IBM System/370 Manual page 172

Performance improvement occurs (1) as a result of overlapping
the rotational positioning time of disk devices and (2) because
more I/O requests can be initiated in a given period of time,
since the channel is free more often. When many disk devices
are active concurrently on a block multiplexer channel, there
is more potential for such overlap.
2. Direct access devices with RPS should be placed on separate
channels from I/O devices without RPS. Alternate possibilities
are as follows:
a. If i t is necessary to place non-RPS devices on the same
block multiplexer channel with RPS devices. give first
choice to non-RPS devices with a buffered control unit,
such as the 2540 Card Read Punch and the 1403 Printer.
These devices disconnect from a block multiplexer channel
during the relatively long mechanical portion of their
cycle, thereby freeing the channel for other operations.
b. Tape units should not be placed on a block multiplexer
channel with RPS devices unless absolutely necessary!, because
channel disconnection does not occur during any of their
channel operations. If this is not possible. try to plan
job scheduling to avoid having jobs using tape units and
jobs using RPS support active on a block multiplexer channel
at the same time. If this is not feasible. try to assign
very lo~activity data sets to these tape units.
A device without channel disconnect capability can monopolize
the block multiplexer channel for relatively long periods
of time, thereby preventing (1) the initiation of other
I/O operations on the channel and (2) the reconnection and
completion of disk RPS channel programs already in operation
on the channel. For example. a direct access device without
RPS retains use of the channel during its search operations
as well as during its reads and writes. If the device is
a 2314 and block size is half a track, the channel is busy
for 25 ms on the average (12.5 ms average rotational delay
plus 12.5 ms read/write) for each I/O operation started
for the non-RPS 2314 facility. Even if the block size used
is relatively small. the channel can still
be
monopolized
by the non-RPS device if there is high activity on the
device.
3. The 2305 facility normally should not be placed on a block
multiplexer channel with other devices.
Exclusive use of a channel insures optimum performance of the
2305 facility as a system residence device.
The following should be noted in regard to specification of priority
and ordered-seek I/O request queuing options for RPS devices at OS
system generation. The priority queuing option insures priority I/O
request initiation for the device, but because of first-corne, first~
served handling of I/O operations on the block multiplexer channel,
this option does not insure that priority device channel programs will
complete sooner than other RPS channel programs that were started later
on the channel. However, the objective of specifying the ordered-seek
queuing option (minimization of arm movement on a disk drive) can still
be achieved when RPS is used.
90