[1] Quality of Service Support for Real-Time Storage Systems

[1] Quality of Service Support for Real-Time Storage Systems

Dimitrijevic, Z. and R. rangaswami (2003). Quality of Service Support for Real-Time Storage Systems. Proceedings of the International IPSI-2003 conference, citeseer.

ArticleNotes:

What are the requirements of applications for storage QoS?

First, some applications require latency for storage for some I/O requests, while others require best effort, which requires stable transmission rates for some applications, that is, the throughput requirement and the demand for continuous application supply.

Real-time applications have these requirements for storage systems

1. differentiate different I/O requests of services

2. Ensure the response time for some I/O requests

3. Some Applications ensure their throughput (that is, they do not care about the time of a single I/O request or the overall throughput)

4. Ensure low latency and high throughput for applications that are best served.

Figure 1 requirements of real-time applications for Storage Systems

There are many studies on disk scheduling, including FCFS, sstf, satf, scan, and csan. These scheduling methods are best effort service models,

FCFS is very simple, but due to different I/O requests, the disk seek time varies greatly. FCFS does not optimize the disk seek time, resulting in low disk bandwidth utilization.

Sstf and satf improve disk bandwidth utilization by optimizing disk seek. However, this method cannot avoid hunger. The latency of a single I/O request is not limited.

To solve this problem, scanAlgorithmAnd the csan algorithm were proposed to solve the hunger problem, but the maximum latency of a single request may still be large, at this time, the operating system usually limits the length of the disk queue to ensure the maximum latency of a single request.

There are also some scheduling for real-time I/O requests, such as EDF, SCAN-EDF, etc.

There is also a category of hard-to-service I/O and real-time I/O types, such as GSS and cello.

Figure 2: Various disk scheduling algorithms