Shmuel said > >Wouldn't the TCB in the primary address space get charged for the > >time? > > No. The issuing TCB would be charged for time spent in the issuing > TCB; it would not be charged for time spent in another address space > unless there is enclave processing for the CAS. I assume that a good > chunk of the work done by CAS is an TCB mode, e.g., DYNALLOC, OPEN.
Well yes, sorta. When a TCB is running, almost all of the work that is directly executed under that TCB is charged to that TCB and either the home address space, or the enclave if the TCB happens to be joined to an enclave. A tiny percentage of time might be uncaptured, e.g. some of the interrupt handler time does not get reflected in the TCBTTIME field and just disappears as "uncaptured" system overhead. The complication for CATALOG and many similar applications, is that some of the work is done under the requestors TCB (and is charged to that TCB) and some of it is done under a completely separate TCB in CAS. In approximation; the requestor accesses CAS services via an SVC that ultimately space switches to CAS via PC. Then it posts a worker task in CAS to complete the request and waits for the result. All of the work that runs under the requestor's TCB is charged to the requestor's TCB, but the work that is done in CAS by the CAS TCB is charged to CAS - which is why CAS shows up as a significant CPU consumer even though it only exists to serve clients in other address spaces. Now in theory, CAS work -could- be charged to the requestor's enclave (if any), but that requires a fair bit of WLM API code in CAS and I am fairly sure (Mark will correct me if I'm wrong) that CAS just does not do this. So it doesn't matter whether the requestor is running under an enclave or not, the CAS time won't get billed to the caller. > >And don't forget that CAS may also PC into VLF to suck its catalog > >blocks out of cached catalog storage so as to avoid I/O if > >possible. > > Yep; multiple address spaces might be involved for the single request, > and each would rack up CPU time separately. Or not. Depending on how the required functions are implemented. In the CAS case yes. DB2 DDF, WAS and other similar applications maybe yes, and maybe no. CC ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
