1. Input and Output Optimization in Linux for Appropriate Resource Allocation and Management James Avery King

2. Abstract  There is one evident area of operating systems that has enormous potential for growth and optimization. Only recently has focus been put on upgrading resources in the input/output (I/O) mechanisms of Linux operating systems.  Energy and time allocation are integral parts to consider during computation  Different advances must be made  Higher throughput during run-time can be maintained through coupling  With the advent of advancements in different facets of I/O in Linux, the operating system can be utilized for more optimized potential  Setback of research: advancements typically are incremental

3. Introduction  General Input and Output  Two Primary Conditions for I/O to overcome:  Developers prefer existing types of I/O hardware and software  Three primary classifications of hardware  Synchronous I/O  Asynchronous I/O  Two Basic Functions of Schedulers  Unique aspect of Linux concerning I/O

4. Cooperative I/O  Minimized time in active state by 60%  Common limitation of I/O in Linux: energy  Varying threshold for state transitions  To solve energy limitation, the team from University of Erlangen created a new interface specifically for batching I/O requests  Resulted in more concentrated periods of active state

5. Cooperative I/O Experiments  First Experiment: used read operations and variable time constraints for the idle state of the computer to be implemented  Resulted in lower frequency of mode switches  Second Experiment: used varying lengths of state and idle time constraints for write operations  Proved Linux update did not match power saving requirements

6. I/O Burstiness for Energy Conservation (IBEC)  Problem: storage devices account for almost 27% of total energy consumption of computing systems  Solution: aggregate similar I/O requests into larger contiguous blocks of requests when the disk is active

7. I/O Burstiness for Energy Conservation (IBEC) Experiment  Compare IBEC with three prominent strategies (Earliest Deadline First and two of its modifications) in terms of power consumption by the disk  Results: IBEC reduced power consumption of real-time embedded disk systems by up to 60%  Increased longevity of battery life in Linux system  Minimized sum of power state transitions in the hard disk while it was executing a request-stream

8. Scalable I/O Forwarding  Aggregated methods on I/O requests  Designed I/O node in conjunction with compute node  Alleviated compute nodes of a majority of their burden of I/O handling  Allowed for circumventing the lack of direct connectivity to the file system for the compute nodes

9. Enhancements to I/O in Linux from IBM Linux Technology Center and University of Texas  Four pre-defined I/O schedulers in Linux  User selects one based off of workload  Cooperative Anticipatory Scheduler built on already existing I/O Scheduler  Experiment:  Implemented Program A and Program B in several different scenarios to test I/O scheduler’s ability to execute synchronous read requests by a single process and a sequence of chunk read requests, each of which was generated by a different process respectively  Results: CAS could run up to 62% faster in terms of run-time

10. Conclusion  Common denominator in first three sources for evidence: aggregation  Fourth source built on foundation of already used I/O scheduler  Diversity is lacking in the advancement of not just I/O scheduling in Linux, but the entirety of the operating system  Purpose of using these sources: show that a majority of work being done is original in some ways but built on the same foundation of knowledge  Until researchers realize that they are only incrementally solving problems there will be no profound breakthroughs in I/O mechanisms of the Linux operating system