Re: SMP Support
"R. Tyler Ballance" <[email protected]>
| Newsgroups | gmane.comp.micro-kernel.l4.l4ka.general |
|---|---|
| Message-ID | <[email protected]> |
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 This topic came up yesterday in #l4bsd on freenode, so I'll paste the transcript from the general "gist" of the conversation - --------------------- [11:12a] rtyler|intel: benno: how does Iguana (and in turn N1 pistachio) handle SMP? [ snipped intermediate conversation ] [6:28p] benno: rtyler|intel: umm, it handles it just fine [6:38p] froody left the chat room. (Read error: 110 (Connection timed out)) [6:41p] ggonzahome left the chat room. (Read error: 113 (No route to host)) [6:56p] rtyler|intel: benno: I have no doubt about that, it's more of a question of how Iguana handles threads and memory in a multi-processor environment [6:57p] benno: how exactly do you mean? on smp iguana will give you the option of starting threads on different CPUs [6:57p] benno: iguana itself will stay single threaded [6:58p] benno: (as most of its operations would require holding a lock anyway, using IPC as the syncronising method is basically good enough..) [6:58p] rtyler|intel: benno: I guess just related to how FreeBSD does things (SMPng is my basis for any of this knowledge [6:58p] benno: no idea how freebsd does things [6:58p] rtyler|intel: i.e. 4.xx had the giant lock, and SMPng has brought fine-grained locking to 5.xx and 6.xx [6:59p] benno: right, iguana doesn't have fine grained locking, its single threadedness is basically a giant lock [6:59p] benno: but, you don't really call iguana nearly as often as you would call the bsd kernel [6:59p] benno: so it isn't clear that fine-grained locking would actually buy you anything (and in fact could have worse performance) [6:59p] rtyler|intel: but then what about services that it serves? do they share the same giant locking mechanism, in regards to the threading that iguana provides [7:00p] benno: well each server running on top of iguana can implement the policy it sees best [7:00p] benno: but really for most of them, the services then selves are fine grained [7:00p] benno: so each of these having a giant lock is kind of the same as having fine grained locks in a traditional kernel [7:01p] benno: well, not fine grained, probably coarse grained, but its not like the giant lock [7:01p] rtyler|intel: benno: I guess where i'm losing the train of thought is in that Iguana essentially operates as a giant lock, and provides threading to it's services, as to how those iguana services would run atop SMP systems [7:02p] benno: .. i need a whiteboard... [7:02p] benno: it is really L4 providing the smp support [7:02p] benno: it provides mechanisms to let you start threads and different CPUs [7:02p] benno: and migrate threads between different cpus - ------------------------------------------- Hope that clears things up :) Cheers, - -R. Tyler Ballance On Jun 9, 2006, at 10:32 AM, Michael Emswiler wrote: > I am interested in Multiprocessor Support in Pistachio. > > I was wondering if anyone could point me to a link where i could > find current status on SMP support and what's in-progress? > > I am primarily interested in EMT64 and AMD64. > > Thanks, > MikeE > > -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.3 (Darwin) iD8DBQFEia4nqO6nEJfroRsRAlksAJsHYe5FGuaUcw604ySS+vW4ECtH8QCfYN0r vZyHZvLmCjOvd5LGCdPIzws= =K7r2 -----END PGP SIGNATURE-----