The Pankian Metaphor 3061

so true randomness may be hard to achieve ... all it has to be is practically random ... and in the case of something like page replacement algorithm ... have sufficiently random characteristics that the choice is better than an extremely deterministic implementation (like least recently used) ... when the buttumptions inherit in the deterministic implementation are wrong, i.e. LRU is based on the buttumption that storage used in the recent past will be in the near future. The LRU buttumption is correct across many operational environments and workloads ... but may be incorrect for something like sequentially accessing some data.

this is one of the areas that I got tss-360 with my cp67 enhancements ... both operating systems ran on the same 360-67 hardware.

they were testing early version of tss-360 at the univ. on the same 360-67 that I was using for testing cp67. tss-360 had a one-level-store paradigm for disk-file operation .... not only were binary executables memory-mapped but all files (even data files with sequential access patterns) were memory-mapped. however, tss-360 only had LRU-like algorithm for page replacement ... and no special provisions for doing things like recognizing sequential access of data files.

anyway, the univ had a lot of work that was fortran test ... i.e. fortran compile, load, and execute. we put together a representative fortran workload and ran it under both tss-360 and cp67 (using the same hardware). for tss-360 test, it ran four simulated "users" all executing the same fortran workload. On cp67, I ran 30 simulatated "users", all concurrently executing the same fortran workload. The avg. interactive response on tss-360 for four users was several seconds. The avg. interactive response for cp67 for 30 uses (running the same workload) was subsecond.

One of the interesting gambling examples is comparison of Los Vegas and gov. lottories. Supposedly the psuedo randomness in the Los Vegas slots is setup to return something like 98-99 percent of bets. The casinos slots keep only one to two percent on each slot played. The casinos make their money on the churn ... expecting the slots to be played over and over. On one hundred plays, the slots randomness would supposedly retain 1-(.99**100) or about 1-.37 = .63 of the money.

Now a number of the state lottories claim to return on the order of 80precent (keeping 20percent on each play). Doing a similar type of analysis (used for slots) of a weekly lottory ... if it were four weeks of churn ... then it would be 1-(.8**4) or about 1-.4 = .6 of the money (buttuming constant starting amount for buying tickets and no new money introduced each week).

Governments seem to have a strong preference for such lottories, especially big winning lottories, since they also get to tax the lottory winnings as income. As a result, the actual amount of money retained by the gov. in each weekly lottory play is probably closer to 40 percent (20 percent that they take directly off the top as part of the play and possibly another 20 percent that they take in taxes on the lottory "winnings"). That makes the government take on four week (weekly) lottery churn calculation closer to 1-(.6**4) or about 1-.13 = .87.

In the slots churn, an individual might start with hundred dollars in quarters and after one hundred plays (on the avg), only have 37 dollars left. The weekly gov lottory churn calculations is somewhat obfuscated being spread across a large number of individuals and with "new" (additional) money typically being introduced into the lottery ticket infrastructure each week. Attempting to translate the gov. lottery analysis into the casino slot scenario ... the example might be a community decides to allocate $1m for buying lottery tickets. After the first drawing, the gov. keeps $200k and returns $800k in winning. However, the gov. also possibly gets another $200k in taxes on the $800k distributed in winnings ... so effectively there is only $600k to buy lottery tickets the next week. This continues for four weeks. In theory, after four weeks, the gov. has $870k (of the original $1m) ... in effect, the gov. gets to skim on the order of 40percent of the total bet each week (in a combination of what it takes off the top and what it gets to take as taxes on winnings treated as income).

--