The Trilemma
Progressive Taxation as a Political Shield for Globalization

Hoisted from Comments: Life Is Not Perfect

Hoisted from comments: Maynard Handley says life is not perfect:

Grasping Reality with Both Hands: Brad DeLong's Semi-Daily Journal: Not Your Father's Microcomputer: But life is not perfect yet.

Safari fragments memory in an ungodly fashion, meaning that after it's been open and used aggressively for a few weeks, you really want to restart it --- either than or accept that massive paging that occurs every time a page is loaded or closed. (And it occasionally crashes.) Obviously the ideal would be a Safari that did not fragment memory in this way. Second best would be a Safari that, on quit or crash, could restart with the same set of open pages. Unfortunately the existing Safari gives us neither. Rumors differ on the Safari 3 of the future. No beta report I've seen mentions long term memory behavior, and while one report implies you can restart everything after a quit, I've seen no confirmations, and no indication of how well it works after a crash.

What we really need, longer term is

  • the ability to restore state after a quit or crash as part of every program and
  • the ability to restore all programs after those occasional necessary reboots (OS upgrades, hardware installation)

I'd like to see Apple or MS get to work on these since they both seem to be reduced to pretty pathetic OS-wide improvements, given their most recent offerings. And honestly, both of these really should not be that hard. Opening all apps on reboot is trivial, as is providing some UI guidance and example in all the Apple apps of what is expected in terms of saving state when an app reopens.

(There do remain the occasional weird things that can crash OS X. The last one I found was making some fctl() calls on a file of length zero. That one is definitely fixed in Leopard, and probably fixed in a recent Tiger update.

The more serious one is the software RAID support which is so crappy, even in Tiger, as to be useless. Mirrored RAID, at least over FW400, if one of the drives goes bad, will hang the machine --- which kinda makes you wonder what the point of the RAID was. I have seen this happen on two different machines with two different collections of drives.

This is specially galling as a paper from Usenix 2000 or 2001 discusses problems with software RAID, describes a method for injecting faults into the system, tests across a variety of OSs, and concludes that SW RAID for Windows is pretty damn solid, while for Linux [at that time] it's pretty damn flaky; OS X did not offer SW RAID at the time. It would be nice if someone at Apple read that paper and followed its methodology.)