咖啡香

Klaus Ma

Team leader, system architect, designer, software developer with 10+ years of experience across a variety of industries and technology bases, including cloud computing, machine learning, bigdata and financial services.

Founding Volcano & kube-batch, Kubernetes SIG-Scheduling co-Leader, CNCF Research User Group & SIG-Runtime Tech Lead. Global Team Lead of IBM Spectrum Symphony CE & L3. Currently, Architect, R&D, Huawei Technologies.

Experience

Huawei, Beijing, China (2018 ~ now)

Founding Volcano & kube-batch; architect of Batch Container Service of Huawei Cloud, lead ~10 size of team to build cloud service for batch workload, including AI, BigData, Gene and so on.

IBM, Beijing, China (2015 ~ 2018)

Open Source Developer of Spectrum Conductor

Kubernetes:

Mesos Contributor, several patches and major proposals:

IBM, Beijing, China (2014 ~ 2015)

Team Lead of Spectrum Symphony L3

Lead ~5 size of team on critical customer issues handling, by working with global Product manager and Support team, to meet business priorities in time

IBM, Beijing, China (2010 ~ 2014)

Team Lead of Spectrum Symphony CE

Lead ~10 size of team on agile development for new requirements, by working with global Product manager, to meet business priorities in time.

Baidu, Beijing, China (2008 ~ 2010)

R&D of PS department; focus on coverage of spider/crawler.

R&D of IT department; focus on “workload form engine”, “user management system” and so on.

Contacts

E-mail: klaus1982.cn@gmail.com; Github: @k82cn; Linkedin: k82cn

Education

Presentation

Publication

Customized Plug-in Modules in Metascheduler CSF4 for Life Sciences Applications

New Generation Computing March 9, 2008

Authors: Zhaohui Ding, Xiaohui Wei, Yuan Luo, Da Ma, Peter Arzberger, Wilfred Li

As more and more life science researchers start to take advantages of grid technologies in their work, the demand increases for a robust yet easy to use metascheduler or resource broker. In this paper, we have extended the metascheduler CSF4 by providing a Virtual Job Model (VJM) to synchronize the resource coallocation for cross-domain parallel jobs. The VJM eliminates dead-locks and improves resource usage for multi-cluster parallel applications compiled with MPICH-G2. Taking advantage of the extensible scheduler plug-in model of CSF4, one may develop customized metascheduling policies for life sciences applications. As an example, an array-job scheduler plug-in is developed for pleasantly parallel applications such as AutoDock and Blast. The performance of the VJM is evaluated through experiments with mpiBLAST-g2 using a Gfarm data grid testbed. Furthermore, a CSF4 portlet has been released to provide a graphical user interface for transparent grid access, with the use of Gfarm for data staging and simplified data management. The platform is open source at sourceforge.net/projects/gcsf/ and has been deployed in life science gateways by projects such as My WorkSphere, and PRAGMA Biosciences Portal. The VJM enables the development of support for more sophisticated workflows and metascheduling policies in the near future.

A Virtual Job Model to Support Cross-Domain Synchronized Resource Allocation

Journal of Software

Authors: Zhaohui Ding, Xiaohui Wei, Da Ma

Although more and more scientists start to take advantages of grid technologies to facilitate their researches, running parallel jobs crossing domains in a grid environment is still a challenge. Even MPICH-G2 is able to run MPI applications on across domain resources, however, the resource allocations are not synchronized which will cause dead lock and other serious problems. In this paper, we introduced a virtual job model (VJM) which achieves synchronized cross-domain resource allocation for parallel grid applications. VJM is able to prevent the resource allocation deadlock caused by multiple parallel jobs competing resource, and alleviate the resource waste by backfilling small jobs. VJM can work with almost all kinds of local schedulers via standard Grid Resource Allocation and Management (GRAM) protocol as it does not depend on resource reservation. We have implemented VJM in meta-scheduler CSF4 and validate the rationality of VJM by mpiBLAST-g2, a parallel bioinformatics application.

Skills

Kubernetes, Mesos, HPC, PMP (#1684623), C/C++, Java, Golang


comments powered by Disqus