----------------------------------------- < SANKARAN SIVATHANU > < 156, Merritts Avenue NW, > < Atlanta, GA 30313 > < Email : sankaran@cc.gatech.edu > < Ph: (404) 418 - 8188 > < http://www.cc.gatech.edu/~sankaran > ----------------------------------------- RESEARCH INTERESTS =================== File and Storage Systems, Virtualization and Operating Systems EDUCATION ========== PhD Student - Computer Science Aug 2007 - Present Advisor: Prof. Ling Liu College of Computing Georgia Institute of Technology, Atlanta, GA Bachelor of Engineering Aug 2003 - May 2007 Computer Science & Engineering, Anna University, Chennai, INDIA. WORK EXPERIENCE =============== Graduate Research Assistant Aug 2007 - Present School of Computer Science, Georgia Institute of Technology (1) Pro-active Storage Systems Designed and implemented a new class of storage systems which facilitates a novel mode of interaction between the storage subsystem and higher layers of the system I/O stack. Instead of viewing the storage system as a passive entity that simply responds to higher level requests, we proposed a model where the storage system can pro-actively initiate control over the higher layers by making use of its internal knowledge (2) Storage Performance Issues in Virtualized Cloud Environments Measured and analyzed storage performance in virtual computing environments through extensive experiments under different workload scenarios. The experiments were aimed at identifying issues that must be considered when managing storage in virtualized infrastructure cloud environments like the Amazon EC2, in order to achieve increased end-user performance. (3) Opportunistic Data Flushing using Pro-active Storage Systems Flushing of dirty pages of data from file system cache is done periodically or during cache reclamation. However, background flushing of data can interfere with more important foreground requests adversely. We used pro-active storage system to detect idle periods in storage during which the dirty pages are "pulled" from the file system cache, thus improving the overall disk utilization. (4) Disk-initiated Data Pre-fetching Modern disks store track full of information in its track-buffer even when a small subset of it is requested by the higher layer. This improves performance by avoiding rotational delays when data from same track is requested next. However, because of smaller size of the buffer, they get evicted very frequently. We used pro-active storage system to detect importance of track buffer contents and promote useful data to higher-level file system cache (which is usually many times larger) during eviction. (5) Idle read-after-write [Work-in-progress] Buggy disk controllers may return successful write operation even without actually writing, or writing a different data, or writing to a different location. To detect and correct these errors, enterprise disks adopt an approach where every write operation is accompanied by a read operation that checks for data correctness. As every write incurs an extra read operation performance is traded-off for data reliability. We use pro-active storage to store those blocks in the file system cache and "pull" them to check whenever the disk is idle. (6) Storage Cache Isolation for Virtual Machines [Work-in-progress] Host memory is partitioned across multiple VMs in an isolated manner. However, the storage cache is still shared across VMs and current storage systems do not enable fair partitioning of its cache. This leads to disproportional usage of storage cache by individual VMs. The goal of this project is to enable proportional sharing of storage cache across VMs. Member of Technical Staff - Intern May 2010 - Aug 2010 (R&D Storage Performance) VMware Inc. Palo Alto, CA Context-Aware Replay of I/O Traces Designed and implemented an I/O trace replay framework that faithfully replays I/O traces in VMware ESX platform. The highlight of this development is the trace replay being context-aware: the replay adapts itself according to the capabilities of the system hardware rather than blindly following original trace characteristics, which is instead a function of hardware in trace-collection system. Research Intern May 2009 - Aug 2009 (Grid Storage) NEC Labs America, Princeton, NJ Performance and Energy Modeling of Storage Systems Given an I/O trace and system parameters, our model would compute the approximate bandwidth, latency and energy consumption values over the timeline of the trace. Our model is extensible to a variety of storage layouts - we tested our model against an accurate simulation of well-known energy conservation scheme called PARAID. Summer Intern May 2008 - Aug 2008 (Advanced Technology Group) NetApp Inc. Sunnyvale, CA Correlation between file systems and energy consumption Analyzed the impact of various types file systems and workloads on energy consumption of NetApp storage servers. Different macro benchmarks were used and a wide range of local and network file systems were considered in this study. Graduate Teaching Assistant Jan 2009 - May 2009 College of Computing, Georgia Institute of Technology Atlanta, GA Graduate Level: Advanced Operating Systems Undergraduate Level: Design of Operating Systems, Emerging Database Technologies OTHER SELECTED PROJECTS ======================== o Design and implementation of content-based file system in Linux Kernel o Implementation and performance analysis of PAXOS algorithm for distributed consensus among un-trusted nodes. PUBLICATIONS ============= 1. Modeling the Performance and Energy of Storage Arrays, Sankaran Sivathanu, Crisitian Ungureanu, Ling Liu IEEE International Conference on Green Computing (IGCC '10) 2. Storage Management in Virtualized Cloud Environments Sankaran Sivathanu, Ling Liu, Mei Yiduo, Xing Pu IEEE International Conference on Cloud Computing (Cloud '10) 3. Understanding Performance Interference of I/O workload in Virtualized Cloud Environment Xing Pu, Ling Liu, Mei Yiduo, Sankaran Sivathanu, Youngyun Koh, Calton Pu IEEE International Conference on Cloud Computing (Cloud '10) 4. Performance Measurements and Analysis of Network I/O Applications in Virtualized Cloud Environment Mei Yiduo, Ling Liu, Xing Pu, Sankaran Sivathanu IEEE International Conference on Cloud Computing (Cloud '10) [Under Review] 1. CART: Context-Aware Replay of I/O Traces Sankaran Sivathanu, Jinpyo Kim, Devaki Kulkarni, Ling Liu 2. Pro-Active Disks Sankaran Sivathanu, Ling Liu AWARDS AND HONORS ================== o Student Travel Grant for USENIX FAST 2010 Conference o 1st Rank Holder in CS Department at St. Peter's Engg. College, Anna University, 2003-07 RELEVANT SKILLS ================ o Linux Kernel Programming o Languages: C, C++, Perl, Matlab o Packages: DiskSim Simulator, Intel ISCSI driver o Operating Platforms: Linux, Xen Hypervisor, VMware ESX, Solaris, Windows PROFESSIONAL ACTIVITIES ======================== o Externally reviewed papers for, VLDB 2010 ICDE 2011 o IEEE Member o Co-organized a national-level technical symposium in CS department during undergrad REFERENCES =========== Available upon request.