isham research
For those who read this page between its initial upload on 15 March and the announcement itself, there will have been few surprises. In general, IBM has done very well to bring out systems around 35% more powerful than their predecessors in only nine months - and many aspects of the software story are very strong. However, it's an industry analyst's function in life to pick nits ...
The Announcement Letter is one of the largest of all time (it's about eighth longest in the all-time rankings) and contains a fair amount of redundant or uninteresting information, as well as the odd gem. Does it really interest us if the cooling is so high-tech - or do we just assume it will be adequate? Is a company like IBM so likely to bring out products with design weaknesses that it's necessary to eat up tens of pages of announcement material with reassurances to users?
Personally, I'd rather see the nitty-gritty. For me, something like: "The G6 turbo 12-way is about half as powerful again as the G5 10-way" tells me as much as I need to know about performance. But "1614 MIPS" implies a level of precision at least two orders of magnitude better than benchmarking technology has ever been. 0.01% accuracy? From a benchmark?
There are some subtleties. There is, despite all to the contrary, a minor differentiation between G5 and G6 in the Capacity Upgrade on Demand area. Smaller G5s use 6 PU (Processing Unit) and 8 PU MCMs - CUoD is not possible over these boundaries, but is possible throughout the G6 range. IBM's 12-way MP ratios may be the best in their industry, but they're beaten by every Amdahl 12-way ever shipped - 8.03 plays 8.12, 8.58 and 8.81. Some of the larger storage sizes are not available immediately, and rumours suggest that Samsung (IBM's storage supplier) may have difficulty with high volumes of Zx7 storage. We will see - but if users are sold 6-way turbos and then offered 7-way non-turbos for earlier delivery ...
FICON is listed for availability on 31 August 1999, even though it's already shipping for the G5. It's also missing from the EMIF statement. Does IBM have any issues here?
IC Channels now manage only 700MB/sec on a G5 - they were touted for up to 6GB/sec in the original G5 announcement (198-115).
VM and VSE now have their own version of IMLC - but the announcement doesn't say whether the PSLC C and PSLC D price breaks will be implemented.
And is Sun getting on IBM's goat? The veneer has cracked in a couple of places:
'Mainframe "wannabee" servers may claim effective utilization for only a specific tuned workload in a sheltered benchmarking environment.'
"OS/390 is not a vague promise of some future enterprise computing greatness, such as that offered by alternate platforms who are 'Mainframe-like.'"