On 29/07/16 12:59, A. Soroka wrote:
Andy,
I seem to remember that there are some abilities to cluster in TDB2-- is that
correct or am I getting different things mixed up?
TDB2 is single machine .
Another part of the grant funded project was/is clustering butthat is
not this code (and for that matter, the Fuseki2 UI by Ian is also an
outcome of grant).
----
I have a strong interest at the moment in change propagation - keeping a
number of RDF triple stores in sync. I've been working with an upgrade
to "RDF patch" [1], simplifying it and also adding transaction markers.
Combined with a router machine to coordination overlaps, replay lost
updates etc etc you get a collection of replicas that are eventually
consistent.
Andy
http://afs.github.io/rdf-patch/
---
A. Soroka
The University of Virginia Library
On Jul 29, 2016, at 5:48 AM, Andy Seaborne <[email protected]> wrote:
There are snapshot/preview builds of TDB2 now available.
* TDB2 has scalable transactions
* TDB2 has write-once transactions (no RDF data in the journal)
and generally better written for long term maintenance and stability.
** TDB2 is not compatible with Apache Jena TDB (TDB1) databases. **
This is not part of Apache Jena - it is a personal project and does not come
with the backing of the Apache Software Foundation. (Whether it will be in the
future is for discussion with the PMC.)
This is a preview release - it is subject to change and while "it works for
me", it is likely to have rough edges.
TDB2 from Java:
https://github.com/afs/mantis/blob/master/use-tdb2.md
TDB2 in Fuseki:
https://github.com/afs/mantis/blob/master/use-fuseki-tdb2.md
Codebase:
https://github.com/afs/mantis
Note that the artifacts come from the Sonatype snapshot repo so you need to
configure maven/gradle/... appropriately.
Example:
Loading 100m of BSBM data into a live Fuseki (a single write transaction while
the server was able to answer queries at the same time) with a default heap
size:
SSD: 70K triples/s (about 24 minutes)
Disk: 47.5K TPS (about 35 minutes)
(That said, my SSD is in a desktop machine it's not particularly fast. Modern
servers have a much architecture.)
This would not have been possible without:
* The UK Government - InnovateUK (the Technology Strategy Board as was)
most of the work is a spin off from a grant-funded project
* Epimorphics for letting me work on that project
* GitHub for the code repository
* Sonatype for the maven repository and route to Maven Central
* TravisCI for the continuous integration server.
Andy
