Re: Archiver question

zhu chao wrote:
> "Joel Garry" <joel-garry_at_home.com> wrote in message
news:<1114105599.699424.254430_at_z14g2000cwz.googlegroups.com>...
> > chao_ping wrote:
> > > OK, to summarize it, since the whole post is a little lengthy.
> > > >From lsof result, you can see file handle 12 is is the first
member.
> > > >From strace result, you can see all the read from the logfile is
> > from
> > > file handle 12, which is the first member.
> > > Other pread only read 8192 blocks, maybe only the header.
> > >
> > > Yong Huang provide the suggestions to do the test and we verified
it.
> > > Not all the meatlink notes are correct. Some can be wrong. Not
all
> > the
> > > words the experts said are correct, some can be out of date:)
> >
> > Thanks guys, that is good to know. Obviously, O8 changed the
archiving
> > greatly. It will take a day or two for me to go through the
> > registration rigamarole at that site, so I'll ask here: Did you
test
> > multiple archiver processes under severe load?
> >
> > jg
>
> Ok, how do you define severe load? When the archiver can't catch up?

Yes, the general idea is that if archiving can't keep up, additional archiver processes are spawned. So more than one additional process would be considered severe. I'm not counting extra arc processes spawned by rman, those "Log actively being archived by another process" messages tell me rman is just dumbly adding a new process.

> One logfile sequence is always only archived by one archive process.

That is what I can't find documented anywhere, why do you say it?

> So do you think multiple archive process will make any difference?

If they don't, why would they happen?

>
> Did you do some test to verify it yourself?

Unfortunately I'm not set up to test any of this, so I'm dependent on the kindness of strangers, and observing what happens on production systems.

As a thought experiment, I can imagine reasons to be able to have multiple archivers work on a single groups different members, and reasons not to (such as, maybe whoever wrote the O8 archiving code decided a simple read/append on modern hardware would generally work better) - and of course, the O7 single archiver did _used to_. What bothers me for the real world is, if single processes work on single members, wouldn't you want to have many smaller log groups so when there is a need for multiple arc processes, each can be done with the members faster? This is the opposite of the generic advice given by Oracle in the manuals, not to mention most DBA's habits. And that I can't find any docs that answer the question, even if wrong, is a red flag (of course, I may just be missing something obvious and would be glad if that were pointed out - maybe the answer to this is, er, archived somewhere from long ago). Unsubstantiated assertions are worse than useless, they are myth generators.

Mr. Cadot's test is very good as far as it goes, but I'm not at all convinced it can be generalized to multiple archiver processes. I hope someone does test such a situation, so we can know one way or another. And I can't help but point out, it may differ based on, say, different filesystem capabilities or configurations.

jg

--
@home.com is bogus.
The word is, beer:
http://www.signonsandiego.com/uniontrib/20050503/news_1b3strauss.html