[Gllug] Comparison

[John Hearns]

David Damerell wrote:

>On Saturday, 27 Oct 2001, Andrew Richardson wrote:
>
>>Linux User magazine has an article this month about a company using Linux on 
>>IBM mainframe.  Cost wise , I recall it estimating that benefits were to be 
>>had if you run more that 20 or so machines.
>>
>
>That's got to be too low. 20 nice machines (but headless, with crappy
>graphics cards) will set you back L20,000; but S/390s start at about
>half a million, before you touch ongoing maintenance fees. I suspect
>your L480,000 change will pay someone to look after 20 beige boxes.
>

What you say is of course totally true -
but more biased towards the case where you want lots of CPU, say to perform
simulations or Monte Carlo calculcations. That is of course the best
'bangs per buck'.


However, the case here is more of a telecomms firm, which I expect will
be running billing programs and databases.
In that world, and others like airline reservations etc,,
maybe reliability is much more important than CPU power
- as you said earlier, in a big farm
of PC class boxes, you can expect one to break every so often. 
Mainframes have
hardware in them to monitor and e.g. vote out malfunctioning CPUs.

Mainframes also offer huge I/O bandwidth for transactions - think of the big
storage arrays, and things like snapshot backups. IIRC, channel 
connections to
peripherals on IBMs have their own CPUs - so taking load away from the main
machine.


Also there are advantages in running Virtual Machines - you can run up test
servers easily, and can quickly re-IPL (IBM speak for reboot) systems.
And you can do things with allocating storage pools - sorry to put on my
old VM/CMS hat here, but I remember being able to ask for large temp scratch
disks on demand on an IBM mainframe, for use during analysis jobs.
And encountering HSM for the first time - if you didn't log in for a few 
days,
your home directory was migrated off to a tape silo. So when you logged in,
there was s short lag while it was restored to main disk.

Of course, big storage arrays work with Unix machines these days too.
Maybe running off topic, but think of virtual storage arrays - where you put
all the storage for your mainframe in one big storage pool. You can 
'snapshot'
this and do what amounts to an instant backup (ie. you backup the copy) so
you don't have massive downtime on your databases. Etc.
Im sure Simon Trimmer can tell us much more about this sort of thing.



>


-- 
Gllug mailing list  -  Gllug at linux.co.uk
http://list.ftech.net/mailman/listinfo/gllug