Dear all:
i want a database which read all data in memory when opened. can
berkeley-db do that ?
thanks in advance!


-----------------
Regards.

Your title in your question is asking how to create an in-memory
database in Berkeley DB. Documentation for how create a RAM only
database can be found here:

http://www.sleepycat.com/docs/ref/program/ram.html

In your message you ask about how to pre-load a file backed database in
cache. At least that is what I think you are asking. In order to do
that you should create a cache large enough to hold the database and
create a cursor that will walk the database. You can use the
DB_MULTIPLE flag to make this go faster. For more information:

http://www.sleepycat.com/docs/api_cxx/dbc_get.html

Ron

--
Dear :
thanks for your reply. acturally i am using OpenLdap which takes
berkeley-db as its backend db. and for performance reason i want to read all
entries in memory when startup .but i dont know whether i can do it , could
you give me some advise ?

-----------------
Regards.

"William" <W (AT) abc (DOT) com> writes:

I think that OpenLdap developers will be vastly more competent to
help you with this idea. Or talk you out of it, if appropriate.

Berkeley DB, as far as I know, does not contain an application-independant
"magically pull everything into RAM as fast as possible" knob.

best regards
Patrick

Patrick Schaaf wrote:
But something that may be almost as good is to define a BDB cache that
resides in shared memory. If the BDB cache is large enough to contain
all of the data, then once the database is primed everything will stay
in memory; on the next application startup it will still be there.

--
-- Howard Chu
Chief Architect, Symas Corp. http://www.symas.com
Director, Highland Sun http://highlandsun.com/hyc
OpenLDAP Core Team http://www.openldap.org/project/

Howard Chu <XYZ.hyc (AT) highlandsun (DOT) com> writes:

If the OS is worth its pagecache algorithms, the data in the backing
file will be there just as well. A file backed shared memory segment,
as is usually used for the cache, is practically identical in handling
to the backing file itself.

None of this is relevant for cold start first time reading of
the data, and possible latency stabilization one hopes to
gain by doing it in one go. As seemed to be the topic.

best regards
Patrick

Patrick Schaaf wrote:
That is a Very Big IF.

There is a marked performance boost on Solaris using SysV shared memory
regions instead of shared mmap'd files for the BDB cache. The difference
on Linux is unmeasurable, but not everybody runs Linux. There are
probably many other platforms where the difference is also significant;
you don't have enough information to make such definitive statements.
--
-- Howard Chu
Chief Architect, Symas Corp. http://www.symas.com
Director, Highland Sun http://highlandsun.com/hyc
OpenLDAP Core Team http://www.openldap.org/project/

Howard Chu <XYZ.hyc (AT) highlandsun (DOT) com> writes:

Note that I explicitly wrote 'file backed shared memory segment'.
SysV shared memory is not file backed.

But, anyway, point taken. I was indeed speaking from my Linux experience,
only, as I haven't had any other meaningful experience in the last >10 years.

best regards
Patrick