[Info-vax] Assembly languages

[Johnny Billquist]

On 2022-04-12 10:44, gah4 wrote:
> On Monday, April 11, 2022 at 8:00:45 AM UTC-7, Dennis Boone wrote:
> 
>>> That is obviously incorrect. Every disk used today is divided into
>>> blocks. So that is what a file on a disk is, in the end. And that is not
>>> the same as a string of bytes. The string of bytes abstraction is
>>> implemented on top of this.
> 
>> Er, not a very convincing argument. In that vein, every magnetic media
>> disk used today is a string of flux transitions. The bytes and blocks
>> abstraction are implemented on top of this.
> 
> Blocks are real on floppy disks, even today.  (Well, they might not be
> so easy to find.)  I suspect, though, that on the usual hard disk these
> days, that blocks might be mostly virtual.

No. They are still blocks on the spinning rust. SSD is a different 
story. But that's all hidden inside the electronics of the drive. The 
exposed interface to the rest of the machine is still talking blocks.

> Traditionally, the important part of disk I/O is the write splice, where
> one starts and stops writing a block. The controller/drive has to
> start writing just after the appropriate block header, and stop within
> the gap after the block. But as internal disk buffers got bigger, it is
> now possible for a whole disk track to go into the buffer.  Though
> last I knew, disks knew how to start reading in the middle of a track,
> such that it could supply blocks to the system before reading the
> whole track.

Of course. And while disks might have large caches, they still operate 
on blocks, and will not rewrite blocks just because they happen to be on 
the same track as some other block which is actually supposed to be 
written out. And they will not read the whole track in just because they 
write a block. That would be pretty bad for performance. But the disk do 
try to guess what you might want to read, and when idle, it will such 
more data in. Same track is a good start.

> Some earlier drives are likely still running, though.

All kind of drives are still running.

> Note that the rotational speed of disks hasn't changed much over
> about 50 years, and rotational latency is a big part of disk reading.

It has, but not as much as most other parts of computers. And seek times 
have improved more than latency, but still not much compared to anything 
else.

An RP06 from around 1975 or so spun at 3600 RPM.
Nowadays, the most crazy ones are at 18000 RPM. So that's about a x5 
increase in close to 50 years. Rotational latency is definitely a part 
of the read times, but seek times used to be the dominant part.

On the RP06, average latency is 8.33ms, while average seek time is 
28.5ms. So more than 2/3 of the time is seek time.

On modern drives spinning at slow speeds, latency is the majority of the 
time, while on the fast spinning disks, seek and latency are about 
equal. Which is why faster spinning disks really gave a boost in 
performance.

With SSD you get rid of both, though. :-)

   Johnny