Paging User Bias Bias 8oX'1 ... 30X3 == 20 X 1 + 50 X 3 =
Weighted
Bias 170/4 = 42 170/4 = 42
The above examples illustrate the following general points about the use
of the bias factors, the delay factor, and the user priority value:
1. The paging and user bias factors are a measure of the relative importance of the bias value. A high user bias allows greater
discrimination via the assigned priority; while a high paging bias makes storage requireaent the primary scheduling parameter. 2. The virtual machine priority
overridden, and is the means
improved perforaance.
value, in the directory, aay be
through which selected users obtain
3. The priority delay factor is the aeasure of the impact that the
paging and user biases are to have. The greater the delay value,
the greater is the maximum delay that can be experienced by a given
user.
4. The interactive bias factor is a tool that enhances coaaand
response to conversational commands that require disk I/O, and that may be partially executed in Q2.
If the paging ·bias factor is nonzero, the net effect of the priority scheme is to discriminate against virtual machines that require large
aaounts of real storage. This discrimination results in a higher leyel of multiprogramming and increased processor utilization; however, it must be traded off against poorer throughput for large storage users.
The distributed scheduler is not biased; the bias factors are as
follows: --- Paging bias factor (DMKSCHPB) = 0 User bias factor (DMKSCHUB) = 1
Priority delay factor (DMKSCHPD) = 0 Interactive bias factor (DKKSCHIB) = 0 Thus, the basic YM/370 scheduler schedules virtual machines FIFO within
user priority.
Figure 19 is a graphic breakdown of the user states, showing the
relationship between interactive and non-interactive states, in-queue
and not-in-queue states, and in-list and not-in-list states. , i 1 In-Queue 1 Bot-in-Queue 1 1------------------------------ I IDispatch I Bo IEligible I Bo 1 1 List IListl List IListl r I 1 In teracti ve 1 1 2 3 4 I 1------------------------------------------ I I Non-Inter- I I 1 active I 5 6 7 8 1 Figure 19. User Dispatching States CP Introduction 1-131

Figure 20 shows the possible user-state changes and the reasons for thea; any changes not described are not possible.
r Status Change I I I I Froa To Reason for status Change I -------------------------------------------------------------------1
1
1
1
1
1
2
2
3 4 4 5
5
5
5
5
6
7
8
2 4 5
1
8
1
5,1
1
1
3
1 4 6
1
8
5
5
5,1
Pagewait, SIO-WAIT, or enabled wait for any busy channel
Enabled wait for interactive terminal read or write
Exceeds in-queue time slice Salle as 1 to 5 except that queue2 is full Wait without active I/O, disabled WAIT or hit ATTN Wait condition cOllplete Wait coapletes, but in-queue time slice exceeded
Another user drops froll queue1 and now there is room Ter.inal I/O completes While user is waiting
Terminal I/O completes, but queue1 is full
Terainal I/O completes while user is active in queue2 User puts up terainal read or write and enters wait Pagewait, SIO-WAIT, or enabled wait for busy channel
Dropped from queue2 due to in-queue tiae-slice end Wait without active I/O, disabled WAIT, or hit ATTW Wait condition coapletes
Rooa is found in queue2
Asynchronous I/O or external interruption or BEGIN
Figure 20. User Status Changes
To control the nuaber of virtual aachines allowed in queue, the
scheduler monitors the paging activity of all virtual aachines and of
the total systea. A decision as to whether or not to aove a potential
virtual aachine froa the eligible to the dispatch list is based upon
whether or not that its projected working set exceeds the systea's
reaaining capacity. Individual virtual machines' working sets are
calculated and projected at queue drop time according to one of the
following foraulas: P=(A+P)/2 If (LP-LA) * (P-A) < 0 -- or -- P=A If (LP-LA) * (P-l) 0 A Actual working set at queue drop tiae LA Last actual working set LP Last projected working set P Current projected working set
1-132 IBK VK/370 Syste. Logic and Proble. Deterlination--Voluae 1