home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.isis
- Path: sparky!uunet!spool.mu.edu!darwin.sura.net!jvnc.net!news.edu.tw!news!nchud5.nchu.edu.tw!dec8.ncku.edu.tw!eembox!jsw
- From: jsw@eembox.ncku.edu.tw (Master Jyh-Shi Wu (82.9))
- Subject: More about Causality
- Message-ID: <1992Dec27.124442.18404@dec8.ncku.edu.tw>
- Sender: usenet@dec8.ncku.edu.tw (USENET Manager)
- Nntp-Posting-Host: eembox
- Organization: National Cheng-Kung University
- X-Newsreader: Tin 1.1 PL3
- Date: Sun, 27 Dec 1992 12:44:42 GMT
- Lines: 58
-
-
- After read the answers of Mr. Birman, here's a lot more questions of
- causality in isis:
-
- 1) As I know, isis v2.1 multicast ordering is arranged in the unit of UNIX
- processes (i.e. heavyweight), Is lightweight mulitcast ordering (i.e.
- in the unit of "tasks") implemented in v2.2.7 ?
-
- 2) There are still little documents in bypass modes of isis [a|g|c]bcast(
- "B",...) . It's very glad to know that causal abcast is implemented
- in bypass mode. But, what's the difference in the bypass mode of
- abcast, cbcast and gbcast?
- what's the essence of "bypass mode" (more than speedup in delivery,
- what I concerned is ordering)
-
- if the bypass mode of abcast is done by cbcasts, then, is "bypassed
- abcast" can show causality when used in conjugated with cbcasts?
-
- If abcast & gbcast are causal within a group, what's the difference
- in abcast and "bypass abcast" when used within a group?, and bypass
- cbcast & ordinary cbcast, gbcast...?
-
- I think, when specifying "Bypass" mode in [a|c|g]bcast(), isis would
- perform bypass mode if situations allows that ( example: multicast
- within a group), if don't specify bypass mode, isis would "sure to"
- use the old protocol. Is there any possibility that isis "automatically"
- turns to bypass mode when user don't specify that?
-
- 3) A causality case:
- consider 3 proesses (p1, p2, p3), first, p1 send msg A to p2, then
- send msg B to p3 after A is sent, and, after p2 receives A, p2 send
- msg C to p3. All three processes start from initial state (i.e. start
- from vector time [0,0,0])
-
- obviously C and B would be delayed in cbcast queue of p3, but, what would
- be the correct sequence if they are finally delivered?
-
- if, considering "causality" as the extension of FIFO, B should be
- delivered before C. (just consider "p1 and p3" pair)
- if, comparing the vector timestamp of B & C by cbcast protocol, B
- should precede C too.
-
- But, if we check the "happened before" chain, we would see A -> B
- and A -> C, then, a question arises: it's hard to decide the "happ-
- ened before" relation between B and C, then, how can we say B should
- be delivered before C? what's the correct answer of the causal delivery
- of B & C?
-
- Sorry for my tedious sentences, but I think, those are all important concepts
- which I cannot find in documents.
-
- Next time I would ask some technical , programming questions. If there's any
- programming template of spooling (as checkpoints) please post it (that may
- reduce my qty of problems in next posting) Thanks!
-
- Jyh-Shi Wu
-
-
-