Queue 3
eMS BLIP Facility
Q2 delay factor
is calculated dynamically based on configuration and load, and is the average
elapsed time required by a virtual machine to receive an amount of processor
time equal to one Q2 time slice.
For Q 1 virtual machines, the scaled bias is divided by 8 (since the Q 1 processor
usage time slice is 1/8th the Q2 time slice). The difference between scheduling a
virtual machine in Q1 instead of Q2 is that it receives 1/8th the amount of
processor,8 times as often.Operating constantly in either queue, a virtual machine
should receive the same amount of processor resources over an extended period of
time. The only preference given Q 1 virtual machines is when they are being moved
from the eligible list to the dispatch list. They are moved ahead of Q2 virtual
machines with the same or even slightly better deadline priorities.
Q3 is an extension of Q2 scheduling. It helps to distinguish between
non-interactive virtual machines and those that are frequently switching back and
forth between Q2 and Q 1. Virtual machines that have cycled through at least eight
consecutive Q2 processor time slices without a Q 1 interaction are labeled Q3. Q3
virtual machines are kept in the same lists (or queues) as Q2 virtual machines and
for most purposes are treated identically. The differences between Q2 and Q3 vir
tual machines are reflected in their deadline priority calculations and the amounts
of such processor time they are allowed in queue. Q3 virtual machines are allowed
eight consecutive Q2 processor time slices before they are dropped from queue.
Because of the eight-fold increase in processor time allowed each time in queue,
the scaled bias is multiplied by eight before adding to the current time-of-day to
form the deadline priority. Q3 virtual machines should receive eight times as much
processor time each time in queue as Q2 virtual machines, but only 1/8th as often.
To reiterate the Ql/Q2 statement, which is also true for Q2/Q3:Operating con
stantly in any queue, a virtual machine should receive the same amount of process
or resources over an extended period of elapsed time. This does not necessarily
mean that a virtual machine performs the same when operating in Q3 mode as
when operating in standard Q2 mode. An amount of overhead (roughly propor
tional to the small number of resident pages) is used for each virtual machine when
it drops from queue. When operating in Q3 mode, a virtual machine may perform
much better than in normal Q2 mode because it is undergoing fewer queue drops.
For some very large virtual storage programs, the total processor resources used
has been cut in half by operating in Q3 mode as compared to standard Q2 mode.
TheCMS BLIP facility causes CMS to perform a write operation to the terminal
after every 2 seconds of virtual processor use. This feature effectively cancels
Queue 3 use for normal, connectedCMS virtual machines, regardless of what types
of programs they are running. TheCMS BLIP facility can be turned off with the CMS SET BLIP OFF command or it can be disabled with the CP SET TIMER OFF command.
Using Processor Resources 17
eMS BLIP Facility
Q2 delay factor
is calculated dynamically based on configuration and load, and is the average
elapsed time required by a virtual machine to receive an amount of processor
time equal to one Q2 time slice.
For Q 1 virtual machines, the scaled bias is divided by 8 (since the Q 1 processor
usage time slice is 1/8th the Q2 time slice). The difference between scheduling a
virtual machine in Q1 instead of Q2 is that it receives 1/8th the amount of
processor,8 times as often.
should receive the same amount of processor resources over an extended period of
time. The only preference given Q 1 virtual machines is when they are being moved
from the eligible list to the dispatch list. They are moved ahead of Q2 virtual
machines with the same or even slightly better deadline priorities.
Q3 is an extension of Q2 scheduling. It helps to distinguish between
non-interactive virtual machines and those that are frequently switching back and
forth between Q2 and Q 1. Virtual machines that have cycled through at least eight
consecutive Q2 processor time slices without a Q 1 interaction are labeled Q3. Q3
virtual machines are kept in the same lists (or queues) as Q2 virtual machines and
for most purposes are treated identically. The differences between Q2 and Q3 vir
tual machines are reflected in their deadline priority calculations and the amounts
of such processor time they are allowed in queue. Q3 virtual machines are allowed
eight consecutive Q2 processor time slices before they are dropped from queue.
Because of the eight-fold increase in processor time allowed each time in queue,
the scaled bias is multiplied by eight before adding to the current time-of-day to
form the deadline priority. Q3 virtual machines should receive eight times as much
processor time each time in queue as Q2 virtual machines, but only 1/8th as often.
To reiterate the Ql/Q2 statement, which is also true for Q2/Q3:
stantly in any queue, a virtual machine should receive the same amount of process
or resources over an extended period of elapsed time. This does not necessarily
mean that a virtual machine performs the same when operating in Q3 mode as
when operating in standard Q2 mode. An amount of overhead (roughly propor
tional to the small number of resident pages) is used for each virtual machine when
it drops from queue. When operating in Q3 mode, a virtual machine may perform
much better than in normal Q2 mode because it is undergoing fewer queue drops.
For some very large virtual storage programs, the total processor resources used
has been cut in half by operating in Q3 mode as compared to standard Q2 mode.
The
after every 2 seconds of virtual processor use. This feature effectively cancels
Queue 3 use for normal, connected
of programs they are running. The
Using Processor Resources 17