Performance Analysis & Workload Monitor

Why does SAP recommend setting stat/max_files to 99 in production rather than leaving the default 48?

Average database read time is around 0.1 ms per row vs another transaction at several times that. What does each number actually tell you to do?

Response time R is measured on the application server. Does it include the time SAP GUI itself spends working on the dialog step, and why does that matter for a diagnosis?

TDB is high. Before you blame the HANA engine, what else is folded into that number?

Why does SAP write per-dialog-step statistical records to binary flat files in the instance directory instead of database tables?

At 2pm you need system-wide (TOTAL) workload across all instances, but the TOTAL figures look like yesterday's. What's going on and how do you get current numbers now?

A STATS record shows a slow dialog step, but it was triggered by an RFC from another system. What in the record traces it back to the true origin?

Two transactions contend for the same lock and one stalls. Will that stall show up under enqueue time TENQ?

An admin proposes setting stat/level to 0 in production to save overhead. What does SAP say, and what is the one statistics feature that actually carries overhead?

You want the single program consuming the most resources across the system. Why can the Standard transaction profile mislead you, and which view gives the truth?

To measure how long a dialog step waited on RFC calls, why is CPIC/RFC time more reliable than roll wait time?

For which workloads is the ST03 response time distribution actually meaningful, and which one should you NOT read this way?

A dialog step waits on an RFC call. Why doesn't that waiting tie up the work process, and why is TRW always 0 for background steps?

An audit will ask about a performance incident three weeks from now. What must you do in STATS today that its predecessors couldn't, or the detail is gone?

Users report a transaction 'froze for 30 seconds yesterday' but ST03 averages look fine. Why do you go to STAD/STATS, and which is the better choice for new work?

Dialog steps show high wait time TW. Does this point you at the program, the database, or somewhere else?

You need table-access statistics to chase a DB problem. Why must you NOT just leave them on, and how do you scope them?

Performance fell off a cliff sometime in the last hour. Why won't the standard hourly time profile pinpoint it, and what ST03 function will?

A dialog step shows high GUI time TGUI. What is it actually counting, and why does that send you to the frontend rather than the server?

Why is roll-in time TRI typically tiny in S/4HANA, and what older mechanism does this replace?

The ST03 memory profile shows climbing work-process 'reservations' for a transaction. What does that actually tell you to fix?

Response time is always 'per dialog step' — but a dialog step means different things by workload. What counts as one for online, background, RFC, web, and update?

In the ST03 workload overview, why are UPDATE (V1) and UPDATE2 (V2) split into separate task types instead of one 'update' bucket?

A STATS single record shows high response time. Which tab confirms the cost is actually in the database, and which one tells you what KIND of access to attack?

A nightly batch issue happens around 3am, but the ST03 time profile lumps your overnight hours together. Why, and what do you change so future analysis has hourly detail?

The hourly job SAP_COLLECTOR_FOR_PERFMONITOR has been failing for two weeks. What breaks, and what part of the loss is unrecoverable?

You want to compare this quarter's workload to last quarter's in ST03, but the data only goes back weeks. Why, and what do you change so the comparison is possible next time?

A dialog step's processing time is far larger than its CPU time. What does that gap mean, and what causes should you rule out before blaming the code?

ST03 shows healthy application response time R, yet remote branch users feel the system is slow. What metric reflects their reality and what does R leave out?

You're hunting a long-running job in STATS. Why is it a mistake to select a wide interval spanning its whole runtime, and what interval should you pick instead?

When does an ST03 weekly workload aggregate begin and end, and who gets tripped up by it?

Processing time TP is high. Why is it wrong to read that as 'the ABAP code was busy computing', and why is TP slippery in general?

A user reports an abnormally long activity from two days ago, but its raw stat record has already aged out. What ST03 feature still lets you find it?