Retrieved [?] times since 06/13/2000
On May 30, 2000, Getronics (formerly Wang Global, formerly Wang Laboratories) announced two new additions to the VS6000 line — the VS6760 and the VS6780. Both use the new Central Processor type 15 (CP15), an adaptation of the CP18 used in the VS18000 series. Available as board upgrades to existing VS5000 and VS6000 systems or as complete new systems, the new processors extend the reach of the VS6000 line into new territory.
| New Model | FAST Rating |
| VS6230 | 1150 |
| VS10100 | 1460 |
| VS7380 | 1500 |
| VS6230T | 2025 |
| VS12650 | 2475 |
| VS6760 | 2850 |
| VS16850 | 3700 |
| VS6780 | 4250 |
| VS18950 | 6300 |
Graphically, the new lineup looks like this:
We see that the VS6760 and VS6780 clearly eclipse their VS6000 predecessors, providing a very natural upgrade path for sites that are straining the limits of their VS6230 or VS6230T.
Prudence requires me to point out that although the nominal FAST ratings exceed those of several large VS models, Getronics is recommending the new models only for upgrade from VS5000 and VS6000 models, not for replacement of large VS models. There are good reasons for that:
First, large VS models (7000 and higher) have wide bus widths of at least 64 bits, fast I/O bus speeds, and most have high upper memory expansion limits. Systems that actually use significant fractions of those capacities are not suitable for the VS6000 architecture, with its slower, 32-bit bus. Large VS systems handle 14 devices per SCSI IOC; small systems handle 7 per IOC.
Second, the impact of I/O on processor performance is greater in the new models than in their larger siblings. This has to do with the fact that I/O must be coordinated with cache in a process called "cache invalidations," in which the cache memory is updated to reflect memory locations altered by I/O operations. The more I/O there is, the more time is spent processing cache invalidations, which are done differently in the VS5000/6000 architecture than in the large machines.
That said, we out here in the field know that there are large VS systems whose workloads have diminished in recent years as applications and users have moved to other platforms. It remains possible, then, that some large VS systems now carrying only the workload suitable for a smaller system may be candidates for replacement by any suitable member of the VS6000 line, particularly where the only alternative is retirement of the large VS. VS6000 costs are more in line with common servers found in most computer facilities, and therefore should have survivability in IT environments where large systems are under constant assault.
I have had one of the new processor boards running in my VS6000 for about a month now. It was consigned to me by Getronics for evaluation while beta tests were being concluded in the field, in the weeks leading up to the announcement.
My first impression upon IPLing the new board (keyed to be a VS6780) was that it was very snappy. I say that from the standpoint of one who has become accustomed to the speed of a VS6230T.
Processor speed manifests itself readily and graphically in EDITOR when we do PF4, 5, 6 and 7 operations to move around in the source. Response of the VS6780 is noticeably faster than the VS6230T, so fast that scrolling is possible by holding down the PF6 or 7 key. Renumbering of modest source programs is virtually instantaneous. Compiles and links are faster, too, even though they are also affected by disk throughput, which remains largely the same regardless of processor speed. Starting GENEDIT and watching the progress arrow move down the screen is another subjective measure that reflects processor speed, and the arrow moves faster with this processor driving it. Subjective measures, though, are not as convincing as benchmarks, and I have been doing a lot of work with VS benchmarks in the last year or two.
The raw speed of the processor doing a tight counting loop, the basis for my measurement and classification of VS processors, is precisely 20% better than the VS12650 and about 3-1/3 times the speed of the VS6230T. That's impressive, because the VS12650 is probably the fastest VS processor for its official FAST rating ever produced. The FAST rating, though, is complex and includes performance meausures far more comprehensive than my raw processor speed measures. The official rating of the new models reflects their use of the same CPU chip that is used in the VS18950, adapted to live on a VS6000 CPU board. That CPU design includes advanced features such as pipelining and predictive branching as well as larger and faster cache memory than earlier designs. As a result, it gets more work done per clock cycle than older chips, whereas my tight-loop tests tend to reflect the CPU clock rate instead of the real work of executing complex instructions. The marriage of the large-VS CP18 to the VS6000 architecture has produced two models with speeds somewhere between the VS12650 and the VS18950, but without the very robust I/O capacities of the large VS models.
Since I don't have any real user population or real batch programs to run, to get some idea of how the new models handle real work I had to resort to using CREATE and COPY to generate file-based activity. While these activities are significantly disk-bound, there is sufficient processor activity in the application and in XDMS to show significant improvement if the CPU is upgraded, as most of us have experienced through the years.
A difficulty in evaluating a processor upgrade is that jobs are likely to finish sooner and use less CPU time. Jobs that are more CPU intensive will show the improvement more in shorter run times than CPU usage, while jobs that are more I/O bound will show the improvement mostly in reduced CPU usage. Thus, one job might complete in half the time while using about the same percentage of CPU, while another might complete in about the same time while using only half the CPU it did before. Reconciling the differences in the two measures can be done arithmetically if the before and after run times and CPU usages are known.
Each effect -- on run time and on CPU usage -- is of interest in its own context. Reduction of baseline CPU usage is of interest in interactive environments, and reduction of job run time is of interest in batch environments. There is no way to quantify an overall measure that will allow complete prediction of results across all environments, but our general experience shows that a faster processor will reduce interactive CPU usage and shorten batch run times. In examining a new processor it is interesting to see the two measures individually as well as combined into a single, normalized performance number.
My first test was to CREATE a consecutive, fixed, not compressed file of 350,000 100-byte records. That comes to 17,500 VS disk blocks, or 34 MB of raw disk data. The results were:
VS6230T (CP11): 62.09 elapsed secs, 22.84 CPU secs VS6780 (CP15): 44.69 15.68
Records per second improved from 5,637 to 7,832, an increase in speed of almost 39%, while the CPU seconds dropped by 31%. The CPU usage (CPU secs / elapsed secs) was very close to the same. If we adjust the speed of the faster machine upward to compensate for stretching out the elapsed time to normalize it to that of the slower machine, we get 1.93 as the effective speed of the faster machine over the slower one -- almost twice as fast, which is very close to the relationship between the two official FAST ratings of 2025 and 4250.
The next test was to CREATE 775,000 consecutive, variable, compressed records:
VS6230T (CP11): 57.15 elapsed secs, 48.96 CPU secs VS6780 (CP15): 41.11 32.51
Records per second improved from 13,561 to 18,852, 39% faster, while CPU time dropped by almost 34%. The normalized speed of the VS6780 is 1.93 times that of the VS6230T.
The most extreme increase in my tests was seen when creating an Indexed-Plus file of 50,000 100-byte records:
VS6230T (CP11): 55.88 elapsed secs, 16.18 CPU secs VS6780 (CP15): 29.12 7.76
In this case records per second improved from 895 to 1,717, 91% faster, while CPU time dropped by 52%. The normalized speed of the VS6780 here is 3.68 times that of the VS6230T.
The lowest figures were turned in by the large-record consecutive file and simple Indexed file CREATEs.
The results of measuring COPY runs were not very revealing, probably because COPY does very little processing and is almost completely disk bound.
I use an evolving set of COBOL timing tests to compare COBOL verb execution times between various VS models. These tests reveal the power brought to bear in the new CPU chip. GOTO runs 3.3 times as fast on the VS6780 as it does on the VS6230T. PERFORM clocks at 2.3. MOVEs range from 1.1 to 2.1, with all MOVEs to edited numerics clocking consistently at 2.0-2.1. Most fullword and halfword operations (MOVE, ADD, SUBTRACT of BINARY PIC(4) and PIC(8)) show modest improvement of 1.2-1.7, but MULTIPLY and DIVIDE clock at a respectable 2.0-2.4 times the VS6230T speed. I would expect real-world COBOL programs to run 1.5 to 2 times faster or use 1/3 to 1/2 the CPU seconds on the VS6780 than on the VS6230T, or some combination thereof. Improvement over any models slower than the VS6230T should be breathtaking, with the possible exception of VS7000-10000 systems doing large volumes of I/O and older systems with very high task counts (the task limit of 384 for the high-end VS6000s has not been increased that I know of).
The new models ship with 128 MB of main memory. The old limit for the VS6230 & Turbo was 64 MB. I believe that was raised sometime in the last year or two, but I missed the memo. In any case, 128 MB is the new standard for the high-end VS6000s.
I can't resist pointing out that the very first VS, the VS80, released in 1977 or 78, was commonly shipped with 256 KB (yes, that's k-i-l-o-bytes), and had a maximum capacity of 512 KB -- half a megabyte. The new VS6780 is about 43 times faster, handles 256 times the memory, and manages far higher performance peripherals than the VS80.
Ever since the introduction of the VS6230T and VS16000 models, Getronics has been using a clever trick to minimize the complications of otherwise simple CPU board upgrades: CP type spoofing. That means that all the newer CP types report themselves to the software as the older CP type they are replacing. In the case of the CP15, like the the VS6230's CP11 before it, the software sees it as a CP9 (VS5000). That means that no software will object to the upgrade on the basis of CP type. The same technique is used in the VS16000 CP16 and VS18000 CP18 to mimic the CP8, 10, 12 or 14.
The upgrade itself consists of preparing the VS by installing new CP microcode that recognizes both the older CP9 or CP11 CPU and the new CP15 CPU, verifying that all elements of the system are working, then swapping out the old CPU board for the new one. After IPL the VS will run at the new speed, which is determined by the VS6000 license key, a piece of hardware remarkably like a conventional doorlock key that installs in the CPU board.
The VS6760/6780 CPU can be installed in any of the VS5000/6000 cabinets. For the benefit of those unfamiliar with the VS5000/6000 cabinets, they are narrow or wide towers not unlike PC server towers. The narrow cabinet offers five slots while the wide cabinet offers ten. Each can be expanded with another five or ten slots. All cost factors of operating the VS5000/6000 systems are significantly lower than those of the large VS cabinets -- capital cost, software licensing, maintenance, electric power, heat load and footprint. At the same time, reliability is higher than that of the older, larger models.
You may view an approximately scale drawing of the various VS cabinets in the VS Family Portrait. The new VS6700 models, if ordered as complete new systems, are delivered in the 10-slot cabinet shown in the drawing as "6010" except that the current cabinets have a more modern style than that shown.
You may also view the Getronics page on the VS6700 series.
When ordered as a complete system, the VS6760 and VS6780 are packaged in the 10-slot Century Server cabinet, as was the VS6230 before them. Unlike the 5-slot cabinet, the 10-slot has a number of internal drive bays that can be wired to the standard RCU SCSI chain or to the internal segment of an optional caching SCSI IOC. This allows the larger cabinet to contain substantial disk storage without resort to external drive enclosures.
As we have come to expect, the engineers who design the VS have come through once again. The new VS6760 and VS6780 clearly extend the VS6000 line into territory once exclusively held by the large VS models. In doing so they provide an upgrade path for customers at or approaching the CPU limit of the VS6230T, for those who wish to jump directly from the VS6230 to something more powerful than the VS6230T, and in general for any customers with older, slower VS models who wish to modernize and reap the benefits of very cost-effective VS operation with substantial processing capacity. In some cases, the new models may also afford a survival path for some large VS systems endangered by internal campaigns to cut operating costs, by reducing those costs more to the level of conventional servers.