... I suggest is that you convert struct proc from zone to kmalloc first, and then look into a possible objcache adaptation. kmalloc ... objcache really wins is in not having to completey reinitialize the allocated structures. Over the ...

dragonflybsd-kernel - Matthew Dillon - Mar 31 2008 - 15:21

... of the proc > structure in the future then moving to objcache now would save us > from having to do ... I'm now running buildworld with procs in zones. I gave a quick look ... an instance inherits lots of stuff from its parent. I'll look at ...

dragonflybsd-kernel - Nicolas Thery - Apr 1 2008 - 03:05

... suggest is that you convert struct proc from zone to kmalloc= > first, and then look ... wins is in not havin= g > to completey reinitialize the allocated structures. Could we ... to move all fast path allocations to objcache? I don't like the ...

dragonflybsd-kernel - Simon 'corecode' Schubert - Mar 31 2008 - 15:39

... ran make buildworld twice with procs in objcache and twice with ... 'm now running buildworld with procs in zones. Same duration with ... instance inherits lots of stuff from its > parent. I'll look ... t think there is much to factor out. Anyway, there are ...

dragonflybsd-kernel - Nicolas Thery - Apr 1 2008 - 13:52

... either, and if the intent is to optimize the initialization of the proc structure in the future then moving to objcache now would save us from having to do it later. So whichever method ...

dragonflybsd-kernel - Matthew Dillon - Mar 31 2008 - 19:11

... Are the significant portions of the proc structure > that can be restored to initialized state on object puts? If ... getting rid of zones. I thought that even without the pre-initialization part, objcache has the advantage of pre-allocating set of ...

dragonflybsd-kernel - Nicolas Thery - Mar 31 2008 - 15:14

... Are there significant portions of the proc structure >> that can be restored to ... even without the pre-initialization part, objcache has > the advantage of pre-allocating ... and is additionally > mp-safe (contrary to zones and, granted, as malloc()) ...

dragonflybsd-kernel - Jeffrey Hsu - Mar 31 2008 - 22:18

... : > Well, the zone allocator is MP-safe. It's not hard to lock up the places that > ... is what triggered me into migrating stuff off zones. Thanks for all ... ll precise my current plan in my forthcoming reply to Matt. Nicolas ...

dragonflybsd-kernel - Nicolas Thery - Apr 1 2008 - 02:54

... Hsu : > >> Well, the zone allocator is MP-safe. It's not hard to lock ... the zlock spinlock in the zone allocator code to > see for yourself that ... twice concurrently on the same ZONE_INTERRUPT zone? I reckon we could end up ...

dragonflybsd-kernel - Nicolas Thery - Apr 1 2008 - 11:25

... I think you misread that post. The zone allocator itself is MP-safe. (Again ... the zlock spinlock in the zone allocator code to see for yourself that ... MP-safe.) The post is refering to the subsystems that use the zone allocators as being not MP-safe. ...

dragonflybsd-kernel - Jeffrey Hsu - Apr 1 2008 - 03:26

... pre-initialized objects. Are the significant portions of the proc structure that can be restored to initialized state on object puts? If not, it's best to just use a general-purpose memory allocator. Jeffrey

dragonflybsd-kernel - Jeffrey Hsu - Mar 31 2008 - 14:41

... Are the significant portions of the proc stru= cture > that can be restored to initialized state on object puts? If not, it's= best > to just use a general-purpose memory ... we use the object cache to allocate struct procs= ? cheers simon

dragonflybsd-kernel - Simon 'corecode' Schubert - Mar 31 2008 - 15:03

... such as kmalloc or zalloc and provides pre-initialized object caching functionality. If you're not going to use the object caching functionality, you can just call the underlying memory allocator directly ...

dragonflybsd-kernel - Jeffrey Hsu - Mar 31 2008 - 15:20