardcopies. Writing Postscript code that draws tiles directly may solve this problem, but this is no trivial task; especially since Postscript uses the "paint" model to draw, and has no concept of XORstyle drawing. What is really needed is the ability to get a condensed overview hardcopy, which is on the drawing board, so to speak! • Long turn-around time for simulation. This is only partly attributable to slow network preparation and tracing. Switch-level simulators are also notoriously slow for large circuits. As for INSTED, besides trimming the fat from ports to reduce paging, another improvement to the :TRACE method immediately suggests itself. Instead of using a list to keep track of ports associated with a node name, use a hash table instead. Hashing on a node name is much faster than doing a linear list search, since it can be done in constant time, given a large enough hash table. The real solution, however, is to avoid using any data structure associating ports with node names, and to simply tag each port with a node name as it is being traced. This requires, of course, that ports with given names be traced first, but that is, in the jargon, a "five-minute hack" that should result in an appreciable, perhaps order of magnitude, speedup in port tracing. • No back-up machine! Unexpected crashes destroy hours of work, while downtime does in the ability to do any. Solution: Either get more Lisp Machines (extremely unlikely) or else implement INSTED's methods on other, more plentiful, less costly Lisp workstations (already underway.) • No interface to other systems. A solution to this problem is in the works. However, if hierarchy is used, the circuit must first be flattened in order to write a file in a format existing PPL editors can read; but since the existing systems do not support hierarchy, there is no easy way for them to write a file that INSTED could read and use to re-create the hierarchy. • No common database. Granted, a common database is desirable; however, the fact that Structured Tiling is a superset of PPL creates an upwardcompatibility problem for PPL-based tools, as noted above. The database management issue is more fully discussed below. • No "undo" command! Actually, deleted regions can be retrieved, but there is no generic undo operation that would allow the user to "unwind" the command history to an arbitrary depth. Nevertheless, this has high priority on the "desired features" list. 77 |
