Amy, on server farms, RAS, etc.
I think the PC Server farm is
great for small to mid-size businesses (who may not have the capital nor desire to
build a large infrastructure), and possibly as temporary expansion/backups for
larger firms, especially during random peak volumes.
PC server farms may not be the right term. Xeon and then IA64 based server farms would drop the "PC" part, which I don't think is even in the description today. I think it's just server farms. A nit, I know. A recent article described server farms used by some large companies, and they mentioned IBM S390 as the central web server. One of these sites could cost the owner millions of dollars per significant outage (like what happened to ebay). AS far as outsourcing server farms goes, I thing ANY company, regardless of size, could be a candidate to do this. Outsourcing is THE trend of the 90's, 00's. Which leads to the next topic, RAS:
Tony, when you say IBM is the leader, what specifically were you referring to
(i.e. for what market segment/application?) I'm interested in learning your
thoughts.
What I meant in the post was that IBM "invented" RAS, or at least the acronym, for all market segments. They put RAS right up there with performance and price in importance as a selling factor for their products used for mission critical applications. I would say they make RAS as important as performance and price in their medium to large system products. Most of all the rest of the hype we hear and read here and elsewhere is on the other two, and RAS is almost never mentioned. IBM's commercials on TV have been RAS oriented for the last several years. Point is, what good is something that runs at xxx megahertz or yyy specints if it goes down even once in a year. One outage a year, even for their most complex system, would be considered abysmal for IBM.
A few key RAS features in IBM S390. e.g.:
- Concurrent hardware and microcode upgrades (no power down, no touch) including CPUs.
- Concurrent hardware and microcode repair (no power down, no touch).
- No single point of failure (NSPF), hardware or software. Pull the plug on that sucker and it keeps going. Probably as much as 10% of overall logic hardware devoted to ECC of all kinds, retry, data integrity testing, etc., to support NSPF.
I know I sound like a broken record on this stuff.
Tony |
