RAIDCOMP375 1Disk ArraysCOMP375 Computer Architecture dO i tiand OrganizationGoals• Understand the different options for improving disk reliability and performanceimproving disk reliability and performance.• Be able to determine the appropriate RAID solution for a given situation. • Be able to calculate the overhead for different RAID configurations.Large Storage Needs• Some applications need more storage th i l di k h ldthan a single disk can hold.• Multiple disks need to be used.• Such application may wish to have quick access to large files.RAID• Redundant Array of Independent Disks•A collection of disks are used as one large unit of mass storage.• Multiple disks operating simultaneously can increase the data transfer rate.Extra data stored on the disks can recover•Extra data stored on the disks can recover the information should a disk fail.RAIDCOMP375 2Historical Note• In the mid 1980’s mainframe disk drives were physically large and expensive, about $50 000about $50,000.• PC disk drives were about 2/3 as fast, held about 1/5 as much and cost under $1,000.• People began to wonder if they couldn’t use a collection of small drivesuse a collection of small drives.•Redundant Array of Inexpensive DisksHardware or Software• RAID can be implemented by the I/O tll b ft ithOScontroller or by software in the OS.• A RAID controller hides the RAID functions from the OS and makes a set of disk drives look like a single large drive.•The OS can implement RAID on a set of•The OS can implement RAID on a set of regular disks. The user is unaware of the RAID features.Failure Rates• As the number of parts in a system i d th b bilit f f ilincreases, so does the probability of failure• A device’s reliability can be expressed as its probability of failure, P.• If a system is composed of N components each with reliabilityPthe reliability of theeach with reliability P, the reliability of the system is PNRAID Types• RAID 0 - Striping• RAID 1 - Mirroring• RAID 2 - Hamming code error recovery• RAID 3 - Bit-interleaved parity• RAID 4 - Block-level parity• RAID 5 - Block-level distributed parity• RAID 6 - Dual redundancyRAIDCOMP375 3RAID 0 (non-redundant)•Improved transfer rate•Decreased reliabilityhttp://www.acnc.com/04_01_00.htmlRAID 1 (mirrored)•Improved ReliabilityImproved Reliability•Slightly slower writes.•Possibly faster reads•Twice the disk space requiredXOR Parity• Consider the exclusive OR of several valuesX=ABCX = A B C• If you XOR any of the three values, you will get the fourth.B = X A C•RAID 3 4 5 & 6 write the XOR of data to an•RAID 3, 4, 5 & 6 write the XOR of data to an additional disk to provide recovery in the event a disk fails.RAIDCOMP375 4RAID 4 (block-level parity)Rarely usedRAID 5 (distributed block parity)•Striping improves read performance•Parity improves reliability•N+1 disks are requiredRAID 6 (dual redundancy)Like RAID 5 b t ith t o parit blocks•Like RAID 5 but with two parity blocks for each data block•Slow writes•N+2 disks requiredSynchronized Rotation• All of the drives in an disk array spin together• Most disk drives have a feature allowing synchronization.• Raid 0 simultaneously reads a block from all drives. If they were not synchronized, the system would have to wait for the last drivesystem would have to wait for the last drive.• Synchronized rotational delay = spin/2• Unsynchronized rotational delay =spin*(n-1)/nRAIDCOMP375 5RAID ComparisonRAID Disks Reads WritesSurvives failures0 N faster faster