Menu
▾
▴
re[8]: [Jdbm-developer] Concurrency in translation system / n on-tran sactional items
|
From: 'Kevin D. ' <ke...@tr...> - 2006-03-02 23:04:40
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0
Transitional//EN">
<HTML><HEAD>
<STYLE type=text/css> P, UL, OL, DL, DIR,
MENU, PRE { margin: 0 auto;}</STYLE>
<META content="MSHTML 6.00.2900.2802" name=GENERATOR></HEAD>
<BODY leftMargin=1 topMargin=1 rightMargin=1><FONT
face=Tahoma>
<DIV><FONT face=Arial size=2>Bryan-</FONT></DIV>
<DIV><FONT face=Arial size=2></FONT> </DIV>
<DIV><FONT face=Arial size=2></FONT></DIV>
<DIV><FONT face=Arial size=2>Fair enough.
I think that if explicit clustering is desired
that a container like your logical container
(or a BTree with clustered indexes) would
be a good choice... The clustering
requirements of these kinds of records are
best managed at the application level anyway
(where there is implicit knowledge of which
records should be co-located).</FONT></DIV>
<DIV><FONT face=Arial size=2></FONT> </DIV>
<DIV><FONT face=Arial size=2>The implicit
clustering I was talking about would be for
the general case where there is no application
level knowledge of how best to cluster records.</FONT></DIV>
<DIV><FONT face=Arial size=2></FONT> </DIV>
<DIV><FONT face=Arial size=2>- K</FONT></DIV>
<DIV><FONT face=Arial size=2> </FONT>
<TABLE>
<TBODY>
<TR>
<TD width=1 bgColor=blue><FONT face=Arial
size=2></FONT></TD>
<TD><FONT face=Arial size=2><FONT color=red>>
Kevin,<BR><BR>I was thinking of a persistent
data structure providing a collection,<BR>not
a Java collection class such as java.util.TreeMap.
So there is a<BR>record for the container
and then individual records for each entry
in<BR>the collection. For example,
I use a double-linked list to maintain<BR>many-to-one
associations among records. One of
the records maintains<BR>the head/tail/size
of the many-to-one collection. The
other records<BR>have the prior/next recids.
To the extent that reads are likely
to<BR>traverse the member records in such
a collection, a significiant<BR>performance
benefit can be realized by clustering those
records onto<BR>pages.<BR><BR>More broadly,
my point is that clustering can be deliberate.
To the<BR>extent that deliberate record
cluster improves read performance, then<BR>NOT
using an in-place update will degrade read
performance.<BR><BR>At this point there is
no facility for clustering in jdbm so it
is<BR>not possible to empirically characterize
the performance tradeoff<BR>for any given
application.<BR><BR>-bryan<BR><BR>-----Original
Message-----<BR>From: <A href="mailto:jdb...@li..."><FONT
color=#0000ff>jdb...@li...</FONT></A><BR>To:
JDBM Developer listserv<BR>Sent: 3/2/2006
9:58 AM<BR>Subject: re[6]: [Jdbm-developer]
Concurrency in translation system / n<BR>on-tran
sactional items<BR><BR>Bryan-<BR><BR>I'm
sure we'll have many interesting discussions
about the best approach<BR>for this - but
just to address a couple of points made:<BR><BR>Collections:
A collection is a higher level abstraction
that is treated<BR>as a single record. If
you have multiple entries in a page of a<BR>collection,
and you change one entry, then the entire
collection page<BR>will be written back,
so the proximity is not effected. Similarly,
if<BR>you were runnnig the current jdbm h*tree
in the system I propose, each<BR>page of
the tree would be written as a whole, so
making a change to one<BR>element in the
tree would not impact the hash grouping.<BR><BR>There
are other advantages to always writing changed
records to clean<BR>pages, and I'm pretty
sure that we will eventually find that these<BR>advantages
outweigh any bennefits of writing in-place.
Some examples of<BR>the advantages:<BR><BR>1.
Record header does not have to contain
a 'consumed' byte count. We<BR>are
going to have to add a reverse lookup for
the logical recid which<BR>will increase
the record header size - not having the consumed
byte<BR>count will make the overall hit 4
extra bytes instead of 8. Minor<BR>point,
but definitely an advantage.<BR><BR>2. IO
performance in the safe. If records
are updated in-place, the<BR>number of pages
that must be written to the safe increases
dramatically.<BR><BR>3. Complete removal
of the free row search bottleneck (as discussed<BR>already)<BR><BR>4.
Implicit proximity of records<BR><BR><BR>There
may be other advantages, but I believe that
the above (especially<BR>#2) will far outweigh
any advantages to in-place updates.<BR><BR>So
we can get a handle on it, what advantages
do you see to in-place<BR>updates? If
we can get a list going, then we can do a
side by side<BR>comparison and make a solid
decision one way or ther other.<BR><BR>Cheers!<BR><BR>-
K<BR><BR><BR> <BR> >
Kevin,<BR><BR>It is not a problem if we write
back records in place when they fit<BR>in
place, but only how we recycle records which
are no longer used.<BR>While implicit proximity
can be obtained in the manner you suggest,<BR>it
can also break apart a set of records which
were clustered earlier.<BR>For example, you
write a collection and later update 1 row
in the<BR>collection. Those rows will
now be on discontinuous pages.<BR><BR>I do
agree with the principle that laying out
serialized records onto<BR>blank pages is
much faster and has lower contention. Java
uses a mark<BR>and sweep approach to clear
out historical pages, but Java is not facing<BR>I/O
costs and jdbm does. So the tradeoff
will be somewhat different.<BR><BR>-bryan<BR><BR>-----Original
Message-----<BR>From: <A href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff>jdb...@li...</FONT></A><BR>To:
JDBM Developer listserv<BR>Sent: 3/1/2006
10:27 PM<BR>Subject: re[4]: [Jdbm-developer]
Concurrency in translation system / n<BR>on-tran
sactional items<BR><BR>Bryan-<BR><BR>A perpendicular
thought:<BR><BR>I think that if we just say
that we will never write a record back in<BR>place
(changed records always get written to new
pages, laid down one<BR>after another), then
the need for any type of allocator goes out
the<BR>window. A periodic garbage collecting
run can force rows from partially<BR>used
pages into the cache - when they fall out,
they will get laid down<BR>onto clean pages
with no space in between them... <BR><BR>I
believe that this is actually how modern
GCs operate in primary<BR>memory... The
more I've read about allocation strategies,
the more<BR>convinced I've become that they
are really only needed in systems that<BR>don't
allow relocation of records (i.e. C/C++ where
pointer arithmetic<BR>would get hosed by
a move).<BR><BR>The approach of always laying
changed rows down on new pages has the<BR>added
bennefit of providing an implicit proximity
algorithm - I think it<BR>is safe to assume
that records that are updated by a given
transaction<BR>will also tend to be read
together....<BR><BR>Anyone needing more explicit
proximity control can resort to a higher<BR>level
container (such as your logical container).<BR><BR>Finally,
this removes the free list completely from
the<BR>implementation.... no bottleneck,
no concurrent access worries, etc...<BR><BR>-
K<BR><BR> > Kevin,<BR><BR>Log-structured
stores do relieve most of the allocation
burden, but<BR>we still need to handle recycling
of free deleted rows. <BR><BR>It could be
Fibbinoci or powers of two or whatever. The
important<BR>part is reducing a first fit
linear scan of an unordered list to<BR>being
able to pop the head of the list off in O(1)
and have a <BR>"satisfactory" fit.<BR><BR>-bryan<BR><BR>-----Original
Message-----<BR>From: <A
href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff>jdb...@li...</FONT></A><BR>To:
'JDBM Developer listserv '<BR>Sent: 2/28/2006
6:28 PM<BR>Subject: re[2]: [Jdbm-developer]
Concurrency in translation system /<BR>non-tran
sactional items<BR><BR>Bryan-<BR><BR>Ah yes
- I remember now - sorry about that. I
believe that we had<BR>discussed a Fibbinoci
series approach... hard to believe
it's been so<BR>long.<BR><BR><BR>As an FYI,
I'm digging into the structured log concept,
and it may<BR>completely remove the requirement
and bottleneck for searching for best<BR>fit,
etc... I'll comment more when I'm done
noodling on the idea. I'm<BR>beginning
to see why Alex is saying that the whole
allocation problem<BR>may just go away...<BR><BR>-
K<BR><BR><BR> > Kevin,<BR><BR>I
am with you. However, the hotspot in
the free physical row list is<BR>from scanning
for the first free physical row that is a
sufficient fit<BR>(or a "good" fit). A
linked list is not sufficient for that purpose.<BR>I
think that we need to maintain an array of
free physical rows based<BR>on some sizes
(e.g., powers of 2) so that a request can
always be<BR>directed<BR>to the free list
whose entries are the smallest sufficent
size. At that<BR>point we can just
pull off the head of the list. That
action can be<BR>made<BR>atomic in one way
or another.<BR><BR>-bryan<BR><BR>-----Original
Message-----<BR>From: <A
href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff><mailto:jdb...@li...></FONT></A><BR><A
href="mailto:jdb...@li..."><FONT
color=#0000ff>jdb...@li...</FONT></A><BR>To:
JDBM Developer listserv<BR>Sent: 2/27/2006
9:14 PM<BR>Subject: [Jdbm-developer] Concurrency
in translation system /<BR>non-transactional
items<BR><BR>Here are some comments on discussions
last week:<BR><BR>Concurrency in the mapping
pages - Alex - you asked Brian about how<BR>different
txs would view the mapping system. The
answer is that the<BR>mapping pages will
be part of the non-transactional data set.
This<BR>means that all transactions
will share the same view and (only) version<BR>of
the mapping pages. The trick is how
to design the mapping data<BR>structure so
simultaneous changes by multiple tx don't
invalidate the<BR>data structure from the
perspecive of each tx. With the mapping
system,<BR>I think this is relatively easy
to do. With a b-link index, it will
be<BR>harder (but we have to handle that
situation as well).<BR><BR>Non transactional
writes - the scheme proposed by Bryan is
directly in<BR>line with my proposal from
a few weeks ago. The scanning of the
free<BR>lists can be easily fixed by implementing
the free rows as an in-place<BR>linked list
- this could actually be done in the current
jdbm system<BR>with some minor tweaks to
the system.<BR><BR>Transaction contention
when looking for the next free row would
then be<BR>limited to pulling the head of
the list - this should be quite fast.<BR>The
page of the list that contains the head would
have to be written to<BR>the safe frequently,
but that's it. By adding a third 'reserved'
list<BR>to the translation page, it should
be possible to reserve a record id,<BR>and
only actually populate that entry when the
transaction that made the<BR>reservation
commits. System failure rollback is
achieved by flushing<BR>the reserved list
(easily done by appending the reserved list
to the end<BR>of the free list). Tx
rollback would be similar.<BR><BR>The point
here is that the translation pages are on
par with a b-link<BR>based index, but with
direct access instead of tree walking. The<BR>challenges
of handling transactions in a b-link tree
will be identical<BR>to those in the translation
pages.<BR><BR>The question of how to handle
concurrency with a blink tree is actually<BR>of
interest to me - I'll throw that out in another
email.<BR><BR>- K<BR><BR><BR>Kevin Day<BR>Trumpet,
Inc.<BR><A href="http://www.trumpetinc.com"><FONT
color=#0000ff><http://www.trumpetinc.com></FONT></A>
<A href="http://www.trumpetinc.com"><FONT
color=#0000ff><http://www.trumpetinc.com></FONT></A><BR><A
href="http://www.trumpetinc.com"><FONT color=#0000ff><http://www.trumpetinc.com></FONT></A>
<A href="http://www.trumpetinc.com"><FONT
color=#0000ff><http://www.trumpetinc.com></FONT></A><BR><A
href="http://www.trumpetinc.com"><FONT color=#0000ff><http://www.trumpetinc.com></FONT></A>
<A href="http://www.trumpetinc.com"><FONT
color=#0000ff><http://www.trumpetinc.com></FONT></A><BR><A
href="http://www.trumpetinc.com"><FONT color=#0000ff><http://www.trumpetinc.com></FONT></A>
<A href="http://www.trumpetinc.com"><FONT
color=#0000ff><http://www.trumpetinc.com></FONT></A><BR><A
href="http://www.trumpetinc.com"><FONT color=#0000ff><http://www.trumpetinc.com></FONT></A>
<A href="http://www.trumpetinc.com"><FONT
color=#0000ff><http://www.trumpetinc.com></FONT></A><BR><A
href="http://www.trumpetinc.com"><FONT color=#0000ff><http://www.trumpetinc.com></FONT></A>
<A href="http://www.trumpetinc.com"><FONT
color=#0000ff><http://www.trumpetinc.com></FONT></A><BR><A
href="http://www.trumpetinc.com"><FONT color=#0000ff><http://www.trumpetinc.com></FONT></A>
<A href="http://www.trumpetinc.com"><FONT
color=#0000ff><http://www.trumpetinc.com></FONT></A><BR><A
href="http://www.trumpetinc.com"><FONT color=#0000ff><http://www.trumpetinc.com></FONT></A>
<A href="http://www.trumpetinc.com"><FONT
color=#0000ff>www.trumpetinc.com</FONT></A><BR><A
href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A>
<A href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A><BR><A
href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A>
<A href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A><BR><A
href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A>
<A href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A><BR><A
href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A>
<A href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A><BR><A
href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A>
<A href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A><BR><A
href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A>
<A href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A><BR><A
href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A>
<A href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A><BR><A
href="mailto:ke...@tr..."><FONT
color=#0000ff><mailto:ke...@tr...></FONT></A>
<A href="mailto:ke...@tr..."><FONT
color=#0000ff>ke...@tr...</FONT></A><BR>602-438-7030<BR><BR>-------------------------------------------------------
This SF.Net<BR>email is sponsored by xPML,
a groundbreaking scripting language that<BR>extends
applications into web and mobile media. Attend
the live webcast<BR>and join the prime developer
group breaking into this new coding<BR>territory!<BR><BR><BR><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>><BR><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>2><BR><BR><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>2><BR><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>2><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642"><FONT
color=#0000ff>http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642</FONT></A><BR>_______________________________________________
Jdbm-developer mailing<BR>list <A
href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff>Jdb...@li...</FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff>https://lists.sourceforge.net/lists/listinfo/jdbm-developer</FONT></A><BR><BR><<BR> <BR>-------------------------------------------------------
This SF.Net<BR>email is sponsored by xPML,
a groundbreaking scripting language that<BR>extends
applications into web and mobile media. Attend
the live webcast<BR>and join the prime developer
group breaking into this new coding<BR>territory!<BR><BR><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>2><BR><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>2><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642"><FONT
color=#0000ff>http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642</FONT></A><BR>_______________________________________________
Jdbm-developer mailing<BR>list <A
href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff>Jdb...@li...</FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff>https://lists.sourceforge.net/lists/listinfo/jdbm-developer</FONT></A><BR><BR><<BR> <BR>-------------------------------------------------------
This SF.Net<BR>email is sponsored by xPML,
a groundbreaking scripting language that<BR>extends
applications into web and mobile media. Attend
the live webcast<BR>and join the prime developer
group breaking into this new coding<BR>territory!<BR><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164"><FONT
color=#0000ff><http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=12164</FONT></A><BR>2><BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642"><FONT
color=#0000ff>http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642</FONT></A><BR>_______________________________________________
Jdbm-developer mailing<BR>list <A href="mailto:Jdb...@li..."><FONT
color=#0000ff><mailto:Jdb...@li...></FONT></A><BR><A
href="mailto:Jdb...@li..."><FONT
color=#0000ff>Jdb...@li...</FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff><https://lists.sourceforge.net/lists/listinfo/jdbm-developer></FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff>https://lists.sourceforge.net/lists/listinfo/jdbm-developer</FONT></A><BR><BR><<BR> <BR>-------------------------------------------------------
This SF.Net<BR>email is sponsored by xPML,
a groundbreaking scripting language that<BR>extends
applications into web and mobile media. Attend
the live webcast<BR>and join the prime developer
group breaking into this new coding<BR>territory!<BR><A
href="http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642"><FONT
color=#0000ff>http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642</FONT></A><BR>_______________________________________________
Jdbm-developer mailing<BR>list <A href="mailto:Jdb...@li..."><FONT
color=#0000ff>Jdb...@li...</FONT></A><BR><A
href="https://lists.sourceforge.net/lists/listinfo/jdbm-developer"><FONT
color=#0000ff>https://lists.sourceforge.net/lists/listinfo/jdbm-developer</FONT></A><BR><BR><<BR></FONT></FONT></TD></TR></TBODY></TABLE></DIV></FONT></BODY></HTML>
|