| Popis: |
This paper introduces a scalable I/O model of Lustre file system and propose a dynamic I/O congestion control mechanism to support the incoming exascale HPC systems. Under its control, clients are allowed to issue more concurrent I/O requests to servers, which optimizes the utilization of the network/server resources and improves the I/O throughput, when servers are under light load; on the other hand, it can throttle the clients' I/O and limit the number of I/O requests queued on the server to control the I/O latency and avoid congestive collapse, when the server is overloaded. The results of series of experiments demonstrate the effectiveness of our congestion control mechanism. It prevents the occurrence of congestive collapse and on this premise it can maximize the I/O throughput for the scalable Lustre file system. |
