[email protected] (Ted MacNEIL) writes: > A long time ago, in a lab far far away, IBM made a (possibly > erroneous) decision when CPU was expensive to not measure > everything. The rough guideline, so I was told was it took 4-8000 > instructions to measure an event. So, anything under 20K was not worth > it (again, I'm quoting what I was told from [old] memory). Also, at > the time, no console activity measured (ie: operator commands). Some > things changed when SMF Typ30-6 waw introduced (System Address > Spaces).
re: http://www.garlic.com/~lynn/2014b.html#78 CPU time http://www.garlic.com/~lynn/2014b.html#80 CPU time http://www.garlic.com/~lynn/2014b.html#81 CPU time http://www.garlic.com/~lynn/2014b.html#82 CPU time http://www.garlic.com/~lynn/2014b.html#83 CPU time tymshare was one of the virtual machine-based online commercial service bureau ... among other things they made their online computer conferencing system available free to *SHARE* starting in aug1976 ... vmshare archives http://vm.marist.edu/~vmshare they also wrote a capability-based 370 operating system GNOSIS ... a major objective was to do detailed accounting for detailed user use of 3rd party applications, databases, files, etc ... to be able to remit to the 3rd parties, pro-rated revenue based on user use of each resource. in the 80s, M/D bought tymshare and spun off a number of things, including GNOSIS (into new company key logic). I was brought in to audit/evaluate GNOSIS as part of the spin-off. It turns out all the accounting in the capability boundary crossing was using 30% of the processor. After the spin-off, they removed all the 3rd party accounting from the capability interfaces that significantly speeded up things. They were able to show KeyKOS (aka GNOSIS) running various kinds applications faster than IBM's TPF ... and with significantly higher security than any other operating system. The high level capability abstraction allowed a lot of performance optimization that isn't possible in traditional operating system ... and at the same time significantly increased the performance. It is too bad that they weren't able to interest any IBM mainframe customers. However a number operations have taken the KeyKOS/GNOSIS design and moved them to other hardware platforms ... aiming for EAL7 level security ... *AND* high performance. Note that capability was part of the IBM Future System design ... but done in the hardware which resultetd in horrible performance (no benefits of higher level abstraction optimization). A major factor in performance analysis that 370/195 applications run on an FS machine made from 370/195 components would have throughput of 370/145 (factor of 15-30 times slowdown), which contributed to its eventual failure http://www.garlic.com/~lynn/submain.html#futuresys Capabilty was also part of the IBM's s/38 (folklore that after FS failure, some FS people retreated to rochester and did the s/38). The as/400 which was combination of s/38 & s/36 followon dropped the capability stuff. -- virtualization experience starting Jan1968, online at home since Mar1970 ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
