Tech News

The debate around software as a service usually focuses on whether it's better or worse than conventional licensed software. The answer is neither--businesses that have adopted SaaS will tell you it's good for some things and not for others.

But that's only one lesson gleaned from this bigger truth: You can't use the same metrics to assess SaaS as you use for conventional software. To understand SaaS requires a new mind-set.

When IT and business managers shop for software, important attributes include the ease with which it can be customized and how well it can integrate with other systems. But SaaS offerings typically can't be customized, only configured. And integrating SaaS with on-site applications can be more difficult than just integrating two on-site apps, often requiring vendors' help to deal with code dissimilarities.

This doesn't mean SaaS is inferior to conventional licensed software, only that it has different strengths. Many businesses are coming to accept those differences: In a survey of 471 business technology professionals by InformationWeek Research, a quarter of whom are SaaS users, SaaS got high marks for its upgradability, reliability, and ease of use. SaaS users were less enthusiastic about its cost, with only 32% saying they found SaaS affordable. It also didn't do well on customization, integration, and the ability to switch among vendors.

Tech Type1

With recent advances in local area networks, networks of workstations differ from massively parallel processors primarily in packaging, cost and software emphasis. The key open architectural question is the nature of the network interface or the communication architecture: its hardware organization and logical abstraction as a basis for communication.

Active Messages represent a RISC approach to communication, providing simple primitives, rather than solutions, which expose the full hardware performance to higher layers. Active Messages are intended to serve as a substrate for building libraries that provide higher-level communication abstractions and for generating communication code from a parallel-language compiler, rather than for direct use by programmers. It is currently in use at UC Berkeley by the Fast Communication layers (sockets, RPC and MPI), the xFS parallel file system, the Split-C and Id compilers, as well as in other libraries like Scalapack.

This project investigates Active Messages on a broad range of hardware, including a dedicated message processor per node (Intel Paragon and Myrinet) an FDDI interface at the graphics bus of a high end workstation (HP 735 with Medusa), and a conventional interface to the next generation LAN (Sparc 10 with Sahi-1 ATM). We will thus demonstrate concepts by construction and evaluate them on real programs on real machines. The result will be a clearer understanding of the communication architecture and trade-offs in the hardware organization of the network interface.