* Posts by Abe Kornelis

2 publicly visible posts • joined 30 Mar 2011

Exploiting the mainframe for new workload requirements

Abe Kornelis

Please define "Weak CPUs"

It's easy to state mainframes have weak CPUs. But what exactly do you mean when you say that? Modern mainframe hardware supports 96 CPUs, running at 5.2 GHz. Moreover, the mainframe uses an extreme CISC-architecture, which means that compilers have a wide range of instructions to choose from. This is freedom of choice that enables optimization to a high degree.

Finally, the processors support out-of-order execution. If you know how it works, you can code for efficiency and speed by exploiting its specific features. And guess what - compilers are pretty good at doing just that, and constantly being improved, too.

Of course, it takes quite a bit of storage to support this computing power. You need to load a mainframe to at least 70% to make it run really efficient. So you can have up to 3 TB of internal storage on a box.

So, if you think this is a weak processor, then you must know of a much stronger one. I have not yet heard of such a strong processor, but I am sure you'll be willing to inform me about the specifics.

Thanks in advance,

Abe Kornelis.

==========

Abe Kornelis

Failover considerations

The article sums up some of the major issues involved in deciding on platforming. One issue that is regularly overlooked is the fail-over requirement. On distributed, this usually means acquisition of a redundant set of hardware, including licenses for any software involved. For mainframes, the hardware need to be duplicated as well, but as long as the workload is on the primary machine, no software licenses come into play. These move with the workload from the primary to the secondary machine, should there be a failure.

The duplicate software licenses constiture a major cost factor for distributed solutions. Therefore it is wise to include them in the decision on platforming at the earliest possible stage.

Abe Kornelis.

==========