Alfrânio Tavares Correia Júnior
Places that I've been... Soon everything will be red. Everything will be wild.
Projects and Interests
Currently, I am a member of the GORDA project which aims to foster database replication as a means to address the challenges of trust, integration, performance, and cost in current database systems underlying the information society. In the past, I worked in the following projects:
- MobiSnap - The main goal of this project was to provide methodologies and tools to deploy relational database applications on mobile computers. Our research focused on two main areas: mobile transaction model and divergence control.
- StrongRep - The project proposed to address the problem of maintaining strongly-consistent replicated data in a large-scale setting. To achieve this goal, the project intended to combine recent research of replication management with broadcast protocols optimized for large-scale operation.
- Escada - The project aimed to study, design and implement transaction replication mechanism suited to large scale distributed systems.
- Concurrency Control and Transactional Systems.
- Synchronous Replication Protocols in relational databases.
- Group Communication and Dependability.
Once I've attended a cultural event promoted by Tribunal Regional Eleitoral, a public enterprise that deals with elections in Brazil, at which I was an employee. In such event, a guy whose face and name a barely remember started singing a song recorded by Caetano Veloso and written by Penhinha. What happened before he started singing, however, I clearly remember. He opened a book and started reading a poem. He said that Caetano once did something similar and as a huge fan, he decided to copy this, although not being as enlightened as him.
Caetano is a Brazilian guy whose talent as composer, writer, musician and singer is awesome and known abroad. I am not a singer, a writer nor a musician but and as apprentice in the field of computer science, I've decided to copy this by doing something similar to what Lamport has been doing in his website. Lamport is a famous computer science researcher whose achievements do not require further introduction as his name speaks for himself. Reading Lamport's web site I've seen nice comments on his technical reports and papers. Comments revealing not technical details but ideas and stories behind such texts. So, I decided to the same and in each text in this page one will find what happened before producing it.
- On the use of a reflective architecture to augment Database Management Systems, 2007.
This paper improves the "GORDA: An Open Architecture for Database Replication" and is an excellent reading. It will appear on the Journal of Universal Computer Science before the end of the year.
GORDA: An Open Architecture for Database Replication (Extended Abstract), 2007.
This paper was accepted on the NCA conference as an extended abstract twice. On the first year, we did not published it. However, this year we decided to do so. It was not accepted as a full paper due to the fact there were no good figures showing performance advantages while adopting the GORDA Architecture. We did not have time to run experiments on different database implementations because Susana that was responsible for the development on Derby left the project. Furthermore, as the implementation on PostgreSQL uses sockets: its performance is not good due to the overhead regarding to serialization and context switches.
- Revisiting 1-copy Equivalence in Clustered Databases, 2006.
Edward Archibald pointed out the problem described in this paper and, at the time, we thought that we were the first to notice it. Unfortunately, Gustavo Alonso mentioned about that before us. In contrast to his paper, however, we formally present the problem, possible solutions and its impact on group-based replication protocols. While studying to produce this paper, I started being interested by isolation levels. Soon, you will see results in that field.
A Pragmatic Protocol for Database Replication in Interconnected Clusters, 2006.
From Database System Community, Jon questioned several aspects of our solutions. Among such questions, we can point out the use of atomic broadcast messages and a distributed certification procedure. So, he proposed a centralized certification thus eliminating the need for atomic broadcast messages, an important optimization while replicating information among interconnected clusters. In this paper, I did not contributed so much writing but I helped find errors and discussed ideas and proofs.
- Evaluating Certification Protocols in the Partial Database State Machine, 2006.
Antonio Luis submitted this paper to DILSOS workshop held together with The First International Conference on Availability, Reliability and Security, after re-writing some parts of a technical report. This work was started 6 months right after I had arrived at University of Minho, in 2003. At that time, I did not know how to do research, write technical texts and my English was pretty bad. This does not mean that I do know how to do such things now, but it only means that my English was worse than it is and my writing even in Portuguese was awful. It was my first attempt to produce a paper. For those reasons, it was submitted and reject uncountable times. Finally, Antonio Luis decided to submit it to a non-first rate conference.
- Experimental Evaluation of Database Replication Protocols (Abstract), 2006.
Luis wrote this abstract on our research and presented it in a workshop on performability in France. He said that most of the papers presented there were about Markov chains applied to different areas in the field of computer science. In contrast to this overdose of an analytical approach, ours was hands-on.
- Testing the dependability and performance of group communication based database replication protocols, 2005.
This paper presents in detail the ideas of our simulator. It is an awesome reading and represent improvements after two successive rejects in different conferences.
- Group-based replication of on-line transaction processing servers, 2005.
This paper provides a good review on group-based replication protocols and a numerical comparison among them. This comparison uses our simulator, submitting such protocols to a TPC-C-based workload, a representative of OLTP applications. To the best of our knowledge, this is the first paper that compares such protocols and the only one with TPC-C.
Here, we have an "optimization" in the Postgres-R protocol. In the original Postgres-R, it is assumed that upon atomic delivery a transaction is put in a queue. Upon reaching the head of the queue, it should commit. Transactions might be aborted before reaching the head of the queue by concurrent transaction that conflict with them. When a decision is achieved, commit or abort, a reliable broadcast is used to notify it. In this paper, however, this decision is achieved by means of a certification procedure without waiting for a transaction to be at the head of the queue. Read our paper on serializability to understand the reasons of this optimization.
- Evaluating database replication
in ESCADA (Position Paper), 2004.
This paper shows what we learned while working on replication protocols based on group communication after two years. Some of the open-issues mentioned in the paper were used as ideas to other papers.
- Revisiting Epsilon Serializability to improve the Database State Machine (Extended Abstract), 2004.
The idea of this paper was given by Rui Oliveira, my advisor at University of Minho. While studying and writing code for this paper, I learned something quite important on weak consistency criteria. There are three approaches to handle inconsistency no matter what type of application we have in mind. First, we might assume that inconsistency is introduced and will be later handled with reconciliation techniques. Second, we might use reservation techniques. Third, we might define degrees of inconsistencies that an application might tolerate. Such degrees can be bounded (i.e., epsilon serializability) or unbounded, being the latter a special case of the former. This paper is a nice review on epsilon serializability and shows its benefits by running simple experiments. Furthermore, it shows that any work produced in this field cannot do anything different from the three choices presented above and for that reason any paper in this field should cite Calton Pul's work (Father's epsilon serializability).
We found some bugs in the simulator. Thus, experiments conducted in this paper should be re-done in our current version of the simulator.
- Avaliação de um SGBD
Replicado Usando Simulação de Redes, 2003 (in Portuguese).
When I arrived in Portugal, my colleagues were working on a database simulation system to evaluate replication protocols. In contrast to other simulation systems, however, this one would allow us to mix real implementation of replication protocols and group communication with simulated databases and different network infra-structures. This paper shows the first results with this simulation tool. In particular, my contribution in this paper is restrict to the development of a TPC-C workload generator based on the paradigm of storing and replaying. Obviously, we soon realized that this was not a good way of generating traffic.
- Executing Workflows with Mobile Agents from UML Specifications, 2003.
I've attended a pos-gradution course in Brazil, LaSiD, where I've began my first steps in the field of Distributed Systems. This work was proceduce one year after I left Brazil and uses as building blocks an extended version of an agent-based workflow management system developed by me, Sérgio and João. I should say that I own them an apology because at the time I was so busy and without further explanation I did not appear to several meetings. I won Flávio Assis, my advisor at Brazil, thanks because I did not wrote a single word in this paper but as I contributed to some ideas on the agent-based system, he decided to include me as a co-author.
- Semantic Caching for Mobile Applications, University of Minho, 2004.
This technical report was written right after I finished my Masters. I've tried to publish it once, but it was rejected. It is a good review on Semantic Caching but nothing more. It still requires lots of code to become a contribution.
- Evaluating the Performance of the Database State Machine (DBSM), University of Minho, 2004.
See the paper Testing the dependability and performance of group communication based database replication protocols.
- (Master Thesis) Distributed Transaction Processing in the Escada Protocol, University of Minho, 2004.
Most of the ideas of my PhD research came up while I was doing my Masters. Unfortunately, due to time constraints and knowledge, I did not develop such ideas during my Masters. I had one year to finish it. Of course, I did not accomplished this tight schedule. It took me one year and six months before the defense. I don't like the document but I learned so much while producing it, and that is what matter.
- Evaluating certification protocols in the partial database state machine, University of Minho, 2003.
See a paper with the same title.
- (Master Thesis) Distributed Transaction Processing in the Escada Protocol, University of Minho, 2004.
This presentation summarizes an effort of one year and six months.
- Sistemas Operativos II (in portuguese)
- Informática no Ensino (in portuguese) - (additional material)
- Sistemas Operativos I (in portuguese) - (additional material)
- Conceitos de Sistemas Informáticos (in portuguese) - (additional material)
- EUROPAR 2007, January 26. (France)
- EuroSys 2007, February 8. (Lisbon)
- NCA 2007, February 13. (Boston, USA)
- ADBIS 2007, February 21. (Bulgaria)
- Dexa 2007, February 28. (Germany)
- LADC 2007, March 20. (Mexico)
- VLDB 2007, March 14,21. (Austria)
- SRDS 2007, April 14,21.
- SBBD 2007, April 15 (Joao Pessoa, Brasil)
- PRDC 2007, May 15
- EuroSys, IPDPS, ACM SIGMOD, ICDE, DSN, ARES, ICDCS
Last modified 2008-05-26 09:41 PM