27.12.2012 Views

z/VM: CP Commands and Utilities Reference - z/VM - IBM

z/VM: CP Commands and Utilities Reference - z/VM - IBM

z/VM: CP Commands and Utilities Reference - z/VM - IBM

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

SET SRM<br />

1392 z/<strong>VM</strong>: <strong>CP</strong> <strong>Comm<strong>and</strong>s</strong> <strong>and</strong> <strong>Utilities</strong> <strong>Reference</strong><br />

In each case, the STORBUF values must be decimal numbers from 0 through<br />

999, <strong>and</strong> aaa ≥ bbb ≥ ccc. Any value over 100 is an overcommitment of<br />

storage.<br />

Q0 users are left out of the above discussion to keep it simple. In fact, Q0<br />

users are allowed to crowd into storage without limit. In effect then, all of<br />

storage above the aaa percentage can be regarded as a buffer reserved for Q0<br />

users. (Q0 users should not take too much of it, because this unspecified buffer<br />

also serves to hold the pages of active users who are temporarily “dormant”<br />

[between transactions].)<br />

Although Q3 users may occupy the percentage of storage represented by ccc,<br />

they do not have exclusive rights to it. This storage can be occupied by Q0, Q1,<br />

Q2, <strong>and</strong> Q3 users, depending on which classes are generating a dem<strong>and</strong> for it.<br />

But the Q1 <strong>and</strong> Q2 buffers tend to eliminate some of the Q1 <strong>and</strong> Q2 dem<strong>and</strong>,<br />

so the ccc storage tends to be occupied mainly by Q3 users. Also, after a Q3<br />

user gets in, he or she stays there for a relatively long time, <strong>and</strong> this gives the<br />

Q3 users something of an advantage over other users when it comes to ccc<br />

storage.<br />

Similarly the Q2 users do not have exclusive rights to the buffer represented by<br />

the difference bbb−ccc. This storage can be occupied by Q0, Q1, <strong>and</strong> Q2,<br />

depending on which classes are generating a dem<strong>and</strong> for it. But for the most<br />

part it is occupied by Q2 users, just as the ccc storage is occupied mostly by<br />

Q3 users.<br />

The Q1 users do not have exclusive rights to the buffer represented by the<br />

difference aaa−bbb. Both Q0 <strong>and</strong> Q1 users may occupy it.<br />

STORBuf INITial<br />

sets the percentage parameters back to their initial values, 125%, 105%, <strong>and</strong><br />

95%.<br />

LDUBuf {ddd[%] eee[% ] fff[%]}<br />

partitions the commitment of the system’s paging resources when the scheduler<br />

chooses users to move from the eligible list to the dispatch list. The result is a<br />

paging capacity buffering mechanism based on the transaction class of a user.<br />

(See the description of the STORBUF parameter on page 1391 for a definition<br />

of transaction classes.)<br />

During system initialization, the number of paging DASD exposures is<br />

determined. Users expected to have high paging rates in the dispatch list are<br />

termed “loading users.” They are identified by recent DASD paging activity or<br />

with the knowledge that their working sets have been moved to DASD since<br />

they last ran. One loading user is regarded as keeping one paging exposure<br />

busy. The LDUBUF (loading user buffer) percentages, then, determine the<br />

number of loading users allowed in the dispatch list by transaction class.<br />

The SET SRM LDUBUF parameters partition the paging capacity as follows:<br />

ddd[%]<br />

is the percentage of paging exposures the scheduler is to view when<br />

considering adding a loading E1, E2, or E3 user to the dispatch list. If<br />

the total number of loading users currently in the dispatch list plus the<br />

loading En user under consideration exceeds the ddd percentage of<br />

paging exposures, the user may not be added immediately <strong>and</strong> is<br />

delayed further.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!