The Question is:
Over the past 10 years or so I have repeatedly heard the following statement:
during a full shadow copy operation, the shadow server process limits the
number of IO's to 20 per second for that particular DSA device.
Meaning, appart from the IO's that are done by the shadow server process in
order to complete the shadow coppying operation, the DSA device is limited to
20 IO's / second for users, programs etc.
Is this true?
Best Regards.. Joe
The Answer is :
The Shadow_Server process has NEVER -- since the product's inception
in the V5.4-1 release -- limited the rate at which is issues $QIOs to
the SHDRIVER during a "full shadow copy operation".
Having said that, the terms merge and copy are regularly confused,
though each refers to a distinct and different operation.
The Shadow_Server process does use an algorithm to limit the rate at
which it issues $QIOs to the SHDRIVER during a full merge operation.
That algorithm can be influenced by using system logical names:
For all virtual units connected on a node:
For a specific virtual unit in a node:
Details of this logical name are in the ECO kit release notes for the
shadowing-related kits for V6.2 thorough V7.2-1H1.
As for the reference to 20 I/Os per second, while a very early version
of the SHDRIVER did involve throttling read requests served by the
master member when the UCB$L_QLEN of the members reached 20. That has
long since been replaced with another throttling algorithm.
In any case, the throttling limits the number of QIO's issued by the
shadowing code, so that it will not compete too much with user QIO's.
No throttling is applied to user QIO's, although they will see reduced
performance because of the competing I/O performed by the merge.