Invisible <voi### [at] devnull> wrote:

> How about drum memory? Or the various delay lines?
>

Don't know, was before my time.

> As in, whole memory chunks on a single IC? Or just two latches per IC?
> Or...?

I don't know very much about the actual electronics. You will have to go to
google or wikipedia for that info.  I remember the 3081 had what were called
TCMs.  Fully encased, water cooled modules with about a thousand pins on the
back.  The CEs had lots of fun replacing them without bending a pin in the
process.

> It still somewhat blows my mind that you could do anything useful with
> so little memory. Presumably for processing large datasets, most of the
> data at any one time would be in secondary storage?

Most really large files were on tape, you just read the data serially.
For highly accessed files, or files that needed random access were kept on disk.

> Heh, sounds line fun. According to the account I just read, you actually
> put one statement per punch card. Is that right? I always assumed that
> each card just held X characters of data, and you type until the card is
> full, then move to the next one. I didn't realise there was an actual
> "significance" to card bounderies. Or is that just for the benefit of
> the humans?

The file system for IBM mainframes are record oriented, not stream.
Files are created with a defined record format and record length.
There is no end of record indicator such as CR or LF as in PC files.
Since cards were 80 columns long, many source files today are also 80 byte
records.  The norm was for program source files was column 1 to 72 was data
and column 73 to 80 was sequence numbers. For most COBOL programs the sequence
numbers were in cols 1 to 6.

> What sort of access speeds do you get for reading or writing punched
> card or tape?

As I recall about 60 characters per second.  The tape was paper, was 8 holes
wide, and easily broken or scrunched.

> As in, discrete transistors on PCBs?
>

Again, I don't know much about the electronic packaging.

> Any ideas on typical memory capacity / clock speed? Data path widths?

On some of the older machines I worked with:
  360/50 128KB
  370/158 1MB
  370/168 3MB
  3033U 4MB
  3081Q 8MB
On the latest machine:
  z/9 BC S03 - 16GB

> Nice consistent naming scheme. Heh. :-/

You can always depend on IBM to be consistent.

> Now if I could figure out what a typical MIPS rating for a normal PC
> today is, I'd have something dissimilar to compare it to. :-}

It is rather difficult to compare since they have such different architectures.
I doubt that the internal processor speed is that much different than current
PCs, but the i/o bandwidth is much higher in the mainframe.
For example our two machines, which are on the small side for mainframes, are
connected to a disk storage unit containing 15TB of data.  There are 4 high
speed fiber optic channels from each processor to this unit. We can easily
sustain i/o rates of over 10000 i/o operations per second, with peaks in the
20000 to 40000 range.  It is very rare for any single job to saturate the cpu.
By running many jobs concurrently we make full use of the machine.  If I could
port POVRAY to our mainframe, I'm sure it would be slower than most current PCs.
 But cpu intensive work is not what the mainframe does best.

Isaac

Post a reply to this message