Chas, it has been a long time since I looked at the code, but I believe that
there will be redundent caching in jdbm and you may wind up paying for
redundant IOs because you are not reusing the same cache. -bryan
________________________________
From: Chas Emerick [mailto:[email protected]]
Sent: Thursday, October 22, 2009 11:14 AM
To: Bryan Thompson
Cc: 'Alex Boisvert'; 'JDBM General listserv'
Subject: Re: [Jdbm-general] Read concurrency clarification
Simplicity of implementation. If I've got a number of totally separate modules
that are read-only, there's no reason to introduce any kind of coordination
ceremony if multiple readers are safe (which, by all rights, should be a given
for any data store).
- Chas
On Oct 22, 2009, at 10:59 AM, Bryan Thompson wrote:
I believe that the B+Tree tuple cursor is unsafe under concurrent read -- that
is, updates to its internal state are not thread safe.
Why would you use more than one recman instance for a given store in the same
JVM?
-bryan
________________________________
From: Chas Emerick [mailto:[email protected]]
Sent: Thursday, October 22, 2009 10:45 AM
To: Alex Boisvert
Cc: JDBM General listserv
Subject: Re: [Jdbm-general] Read concurrency clarification
Right, which happens to be my situation: many concurrent reads from different
threads, and when a write needs to be done, we lock globally.
Thanks for the (extended) clarification :-)
- Chas
On Oct 22, 2009, at 10:41 AM, Alex Boisvert wrote:
No, that's not safe at all.... unless all the access is read-only.
alex
On Thu, Oct 22, 2009 at 7:36 AM, Chas Emerick
<[email protected]<mailto:[email protected]>> wrote:
Alex,
Thank you for the clarification. Just to make sure we're understanding each
other, I'm actually asking about multiple readers, each opening and using their
own RecordManagers. Your mention of synchronization leads me to think you're
talking about multiple threads reading from a single RecordManager.
Just FYI, I'm being careful here because I came across a mention of RM changing
files on disk when *reading*, which surprised me (can't find the link right
now)...
Thanks again,
- Chas
On Oct 22, 2009, at 10:27 AM, Alex Boisvert wrote:
Individual concurrent reads are safe. Pretty much everything is synchronized
at the top of level.
However, iterators / tuple browsers are unsafe under concurrent updates. You
would have to synchronize on the RecordManager instance to maintain a
consistent view while iterating.
alex
On Thu, Oct 22, 2009 at 5:10 AM, Chas Emerick
<[email protected]<mailto:[email protected]>> wrote:
I've been plumbing around the list archives for a while, but they're
something of a jumble (thanks, Sourceforge!), so I figured I'd open a
new thread to ask:
What level of read concurrency does jdbm support?
What seems clear is that, to be safe, there should only be one thread
writing to a jdbm database (although it appears that it is possible
for multiple threads to participate in a transaction, I'm perfectly
fine with serializing all write access). However, I came across at
least a number of comments that lead me to believe that concurrent
reads are unsafe as well, although they're all from pre-history (e.g.
2001).
Thanks,
- Chas
------------------------------------------------------------------------------
Come build with us! The BlackBerry(R) Developer Conference in SF, CA
is the only developer event you need to attend this year. Jumpstart your
developing skills, take BlackBerry mobile applications to market and stay
ahead of the curve. Join us from November 9 - 12, 2009. Register now!
http://p.sf.net/sfu/devconference
_______________________________________________
Jdbm-general mailing list
[email protected]<mailto:[email protected]>
https://lists.sourceforge.net/lists/listinfo/jdbm-general
------------------------------------------------------------------------------
Come build with us! The BlackBerry(R) Developer Conference in SF, CA
is the only developer event you need to attend this year. Jumpstart your
developing skills, take BlackBerry mobile applications to market and stay
ahead of the curve. Join us from November 9 - 12, 2009. Register now!
http://p.sf.net/sfu/devconference
_______________________________________________
Jdbm-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/jdbm-general
