Transactional Services

1
Transactional Services

• Ricardo Jiménez-Peris
• Marta Patiño-Martínez
• Technical University of Madrid

1st Adapt Workshop 23rd-24th September
2002 Madrid, Spain
2
Talk Outline

• Work on transactional systems
• Previous work.
• Current work.
• Possible contributions to the project.
• Resources
• Team.
• Computing resources.
• Funding.

3
Previous and current work on transactional
systems

• Our previous work on transactional systems has
  dealt with different aspects of them including
• Design of an advanced transaction model
  supporting multithreaded execution of nested
  transactions and conversational interaction.
• Development of an Ada 95 TP monitor implementing
  the above model.
• Replication middleware for databases (joint work
  at ETHZ with Bettina and Gustavo).
• A low latency commit service (joint work with
  Gustavo).
• Deterministic scheduling of replicated
  multithreaded transactional processes.

4
Multithreaded Nested TransactionsPrevious work

• The model is an extension of the nested
  transaction model.
• Unlike the traditional nested model that only
  allows competitive concurrency (concurrent
  subtransactions are isolated among them), it
  supports cooperative concurrency.
• This is useful, for instance, to add
  transactional semantics to a cooperative
  distributed server or to allow cooperative
  multithreading within a transaction.

5
Multithreaded Nested TransactionsContributions
to the project

• We expect that the knowledge gained in this work
  will help us in developing an advanced
  transaction model suitable for transactional
  workflow part of WP2.
• The experience with the Acta framework will also
  help in formalizing the model and formally
  proving its properties.

6
TP Monitor Previous work

• The novelties of the TP monitor were
• Its flexible architecture that allowed to change
  the concurrency control and recovery protocols
  (inspired by Arjuna).
• It supported logical logging, therefore enabling
  advanced concurrency control like Commutative
  Locking.
• It supported the multithreaded nested transaction
  model.
• It had support for replicated and cooperative
  transactional servers.

7
TP Monitor Current work

• The TP monitor is being reimplemented in Java.
• Currently, the centralized concurrency control
  and recovery has been implemented.
• The distributed part is expected to be finished
  by Christmas.

8
TP Monitor Contributions to the Project

• This work can be extended to contribute to the
  project in the following ways
• The resulting transactional engine is very
  flexible and will (hopefully) be able to support
  a wide variety of advanced transaction models to
  be used by WP2-3.
• It can be used as a base for an EJB
  implementation providing nested transactions, if
  found useful for the project.

9
Replication Middleware Previous work

• The middleware extends previous work from Bettina
  and Gustavo on database replication (Postgres-R).
• Their work focused on how to achieve scalable
  eager database replication within the database
  (white box approach).
• Our joint work with them extended this work on
  how to implement the replication outside the
  database in a middleware layer with a reasonable
  efficiency (gray box approach). More details will
  be given later by Bettina.
• Recently, we have extended the results on online
  recovery from Bettina, Alberto and Ozalp to apply
  them to the replication middleware.

10
Replication Middleware Current work

• Currently, there is a PhD student working on load
  balancing algorithms within the middleware layer,
  and a TPC-W implementation to measure the
  improvement.
• We are currently implementing the online recovery
  within the replication middleware.
• We are currently extending the replication
  protocols to deal with wide area replication.

11
Replication MiddlewareContributions to the
Project

• The results from this research will be directly
  applicable to the BS middleware.
• Availability, online recovery, and load balancing
  are essential adaptive properties.
• The middleware is more or less independent of the
  underlying data store being replicated.
  Therefore, in principle, it can be applied (in
  addition to databases) to object containers such
  as EJBs.

12
Commit Service Previous and Current Work

• This work reduces the cost (in terms of latency)
  of non-blocking atomic commitment to two rounds
  by combining optimistic commit, optimistic
  delivery of uniform multicast (extending previous
  work from ETHZ), and replication.
• We are currently implementing the service in Java.

13
Commit ServiceContributions to the Project

• This work can be extended to contribute to the
  project
• Providing efficient and non-blocking atomic
  commit for BSs.
• Supporting BTP (Business Transaction Protocol)
  for CSs and providing the necessary
  infrastructure for transactional web services.

14
Deterministic Scheduling of Multithreaded
Transactional Servers

• Previous work on active replication restricted
  servers to be single-threaded to guarantee
  deterministic behavior.
• In this work, we extended previous work by
  enabling the replication of multithreaded
  transactional servers.
• Additionally, this work tackles with servers with
  a conversational interface (similar to stateful
  session beans).
• Based on total ordered multicast and a
  deterministic scheduler that schedules all
  threads in the same way at all replicas without
  recurring to inter-replica communication.
• This work can contribute to the project by
  enabling replication of stateful session beans
  and entity beans.

15
Resources

• Team
• 2 Professors.
• 2 PhD Students
• One already working on load balancing for the
  middleware layer.
• A group of undergraduate students.

16
Resources

• Lab
• A 20-site cluster (dual AMD k7) (50 funded by
  the project).
• Funding
• A complementary grant has been requested to the
  Spanish Ministry of Science and Technology to
  cover the 50 of equipment, travel and
  subsistence, and other specific costs not
  funded by EU.