Coerced Cache Eviction and Discreet Mode Journaling: Dealing ...

Coerced Cache Eviction and Discreet Mode Journaling: Dealing ...

Coerced Cache Eviction and Discreet Mode Journaling: Dealing with Misbehaving Disks Abhishek Rajimwale, Vijay Chidambaram, Deepak Ramamurthi Andrea Arpaci-Dusseau, Remzi Arpaci-Dusseau WSRR 11 1 Ordering points are essential All modern file systems depend on ordering points Journaling file systems (ext3, ext4) Data before metadata Copy on write file systems (ZFS) Transaction data before uber-block

File systems flush the disk cache to ensure correct ordering of writes WSRR 11 2 Problem: Flushing doesnt work Disks can fail to flush data upon request One reason: Bugs Errors in the storage stack [FAST 08] Improper propagation of error codes Inadequate failure policies [SOSP 05] Bugs in the firmware [SOSP 03]

WSRR 11 [FAST 08] 3 A v g t im e ( m s e c ) Disks can lie! Caching improves performance File systems are usually blamed for data loss 50 45 40 35 30 25 20 15 10 5 0

4k Sequential writes w/ cache w/o cache 16k 64k 128k 512k 1m Write size From the fcntl man page in Mac OSX: F_FULLFSYNC Does the same thing as fsync(2) then asks the drive to flush all buffered data to the permanent storage device (arg is ignored). This is currently implemented on HFS, MS-DOS (FAT), and Universal Disk Format (UDF) file systems. The operation may take quite a while to complete. Certain FireWire drives have also been known to ignore the request to

flush their buffered data. WSRR 11 4 Summary We present Coerced Cache Eviction We show how to characterize a disk cache We implement CCE in the ext3 file system We find that CCE adds low overhead for most workloads WSRR 11 5 Outline Motivation

Background Coerced Cache Eviction Cache Fingerprinting Discreet Mode Journaling Evaluation Conclusion WSRR 11 6 Background - Journaling Memory M C D Disk cache Disk surface WSRR 11 7

Outline Motivation Background Coerced Cache Eviction Cache Fingerprinting Discreet Mode Journaling Evaluation Conclusion WSRR 11 8 Coerced Cache Eviction

Ensures that cache has been truly flushed Key idea: Write a flush workload to disk which replaces the target blocks in the disk cache, flushing target blocks to disk WSRR 11 9 Coerced Cache Eviction Memory M C D F F F F F

F F F F Disk cache Disk surface WSRR 11 10 Coerced Cache Eviction Desired properties: Low performance overhead High probability of flushing target blocks Need to understand the disk cache to design good workload WSRR 11 11

Outline Motivation Background Coerced Cache Eviction Cache Fingerprinting Discreet Mode Journaling Evaluation Conclusion WSRR 11 12 Cache Fingerprinting Manufacturers dont expose details about disk

caches Disk caches can vary in: Read/Write partition size Number of segments Replacement policy Poorly characterized in literature WSRR 11 13 Cache Fingerprinting Flush micro-benchmark: Write target block Write varied flush workload measure fsync Read target measure

Vary in each workload: Number of writes Amount of data in each write Sequential/Random writes WSRR 11 14 Cache Fingerprinting For each workload Eviction fingerprint Probability of eviction is visually shown Darker region indicates higher probability

Performance fingerprint Time taken to write flush workload Darker region indicates higher time to flush WSRR 11 15 Cache Fingerprinting Optimal workload selection Highest probability of eviction Lowest cost For some disks, sequential writes may be

ineffective at flushing A number of random writes are required WSRR 11 16 Cache Fingerprinting Manufactur er Cache (MB) Number of writes Data (MB) Time (ms) Hitachi

8 1 2.38 50 Hitachi 32 1 11 87 Samsung 8 128 49

1328 Samsung 16 512 128 2872 Western Digital 16 1792 19 5107 Western Digital

64 256 1 7705 Seagate 8 256 31 870 Seagate 16 128 17

342 Seagate 32 128 37 396 WSRR 11 17 Outline Motivation

Background Coerced Cache Eviction Cache Fingerprinting Discreet Mode Journaling Evaluation Conclusion WSRR 11 18 Discreet Mode Journaling Incorporating CCE into ext3 Fingerprint the disk to find optimal flush workload Create flush zone with suitable size Modify ext3 to issue flush zone writes at each ordering point WSRR 11 19 Outline

Motivation Background Coerced Cache Eviction Cache Fingerprinting Discreet Mode Journaling Evaluation Conclusion WSRR 11 20 Evaluation Performance of Discreet Mode Journaling is comparable to regular journaling for most workloads such as Postmark, Filebench webserver and

openSSH Postmark Time (s) 1600 1400 1200 1000 800 600 400 200 0 regular w/o cache discreet regular w/ cache WSRR 11 21 Conclusion

Disk drives misbehave with respect to flush commands Coerced Cache Eviction provides a way to run file systems reliably on top of such disks, with low overhead Trust in disk is weakening Previously flushing was trusted Cloud computing systems Can virtualized WSRR 11 coercion be more 22 hardware be trusted? Will Thank you! WSRR 11

23 Evaluation Workloads which call fsync frequently perform poorly Each fsync causes 4 flushes Filebench varmail writes a small amount of data and calls fsync repeatedly We did a number of optimizations to improve the situation: Transactional Checksum removes 1 CCE Use optimized CCE instead of generic CCE Incorporate Group Commit in varmail WSRR 11 24 Evaluation

Performance of Filebench Varmail after optimizations: 35 Throughput ()mb/s 30 25 regular w/o cache discreet (costly flush) 20 15 10 5 0 Data Ordered Journal type

WSRR 11 25

Recently Viewed Presentations

  • SUPREME COURT CASES - Winston-Salem/Forsyth County Schools

    SUPREME COURT CASES - Winston-Salem/Forsyth County Schools

    Gregg v Georgia (1976) Regents of the University of California v. Bakke (1978) New Jersey v. T.L.O (1985) Bethel School District v Fraser (1986) Hazelwood School District v. Kuhlmeier (1988) Texas v. Johnson (1989) Webster v. Reproductive Health Services (1989)
  • Dia 1 - BIJBELONDERZOEK

    Dia 1 - BIJBELONDERZOEK

    Arial Maiandra GD Trebuchet MS Times New Roman dingbats-Thin VAGRoundedBold Wingdings 3 Arial Black Eggman Impact EklekticNormalLight Standaardontwerp Dia 1 Dia 2 Dia 3 Dia 4 Dia 5 Dia 6 Maleachi 1:6 Genesis 15:2 Dia 9 Dia 10 Genesis 17:1,2...
  • 1.2 "Bringing the future of the church into the present so ...

    1.2 "Bringing the future of the church into the present so ...

    (Ephesians 1:22-23) 2. BIBLICAL TEACHING We desire to teach God's word with integrity and authority so that seekers find Christ and believers mature in Him (II Tim. 3:16) CrossRoads Nazarene www.crossroadsnazarene.org What are the Values of a Healthy Church?
  • CIHR at Queen's - Memorial University of Newfoundland

    CIHR at Queen's - Memorial University of Newfoundland

    Frederick Banting and Charles Best Canada Graduate Scholarships - Doctoral Awards (CGSDA) CIHR Doctoral Research Awards (CIHR DRA) To note: All doctoral students who want to be considered for a Vanier AND a CIHR CGS award, must submit an application...
  • Imagery - Middlesex County Vocational and Technical Schools

    Imagery - Middlesex County Vocational and Technical Schools

    The imagery used the least was gustatory and auditory, which were used about 2 times each. This was mostly because imagery was better in conveying how bad the flames and floods of God are. People are more scared of what...
  • 2016 Training Conference - WCDA

    2016 Training Conference - WCDA

    Another IRS form. Reports specific non-compliance to the IRS. Shows the out-of-compliance date. Shows the back in-compliance date. IRS requires we send it for all non-compliance issues; but only the IRS has the option to pursue an audit or not.
  • Presentación de PowerPoint - OAS

    Presentación de PowerPoint - OAS

    Title: Presentación de PowerPoint Author: ffigueroa Last modified by: jdiazd Created Date: 9/5/2011 4:01:28 PM Document presentation format: Presentación en pantalla (4:3)
  • The External Sector - Mark Ellyne

    The External Sector - Mark Ellyne

    The External Sector. Goals: Understand the balance of payments accounts; How they reflect real and fiscal sector activities ,and. Be able to forecast key elements of the BoP.