用户:Hengfeng-Wei

来自Algorithm Wiki
Hengfeng-Wei讨论 | 贡献2018年1月6日 (六) 16:51的版本 CSS for Jupiter

跳转至: 导航搜索



Profile

Whf-white-2016-12.jpg

Hengfeng Wei (魏恒峰) cv (en) cv (ch)

  • Research Assistant (助理研究员):From 2017-01
 Institute of Computer Software                          (软件所)
 State Key Laboratory for Novel Software Technology      (软件新技术国家重点实验室)
 Department of Computer Science and Technology           (计算机科学与技术系)
 Nanjing University                                      (南京大学)
 Addr: 163 Xianlin Road, Nanjing 210023, P. R. China     (江苏省南京市栖霞区仙林大道163号南京大学仙林校区 210023)
 Office: 302
 Mailbox: H016
 Email: hfwei-nju-edu-cn

Research Interests

I am interested in Distributed Computing Theory and Formal Methods.

Publications

dblp: Hengfeng Wei scholar: Hengfeng Wei


Algebraic Operational Transformation

 Work in Progress.

More details and comments are coming when the work is ready.


CJupiter

 Work Under Review => Rejected from PPoPP'2018 => Under Revision for PODC'2018

This work is inspired by the work on "Specification and Complexity of Collaborative Text Editing" published in PODC'2016 by Hagit Attiya et al.

More details and comments are coming when the work is ready.


Parameterized and Runtime-tunable Snapshot Isolation in Distributed Transactional Key-value Stores

 Hengfeng Wei, Yu Huang, Jian Lu.
 Parameterized and Runtime-tunable Snapshot Isolation in Distributed Transactional Key-value Stores.
 In Proc. of the 36th International Symposium on Reliable Distributed Systems (SRDS), pp. 21--33, Hong Kong, Chain, Sep.2017.
 [Paper:RVSI (not camera-ready)]

The formal specification of RVSI (Relaxed Version Snapshot Isolation) in this work is partly inspired by the work on "Relaxed Currency Serializability for Middle-Tier Caching and Replication" published in SIGMOD'2006 by Philip A.Bernstein et al.

It is the first time we use Aliyun in our experiments. This will be the normal case in the future.

Important Lessons Learnt:

  • Solve the real problems in realistic ways!
 One can always find made up examples of ..., but it is hard to convince that .... 
 This paper does not go beyond ....                          --- From Reviewers
  • System research greatly matters!
 The following is all about the experimental section, which is probably the weakest part of the paper.
                                                             --- From Reviewers

Probabilistically-Atomic 2-Atomicity: Enabling Almost Strong Consistency in Distributed Storage Systems

 Hengfeng Wei, Yu Huang, Jian Lu. 
 Probabilistically-Atomic 2-Atomicity: Enabling Almost Strong Consistency in Distributed Storage Systems. 
 In IEEE Trans. Comput. (IEEE Transactions on Computers), 66(3):502--514, doi:10.1109/TC.2016.2601322, March 2017.
 [abstract@IEEE]  [Paper: PA2AM@TC'2017] [Paper: PA2AM@arXiv (not up-to-date)]

Last night (2016-10-13), I reread the paper "On Interprocess Communication --- Part II: Algorithms" of Leslie Lamport, and found that in Proposition 5, Lamport has proved that in the single-writer model a regular register is atomic if two successive reads that overlap the same write cannot obtain the new then the old value. The phenomenon that "two successive reads that overlap the same write cannot obtain the new then the old value" is exactly the old-new inversion anomaly. Therefore, Lamport has shown that a regular register is atomic if it does not allow old-new inversion anomalies. This is very similar to Theorem 1 in our paper which essentially states that the PA2AM algorithm implements a 2-atomic register and the old-new inversion anomaly is the only cause of atomicity violation.

The major difference between Proposition 5 of Lamport and Theorem 1 in our paper is that Proposition 5 is at the specification level while Theorem 1 is at the implementation level.

Combining Proposition 5 and Theorem 1, we obtain

Conjecture 1
In the single-writer model, 2-atomicity is equivalent to regularity.
Conjecture 2
In the single-write model, PA2AM implements a regular register.

Verifying Pipelined-RAM Consistency over Read/Write Traces of Data Replicas

 Hengfeng Wei, Marzio De Biasi, Yu Huang, Jiannong Cao, and Jian Lu. 
 Verifying Pipelined-RAM Consistency over Read/Write Traces of Data Replicas.
 In IEEE Trans. Parallel Distrib. Syst., 27(5):1511--1523, May 2016, doi:10.1109/TPDS.2015.2453985
 [abstract@IEEE]  [Paper: VPC@TPDS'2016] [Paper: VPC@arXiv(not up-to-date)]
 SCI: DJ5HS

In the Conclusion section, we discussed the VCC (Verifying Causal Consistency) problem and wrote: "Because Pipelined-RAM is a weakening of causal consistency, our NP-complete result also applies to the general problem of verifying causal consistency".

I am sorry to say that it is not right for us to conclude that VCC, in general, (i.e., VCC-SD in our terms) is NP-complete just because Pipelined-RAM is a weakening of causal consistency.

However, I think that VCC-SD is NP-complete and that the basic idea of the polynomial reduction used in the NP-complete proof for VPC-SD will be useful in proving the NP-completeness of VCC-SD.

The NP-completeness proof for VPC-SD (along with VPC-MD) is mainly credited to Marzio De Biasi, the second author of this paper.


Formal Specification and Runtime Detection of Temporal Properties for Asynchronous Context

 Hengfeng Wei, Yu Huang, Jiannong Cao, Xiaoxing Ma, Jian Lu. 
 Formal Specification and Runtime Detection of Temporal Properties for Asynchronous Context. 
 In Proceedings of the 10th IEEE International Conference on Pervasive Computing and Communications 
 (IEEE PerCom '12), pages 30--38, 2012.
 [abstract@IEEE]  [Paper: CTL3@PerCom'2012]
 EI: 20122315082503

Projects

  • Jupiter Simulator [In Plan]
  • TLA+ Book Translation [Work in Progress@Bitbucket]
    • Please contact me if you are also interested in it.
  • LaTeX Editor: Vim-LaTeX Improved
    • Project structure
    • Label autocompletion and renaming
  • TikZ Editor (I am very interested in this project.)
    • Detexify for TikZ
    • (Better) Code recommendation

Links