1 / 24

A Survey of I/O Optimization Techniques

A Survey of I/O Optimization Techniques. Sven GROOT Kitsuregawa Laboratory December 7th 2007. Background. Increase in CPU and memory speed not matched by disk drives Increasing disk and data sizes Disks limited by mechanical components: hard to improve We must optimize I/O accesses.

jennis
Download Presentation

A Survey of I/O Optimization Techniques

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. A Survey of I/O OptimizationTechniques Sven GROOT Kitsuregawa Laboratory December 7th 2007

  2. Background • Increase in CPU and memory speed not matched by disk drives • Increasing disk and data sizes • Disks limited by mechanical components: hard to improve • We must optimize I/O accesses.

  3. Outline • Hard disk drives • Evaluating Optimizations Through the I/O Path [Riska et al, 2007] • File System Level • Device Driver Level • Disk Level • Prefetching • Competitive Prefetching[Li et al, 2007] • DiskSeen[Ding et al, 2007]

  4. Hard Disk Drives

  5. Hard Disk Drives Bottleneck! Disk rotation Track Sector Bottleneck! Head movement Disk head Delay = Head Seek Time + Rotational Latency

  6. The I/O Path [Riska et al, 2007]

  7. Evaluation Environment • Postmark file system benchmark • Measure transactions • Each transaction has two steps: create file or delete file and read file or append file.

  8. File Systems • Ext2 • Block/cylinder groups • Single, double, or triple indirect metadata blocks • Ext3 • Compatible with Ext2 • Journaling filesystem • ReiserFS • Metadata in B+ trees • Journaling filesystem • XFS • Extent-based B+ trees. • Journaling filesystem • Allocation groups

  9. File System Throughput • Reiser performs best • Efficient request merging • Efficient block allocation • Not much journaling overhead

  10. DeviceDriver Level • Requestreordering/merging • Elevator algorithm • Sweep the disk, process all requests when the head passes the location • Shortest Seek Time First • Always process closest request • May lead to starvation

  11. I/O Schedulers • No-Op • First-Come First-Serve algorithm; no reordering • Deadline • SSTF with aging to prevent starvation • Anticipatory (default) • Similar to deadline • Waits for better request under some circumstances • CFQ • Elevator algorithm • Gives each process equal I/O time

  12. Scheduler Throughput • All outperform No-Op • Deadline performs best • No deceptive idleness in Postmark

  13. Disk Drive Level • Requestreordering • Disk drives used:

  14. Disk Drive Results – Throughput

  15. Prefetching • Read data expected to be needed in the future • Prefetching reduces number of I/O switches between concurrent data streams • Optimal strategy: read exactly the data needed • Requires a-priori knowledge of the stream size • Aggressive prefetching: large prefetching depth • May fetch unnecessary data • Conservative prefetching: small prefetching depth • Too many I/O switches

  16. Competitive Prefetching[Li et al,2007] • Prefetching depth: data that can be read during average I/O switch time • Guarantees time taken no more than twice that of optimal off-line strategy • Must measure I/O switch time and transfer rates

  17. CompetitivePrefetching – Results

  18. DiskSeen[Ding et al, 2007] • Problem: file level prefetching has disadvantages • File level sequentialitymaynotbepreserved at disk level • Inconvenientforrecordingaccessinformation • Inter-file sequentialitynotexploited • File metadata blocksnotprefetched • Solution: Block level prefetching • Uses disk logicalblocknumbers • Works next to file level prefetcher

  19. DiskSeen • Sequence detection • Global counter, incremented every block access • Current counter value for block stored: access index • Sequence when access indices on sequential blocks grow uniformly • Prefetch when sequence detected

  20. DiskSeen • History based prefetching • Keeps limited history of past access indices • Look for trails from current block • Unlike sequences, trails can skip blocks or go backwards • When history trail found, prefetch trail blocks 52002 52001 85000 74000 85001 85010 N/A N/A 63110 63111 63200 63290 N/A N/A 52000 48550 43501 43510 N/A N/A 43500 34950 35000 37000 B’3 B’2 B1 B2 B3 B4

  21. DiskSeen OS Caching Area 2 4 Block access information Prefetched blocks On-demandread File-level prefetch Prefetching Move hit blocks Delayedwrite-back Prefetching Area 1 5 3 Hard Disk

  22. DiskSeen – Results

  23. DiskSeen – Results CVS Benchmark BeforeDiskSeen AfterDiskSeen

  24. Conclusion • Disk performance remains bottleneck • Effective optimization opportunities exist at many levels • Active area of research • FS2 [Huang et al, 2005], Preemptive Scheduling [Dimitrijevic et al, 2005], Distributed File Systems [e.g. Weil et al, 2006], Idletime Scheduling [Eggert et al, 2005], etc. • Room for improvement • Use of application/domain knowledge in schedulers/prefetchers • Anticipatory scheduler improvements • Scheduling at disk level

More Related