Replies

Back in the day, we had some software that monitored mainframe performance. It had to be very efficient because you don't want to impact the performance of the system you are monitoring. Unfortunately, the information available from the OS was very arcane to wade through at runtime....to get at over 2000 metrics each 15s monitoring interval, you had to request the latest stats buffer from the OS and indirectly go through various links and references...from 6 to 12 references deep....to find the data for the particular metric. Even with assembler, that was too much overhead.

Our solution was the following: at startup, we requested the buffer initially, found the locations for each metric for that particular OS configuration, and rewrote the instructions to acces each in a DBank (data banks were modifiable; IBanks — instruction banks — we're not). When we were done, we deleted the old IBank, changed the DBank to mark it as an IBank, and the rest was history.

Extremely efficient. Our competitor's performance could come close. Looking back, and if we had wanted to be malicious, we could have done whatever we wanted, as the privileges to obtain those metrics already meant we had access to anything.

This was 30 years ago. Nothing new under the sun.

Kids today think they are doing things for the first time. Heh.