2005 Fiscal Year Final Research Report Summary
Estimation of Temporal Behavior in Distributed Systems
| Project/Area Number |
15500053
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Computer system/Network
|
|---|
| Research Institution | Waseda University |
|---|
Principal Investigator |
NAKAZATO Hidenori Waseda University, Graduate School of Global Information and Telecommunication Studied, Professor, 大学院・国際情報通信研究科, 教授 (30329156)
|
|---|
| Project Period (FY) |
2003 – 2005
|
| Keywords | Real-Time Systems / Distributed Processing |
|---|
| Research Abstract |
We proposed a network architecture for soft-real-time distributed systems. In the architecture, priority scheduling with fixed number of discrete priority levels available in VLAN header in Ethernet packets are used in LAN section of the network. Packet scheduling in WAN section of the network and processor scheduling adopt EDF scheduling. We proposed a priority assignment algorithm for LAN section based on delay constraint in one-hop LAN section which is determined from delay allocation to WAN section and processors, and on distribution of packet priority assignment in LAN section. As processor scheduling, we analytically derived estimation method for response time in EDF scheduling. We also proposed a packet scheduling algorithm called ORC-GPS for WAN section. The algorithm can utilize network efficiently while controlling delay. ORC-GPS provides larger schedulable region than both EDF and GPS. ORC-GPS is a class-based scheduling algorithm. Parameters for scheduling are determined on required maximum delays for classes. Since delay control by priority assignment in LAN section has only limited effect and we cannot vary delay from certain value, delay is mainly controlled in WAN section while delay control by priority classis performed in LAN section. ORC-GPS can be applied to this WAN section delay control. ORC-GPS parameters are determined from the residual delay derived by subtracting delay in LAN section from overall permissible network delay. In this way, the target end-to-end delay canbe realized.
|
