| PLEX86 | ||
|
winscape 2317the original (release 1) cp-67 scheduler (what i initially saw when they brought it out to the univ. in jan. 1968) was effectively a multi-level round-robin scheduler; tasks were moved between levels based on various end characteristics (which had a demon that woke up once a second and checked stuff out). this was (quickly) changed to a much simpler two level structure by somebody at lincoln labs ... in part because all the level movements were consuming significant processor overhead (on the order of 10-15 percent of total cpu with 30-35 users). i then redid that with the initial version of dynamic adaptive fair share (fair share, in part because the default policy was fair share ... but it allowed other policies also, able to support mixed-mode, interactive, batch, etc). it also further significantly reduced required pathlength at the same time and made everything much more predictable (late '60s, while still undergraduate). misc. past posts on fair share somewhat related posts on commercial time-sharing operations using the stuff for service offerings winscape 2318 To relate this EXACT topic to one near and dear to Barb's heart, Tom Hastings' name appears on both the CTSS... twenty some years later, i observed similar in some systems that also had scheduling thing that woke up once a second to adjust things. I always wondered if some of it dated back to having common heritage thread to CTSS. --
|
||||||||