PB,
Tony - Somewhat OT
Can you take a look at this, at let me know if you have a take on the whole
argument, i.e., both Xeon vs. K7 and NT vs. UNIX?
Message 9386783;
What's in the highlighted URL:
Tenchusatsu said: "... Believe it or not, customers who buy servers care more about
RAS (reliability, availability, and serviceability) than
performance ..."
Thomas M said: "Stop spouting the company line. You don't want to sound like a
corporate schil, do you?
If this statement were true, why would so many IT pros be
switching to the legendarily unreliable Windows NT from Unix?"
If that's a company line, it's IBM's, still the biggest and best diversified IT company in the world. Also, IBM came out with the acronym RAS first, and it has caught on at Sun, Intel and all other companies that know what's important. Performance, price and RAS are the three overwhelming spec categories for major IT systems. Certainly, as you move up the food chain, RAS gets more and more important. At the top, entire company operations are dependent on their servers. An outage can cripple the whole thing and essentially stop revenues completely. Airline reservation systems come to mind first. No central server, no new reservations. I have heard of contracts where there are penalties built in, whereby as much as $100,000 is charged to the server vendor in the event of a complete outage. There is nothing that will get you thrown out of a customer's IT shop faster than poor reliability. So, yes, what Ten said is true. Bad reliability will outweigh top performance most every time where continuous operation is imperative. I mentioned airlines, hell, consider even Internet commerce, which we've read could be in the trillions of $$ within 10 years. A server outage can just as well be equivalent to company down time, and lost revenue there also. Schwab and Etrade have had outages recently that got them a ton of bad publicity. An example of an environment that could tolerate outages is universities.
Certainly, there are server companies with reputations for excellent reliability. IBM, no question. Sun is way up there also. Both have the benefit of running their hardware on mature, stable operating systems. On the hardware side, for RAS, I don't know too much about what Sun has done, but I know it's significant. I know a lot more about what IBM has done for RAS, and, it, coupled with their already best of class reliability chips, has them able to claim 30 years as MTBF for their newest, biggest servers. I think this is a huge selling point for them.
As far as Xeon vs. K7, Intel has a huge lead because of several reasons:
1. Their demonstrated, year after year, high quality, high yield process engineering and manufacturing prowess. As I've said before here, the story about the cause and effect relationship between yield and reliability is being spread more and more by IBM and others. Intel has the consistently high yields, AMD doesn't. High yields beget high reliability. I think this is a show stopper right here, i.e., until AMD can tell a consistently good yield story for their high end chips, the smart vendor companies won't look at them. That doesn't mean that some others won't.
2. Infrastructure. Quality, reliable motherboards and chipsets with proven SMP capability. This goes back to the Pentium Pro days, making it, what three years now of demonstrated SMP capability. AMD is brand new to this.
3. More infrastructure. The server vendor has to develop a lot more hardware to "flesh out" the server. Some of these you could say are relatively independent of the CPU chip. However, before investing all the development time and money into these, he has to have faith that the most important part, the CPU, warrants all the work. Note these are from a previous post of mine.
Motherboard and node development based on the µP to support:
• Up to 4 GB or more memory
• Failover
• Clustering/Fabric interconnect
• Hot swappable hard drives, including RAID
• Hot swappable and redundant power supplies and fans
• JTAG
• ECC Memory
• Remote Management/Landesk server manager
- Power Cycle
- Boot
- Diagnostics
- Configuration/reconfiguration
- Software distribution
- Inventory
- Security - Chassis intrusion, password protection
- Event logging
- SNMP trap
From the software side, the rest of the question is why invest in all this new hardware if NT is not reliable. First, don't Xeons run Unix, besides NT? Would the K7 also? Second, NT is "supposed to be getting better all the time, isn't it?" Third, software crashes don't result in lost or corrupted data as much as hardware crashes. For whatever reasons, it seems like IT customers tend to overlook RAS more on the software side more than they do on the hardware side, or they don't know as much about it.
Tony
|
