Re: Will Rich Marcello either come clean or do some proper marketingand
From: Karsten Nyblad (nospam_at_nospam.nospam)
Date: 11/14/05
- Next message: Dave Froble: "Re: FW: OT: Microsoft drop more Itanic support"
- Previous message: Peter 'EPLAN' LANGSTOEGER: "Re: DCPS and form feeds with /copies=n"
- In reply to: Bill Todd: "Re: Will Rich Marcello either come clean or do some proper marketingand"
- Next in thread: Bill Todd: "Re: Will Rich Marcello either come clean or do some proper marketingand"
- Reply: Bill Todd: "Re: Will Rich Marcello either come clean or do some proper marketingand"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Mon, 14 Nov 2005 18:00:59 +0100
Bill Todd wrote:
> Karsten Nyblad wrote:
>> Even software comparison may not be enough.
>
>
> Yes, they are: they are just much slower than hardware comparisons.
>
> The situation you describe is solved by conventional transactional
> mechanisms - it has nothing to do with hardware failure.
>
> - bill
You need software comparisons when ever a process locks a resource or
chances the state of some system wide data. Basicly a process on one
computer trying to lock a resource may not continue before the sibling
processes on the other computers try to lock the same resource, and then
all the sibling processes must get the same answer on the request to
lock the resource. This includes getting resources like virtual and
physical memory. If a process on one of the computers wants to page in
a page of virtual memory, then it should only be allowed to continue
when the sibling processes on the other computers want to page in the
same page. Why? Because there is a small chance that paging a page
will cause something to fail because of lack of virtual or physical memory.
An this is not a conventional transaction problem. In the conventional
transaction world two computer systems are not expected to generate the
same output---at least not expected to always generate the same output.
But I think that we basicly agree that doing what Tandem systems are
doing will be very difficult to do without hardware support.
- Next message: Dave Froble: "Re: FW: OT: Microsoft drop more Itanic support"
- Previous message: Peter 'EPLAN' LANGSTOEGER: "Re: DCPS and form feeds with /copies=n"
- In reply to: Bill Todd: "Re: Will Rich Marcello either come clean or do some proper marketingand"
- Next in thread: Bill Todd: "Re: Will Rich Marcello either come clean or do some proper marketingand"
- Reply: Bill Todd: "Re: Will Rich Marcello either come clean or do some proper marketingand"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|