Performance of a mirrored disk in a real-time transaction system
SIGMETRICS '91 Proceedings of the 1991 ACM SIGMETRICS conference on Measurement and modeling of computer systems
Teaching real time OSs with DORITOS
SIGCSE '99 The proceedings of the thirtieth SIGCSE technical symposium on Computer science education
Performance Analysis of Storage Systems
Performance Evaluation: Origins and Directions
Multi-dimensional storage virtualization
Proceedings of the joint international conference on Measurement and modeling of computer systems
Two phase locking-based algorithm with partial abort for firm deadline real-time database systems
ADBIS'97 Proceedings of the First East-European conference on Advances in Databases and Information systems
| Hi-index | 0.00 |
In this paper, we present two new disk scheduling algorithms for real-time systems. The two algorithms, called SSEDO (for Shortest Seek and Earliest Deadline by Ordering) and SSEDV (for Shortest Seek and Earliest Deadline by Value), combine deadline information and disk service time information in different ways. The basic idea behind these new algorithms is to give the disk I/O request with the earliest deadline a high priority, but if a request with a larger deadline is "very" close to the current disk arm position, then it may be assigned the highest priority. The performance of SSEDO and SSEDV algorithms is compared with three other proposed real-time disk scheduling algorithms ED, P-SCAN, and FD-SCAN, as well as four conventional algorithms SSTF, SCAN, C-SCAN, and FCFS. An important aspect of the performance study is that the evaluation is not done in isolation with respect to the disk, but as part of an integrated collection of protocols necessary to support a real-time transaction system. The transaction system model is validated on an actual real-time transaction system testbed, called RT-CARAT. The performance results show that SSEDV outperforms SSEDO; that both of these new algorithms can improve performance of up to 38% over previously-known real-time disk scheduling algorithms; and that all of these real-time scheduling algorithms are significantly better than non-real-time algorithms in the sense of minimizing the transaction loss ratio.