Upload
stanley-harvey
View
215
Download
2
Embed Size (px)
Citation preview
SATA ExperiencesSATA Experiences
HEPiX Spring 2005HEPiX Spring 2005
FermilabFermilabRay Pasetes, Lisa GiacchettiRay Pasetes, Lisa Giacchetti
Type Of StorageType Of Storage
EnterpriseEnterprise Typically SCSI or Fibre ChannelTypically SCSI or Fibre Channel Very expensive, high performanceVery expensive, high performance
Mid-TierMid-Tier Enterpise SATA, FATAEnterpise SATA, FATA Higher MTBF than commodityHigher MTBF than commodity 90-95% streaming performance of Enterprise drives90-95% streaming performance of Enterprise drives
CommodityCommodity SATA, ATA, IDESATA, ATA, IDE 90-95% streaming performance of Enterprise drives90-95% streaming performance of Enterprise drives
Types of SATA storageTypes of SATA storage
FC Controllers, SATA storageFC Controllers, SATA storage Sierra Logics ATA-FC bridgesSierra Logics ATA-FC bridges Custom ASICsCustom ASICs
PCI/PCI-X controllersPCI/PCI-X controllers Most RAIDMost RAID Some JBODSome JBOD
You Get What You Pay ForYou Get What You Pay For
FirmwareFirmware Fixed 512B blocksFixed 512B blocks
• No checksum per driveNo checksum per drive Bad block allocation timeoutsBad block allocation timeouts
• Some ATA/IDE/SATA fw allows for 2 minutesSome ATA/IDE/SATA fw allows for 2 minutes Head misalignmentsHead misalignments
• Common for disks with high idle times while poweredCommon for disks with high idle times while powered Single processor for head movement and data movementSingle processor for head movement and data movement
• Higher activity in one causes the other to degradeHigher activity in one causes the other to degrade Hot SpotsHot Spots Data corruptionData corruption
• Data checksummed and verified accurate at time t1, read back Data checksummed and verified accurate at time t1, read back corrupt at time t2.corrupt at time t2.
You Get What You Pay For (2)You Get What You Pay For (2)
Dual Controller – just two wordsDual Controller – just two words May not mean controllers failover (at all )May not mean controllers failover (at all )
• Each controller only handles ½ of RAID.Each controller only handles ½ of RAID. May not mean controllers failover May not mean controllers failover
automaticallyautomatically• System needs to be reset to have surviving System needs to be reset to have surviving
controller take over.controller take over. May not be availableMay not be available
• Option does not exist.Option does not exist.
You Get What You Pay For (3)You Get What You Pay For (3)
Stability of UnitStability of Unit May find self as un-willing beta testerMay find self as un-willing beta tester
• FW updates introduce more problems than fixesFW updates introduce more problems than fixes• FW updates fix problem, breaks other features that FW updates fix problem, breaks other features that
were workingwere working Often find bugs already known but not Often find bugs already known but not
discloseddisclosed Loss of data not as rare as one would expectLoss of data not as rare as one would expect
Upgrades are disruptiveUpgrades are disruptive
Under InvestigationUnder Investigation Enterprise class “S/ATA” solutionsEnterprise class “S/ATA” solutions
EMCEMC• Still uses 250GB, 5400RPM PATA drivesStill uses 250GB, 5400RPM PATA drives
LSILSI• 250GB, 7200RPM SATA250GB, 7200RPM SATA• 1.25TB, 5400RPM SATA blades1.25TB, 5400RPM SATA blades
HDSHDS• 250GB, 7200RPM SATA 250GB, 7200RPM SATA
Vendors slow to adapt to new technologyVendors slow to adapt to new technology None offer 400GB drives.None offer 400GB drives. Some claim 400GB drives show 1/8 reliability in Some claim 400GB drives show 1/8 reliability in
internal testing. May never adopt this drive.internal testing. May never adopt this drive.
Under Investigation (2)Under Investigation (2)
Tweaked SATA storageTweaked SATA storage FW changesFW changes
• Allows for 520B blocks – can checksum per driveAllows for 520B blocks – can checksum per drive• Periodic head sweeps – prevents hot spots, Periodic head sweeps – prevents hot spots,
misalignmentsmisalignments• Perceived “better” testing of sw, hw, fw.Perceived “better” testing of sw, hw, fw.
Some claim drives not available in open Some claim drives not available in open marketmarket
Upgrades are non-disruptive (dual-attach)Upgrades are non-disruptive (dual-attach) More expensive…but not too outrageousMore expensive…but not too outrageous
Understand Your NeedsUnderstand Your Needs
Caching storageCaching storage Massive storageMassive storage Loss of disk access and data tolerableLoss of disk access and data tolerable
• Data also on tape or other storage mediumData also on tape or other storage medium Low I/O.Low I/O.
Secondary storageSecondary storage Loss of disk access somewhat tolerableLoss of disk access somewhat tolerable
• Downtimes allowable, usually off hours.Downtimes allowable, usually off hours. Loss of data NOT tolerableLoss of data NOT tolerable Low I/OLow I/O