"William" == William Lee Irwin <wli@holomorphy.com> writes: William> Well, consider it briefly for a moment: William> Q1: what takes the most time during boot? William> A1: inefficient userspace initialization William> but worse yet, firmware stuff On Wed, Jan 23, 2002 at 11:10:04AM +0200, Momchil Velikov wrote: > We don't (re)boot :) Noble, but perhaps more worthy of a system administrator than a kernel hacker given the old "You mess up, you reboot" rule, and the further constraint that testing a new kernel patch requires rebooting... "William" == William Lee Irwin <wli@holomorphy.com> writes: William> Q2: what takes the most time during application startup? William> A2: disk accesses William> A1 and A2 have some impact on what could perhaps be considered William> important (to some people) aspects of performance. For instance, William> boot times contribute to some rather large latencies important William> to clustering folk. On Wed, Jan 23, 2002 at 11:10:04AM +0200, Momchil Velikov wrote: > "Using a cluster to hide the fact that the underlying systems crash > regularly is an extremely dangerous way to manage a computing > environment." > -- Matt Dillon In my heart of hearts I'm in general agreement with this notion. I don't have any particular defense of the clustered computing model offhand aside from the vague notion that with large enough collections of hardware, hardware failure rates start to require full-time teams of individuals dedicated to replacing failed disks nonstop 24/7... "William" == William Lee Irwin <wli@holomorphy.com> writes: William> At any rate, the filesystem profiling + defragmentation doesn't William> seem like a bad idea at all, though I'm wondering how much of it could William> be driven from userspace. Probably the minimal kernel support needed William> would be some kind of inode-based access profiling. So (2) perhaps William> also has an answer. On Wed, Jan 23, 2002 at 11:10:04AM +0200, Momchil Velikov wrote: > Defragmentation may result as a side effect of online logical volume > resizing effort. I think the sort of defragmentation sought here is specific to the on-disk organization individual files determined by profiling to be important, not block allocation in general. Cheers, Bill -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ IRC Channel: irc.openprojects.net / #kernelnewbies Web Page: http://www.kernelnewbies.org/
