As any database grows over time, data storage capacity eventually becomes limited, and the need arises to archive and purge old information. Often, too, it is desired to extract some of the data in machine-readable form for transmission to other databases. These two related functionalities, archiving and data transport, have been implemented by FileMan developers over the years in a variety of ways.

The task has proved to be difficult because there are several competing needs to be considered. Ideally, there would be, for simplicity's sake, one extraction method, and it would be

Generalized (to handle any File, any data type)

Standard (so that it works on any MUMPS machine, with a variety of media)

Scalable (so that it handles very large volumes of data, as well as small)

Selective (so that it can extract precisely what is needed)

Complete (so that it can purge and restore anything it can extract)

and Efficient (so that, in the worst case, a disk can be emptied faster than it is being filled!)

The first attempt at solving the data-extraction problem was the utility known as "DIFROM", a subset of FileMan routines which still exists in both the DHCP and CHCS systems. DIFROM at least had the virtue of generality; one could use it, ad hoc, to transport the Data Definition and/or the contents of any FileMan File. (As one V.A. physician mordantly noted, "FileMan means never knowing what you're going to DIFROM".) And DIFROM was Standard because what it generated was MUMPS routines; every MUMPS system has to be able to load MUMPS routines. The generated routines would simply be executed on a target system to restore the data and the database definition, thus guaranteeing a certain conceptual completeness. Because it generated MUMPS routines, though, DIFROM was not totally scalable; every "package" it created needed to fit into a certain maximum number of routines. Also, the process was not particularly CPU-efficient at either end. But the main reason DIFROM could not be used well for archiving was that, although it can be made selective with respect to which Field data is to be transported, it didn't allow the selection of particular records to be moved. In Relational terminology, the DIFROM user can select table "columns", but not "rows".

After 1986, separate development efforts in the VA and at SAIC led to many idiosyncratic changes in the two versions of DIFROM, often dictated by differing institutional policies. The most general unilateral enhancement of DIFROM in CHCS FileMan is the ability to ask an initialization routine, just before it installs data, to print out a comparison of the data and data structures it is importing with those that already are in existence on the target system. The resulting printout (often quite lengthy) can assure a diligent System Manager as to what an update to his database will entail. On the VA side, DIFROM was enhanced so that its File-installation packages could be transmitted from site to site via the Electronic Mail system called MailMan, which has been part of the DHCP Kernel of utilities since 1984.

Meanwhile, during the 1986-1990 period, DHCP and CHCS programmers created two new packages, each attempting to implement selective data extraction from FileMan databases in a generalised way. The two approaches were quite different. The CHCS package was called "TapeMan", and the VA version (much of which was actually produced by the Indian Health Service) implemented "FileGrams". TapeMan, as its name implied, was conceptualized as a tape-handling utility, storing data on tape in much the same way that it is stored in FileMan Globals. FileGrams, by contrast, are extracts of data in the form of MailMan messages which are machine-built and machine-readable, but are also "expanded" enough from FileMan's internal Global format so that they are human-readable as well. It is worth mentioning that the CHCS project made an effort to implement FileGrams in a limited way, based on format information that was available from the VA around 1992. One could imagine that this work might aid in data interchange between the two systems at some future time.

In Version 21 of DHCP FileMan, the VA has released a prototype archiving system based on the concept of the FileGram, and also on the longstanding FileMan concept of a Search Template. File Entries are selected by a Search, and put in temporary storage in FileGram form. This work is logged in an Archive Activity File, #1.11. Other Archive Options then allow the system manager to write the temporary record to tape, and to purge the archived entries.

The CHCS TapeMan utility evolved into a somewhat similar set of options. It used Search Templates, too, but the "column" specification was done with Print Templates (including the catch-all "[CAPTIONED]" Template), rather than with FileGrams. The output, non-human-readable, went directly to a tape device. An "ARCHIVE TAPE RECORD" File (number .99) kept track of every run, and this on-line File could be queried to see the history of any Entry's archiving. Other TapeMan modules allowed one not only to purge the Entries one had written to tape, but also to restore later one or all of the Entries from a tape. Note that VA Archiving, by comparison, does not allow data restoration.

Although "TapeMan" can still be found in the "DIR*" namespace of CHCS FileMan, it was never widely used in the DOD hospitals. The larger hospitals had immense amounts of data to purge every month, and TapeMan was not much more efficient than DIFROM in extracting information. One could say, really, that TapeMan met four of the six archiving criteria (Generality, Standardness, Scalability, and Completeness), but was too inefficient, and, given the complex interrelationships between various CHCS Files, was still not selective enough to do what the DOD felt needed to be done.

What kind of archiving system supplanted TapeMan at the large DOD sites since 1990?

A very idiosyncratic one, which foregoes the advantages of Generality, Standardness, and Completeness, in favor of Selectivity, Efficiency, and the massive scale of optical-disk technology.

First of all, the DOD (like the VA?) essentially abandoned the requirement that archived data could, at some later date, be reinstalled automatically into the on-line database. This decision was made in light of the fact that the application Data Dictionary definitions in CHCS were evolving very rapidly; it seemed hopeless to expect that data written off at one moment in time would match structurally the data definitions that might be in effect years later, when retrieval was attempted.

It was decided, rather, to specify a new, "permanent" data structure into which all archived FileMan information would be funnelled. The only kind of retrieval from this (non-FileMan) structure is by ad hoc queries (for individual Entries) or searches (across a whole File), the results of which are simply displayed or printed when needed.

The physical underpinnings of the new CHCS system are impressive. At the large sites, a separate piece of hardware (a VAXstation 3100) serves as a dedicated archiving CPU, in order to minimize the impact of archiving on the production cluster. This archiving processor's function is to put archive-ready data out to a write-once ("WORM") optical disk. The index to data stored on the WORM drive is maintained on a conventional magnetic disk that is also accessed by the archiving CPU. There is an Ethernet link between the archive system and the production system; the archiving process runs on the VAXstation 3100, pulling data directly off production disk drives.

Each field in the CHCS FileMan database is flagged with an Attribute (not found in DHCP FileMan) that tells the archiving system what data should be saved to the WORM drive. General criteria have also been hard-coded, specifying, for example, that inpatient data is ready to be archived 60 days after the record of the admission is complete, while outpatient data is held for 18 months. Thus, archive data is selected in a very complex way, application by application, in a process that navigates intricately from File to File.

A full archive run is scheduled once a month. Site staff can tell the system to scan the production disks only at nonpeak processing times, and, again, the scan is done by the 3100 processor, over the Ethernet. When archive-ready data is found by this scan, it is moved to a "staging area" on the archive system, where it is transformed for "WORM" storage. This transformation includes turning internal Pointer numbers into human-readable names. Data is written to the WORM as VMS files, not as MUMPS globals.

It should be emphasized that the CHCS archiving system is built to run on specific hardware and software, viz., the VAXstation, VMS, the Ethernet, and the WORM drive.

The system is therefore extremely vendor-dependent.

A fuller description of the CHCS archive design can be found in Donna Beckman-Cunningham's article, "Designing an Optical Base Archive System", in MUMPS Computing, Volume 22, Number 3, June, 1992. The four paragraphs above are a precis of that article.

In terms of portability of applications, the alternative archiving techniques presented in this Section are probably not crucially relevant. It should be obvious, though, that if a major CHCS application were ported to a DHCP environment, serious consideration would need to be given as to whether large amounts of Global data could be selectively purged in accordance with the designers' intentions by the construction of new "FileGrams". It should be emphasized that hard-code algorithms, not Computed Expressions, are used in CHCS to trace through various Files to find archive-ready data. Going the other way, therefore, if a large DHCP application were ported to CHCS, new archiving code would have to be written to select any of that application's data for long-term WORM storage.