A CDS Standard?

The risk involved with investing time and effort to adopt any new tool or approach is often reduced if that new tool or approach has been accepted as a standard, de facto or otherwise, thereby increasing the likelihood that it will be widely implemented, supported, and accepted, and that the sharing of tools and techniques based upon the approach can be maximized. It is therefore in the best interests of CDS users for CDS (or something like it) to become "standardized" in some way.

Other recent standardization efforts, such as for HPF and MPI-2, can understandably make people wary of becoming involved in such a task for CDS. Those recent efforts required a substantial investment of time and expense overall, principally to fly the many participants from many different institutions and companies and to organize them into a meeting at a common location, approximately 8 times/year. The wisdom of repeating such a large investment with no assurance of eventual acceptance of the resulting standard is arguable. Everything possible would need to be done to decrease the investment and increase the likelihood of acceptance and widespread implementation.

elepar believes that the creation of a CDS standard is an important goal, and it believes that the majority of the standards process can be as effectively (if not more effectively) carried out on line as in person. Success will first hinge on adequate interest, and then adequate announcement of the effort as well as sufficient and appropriate mechanisms for facilitating the on-line process. elepar will be happy to be involved in any or all of these steps, to the extent that it does not interfere with the impartiality of the process (or the appearance thereof).

Small efforts have already been made in this direction. There was a Birds of a Feather (BOF) at SC97 entitled "PICIS: A Paradigm-Independent Communication Interface Standard (or, is it time to standardize an inter-process, rather that inter-processor, API)", chaired by elepar's founder. Both CDS and Treadmarks were discussed as possible directions for such a standard. The attendance of this BOF was small (attributable to other popular BOFs scheduled at the same time), but those who were present were quite interested in the CDS approach, and some are still quite supportive today. Also, recent discussions on the Usenet group comp.sys.super have turned up signs of support from that community as well.

elepar recommends that people become familiar with CDS (e.g. through this web site) and attempt to determine the interest of those around them, either through direct contact or through public fora such as the Usenet groups (e.g. comp.sys.super, comp.parallel, comp.architecture).

There is also now a cds-std mailing list now at eGroups for discussing CDS and its standardization. (eGroups is free and basically is identical to a web-archived majordomo mailing list, but because it is commercial, messages will contain small advertisements. Email addresses in the archive are obscured to prevent serving as a spam creation DB.)