Inter Process Communication over High Speed Interconnects:
Highly multi-threaded applications, such as, Oracle Database Server relies heavily on the message based IPC for coordination
among processes/threads executing within and across machines. As, network message transfer requires a processor interrupt
that results in a pipeline flush; at very high data transfer speed, for such systems to be scalable, interrupt due
to message transfer should be minimized. These projects tackle this very problem.
-
Light weight IPC for Infiniband fabric, enabling low latency and high throughput communication in Oracle Database Server
-
Oracle listener enhancement for Solaris x64 over Infiniband using SDP transport
Disk I/O Scheduling for Exascale Computation
Sharing storage across multiple Databases require adequate allocation of disk bandwidth between consumers to
ensure predictable performance. This project was aimed at benchmarking TPC-C and TPC-H applications for
Oracle Exadata machines and propose a new costing model to partition disk bandwidth across consumers.
- Oracle I/O resource manager (IORM) improvements for TPC-C and TPC-H workloads
On-disk File System Management
On-disk data structures used to manage a disk greatly enhance the reliability of a magnetic storage. Following projects
implement solutions to fix any discrepancy in such data structures caused by a permanent data block failure and provide a recovery
mechanism for the data currently stored on the disk.
-
Disk repairing tool for FAT file system
-
Software driven data recovery for FAT file system
Instruction Set Architecture and Compilers
Program profiler is a very important tool for diagnosing, analysing, and enhancing performance of an applications. These projects
develop a profiler and implement an optimized unwind library for HPUX-IA64 platform.
-
Profiler for HPUX IA64 architecture
-
Optimized unwind library for HPUX IA64 architecture