Chapter 10: Mass-Storage Systems. Operating System Concepts 9 th Edition

Transcription

1 Chapter 10: Mass-Storage Systems Silberschatz, Galvin and Gagne 2013

2 Objectives To describe the physical structure of secondary storage devices and its effects on the uses of the devices To explain the performance characteristics of mass-storage devices To evaluate disk scheduling algorithms To discuss operating-system services provided for mass storage, including RAID 10.2 Silberschatz, Galvin and Gagne 2013

3 Chapter 10: Mass-Storage Systems Overview of Mass Storage Structure Disk Structure Disk Attachment Disk Scheduling Disk Management Swap-Space Management RAID Structure Stable-Storage Implementation 10.3 Silberschatz, Galvin and Gagne 2013

4 Warm-up Virtualization Concurrency Persistency Challenges in making information persist, despite computer crashes, disk failures, or power outages Silberschatz, Galvin and Gagne 2013

5 Warm-up Fast System Architecture Slow 10.5 Silberschatz, Galvin and Gagne 2013

6 Warm-up Hard disk drive The main form of persistent data storage in computer systems Non-volatile random-access memory Optane (Intel) 10.6 Silberschatz, Galvin and Gagne 2013

7 Warm-up 10.7 Silberschatz, Galvin and Gagne 2013

8 Chapter 10: Mass-Storage Systems Overview of Mass Storage Structure Disk Structure Disk Attachment Disk Scheduling Disk Management Swap-Space Management RAID Structure Stable-Storage Implementation 10.8 Silberschatz, Galvin and Gagne 2013

9 Moving-head Disk Mechanism 10.9 Silberschatz, Galvin and Gagne 2013

10 The Interface A modern disk drive consists of a large number of sectors (512-byte blocks), each of which can be read or written. The sectors are numbered from 0 to n 1 on a disk with n sectors. Thus, we can view the disk as an array of sectors; 0 to n 1 is thus the address space of the drive. Multi-sector operations are possible; indeed, many file systems will read or write 4KB at a time (or more). A single 512-byte write is atomic (i.e., it will either complete in its entirety or it won t complete at all Silberschatz, Galvin and Gagne 2013

11 Basic Geometry Components of a modern disk A platter is a circular hard surface on which data is stored persistently by inducing magnetic changes to it. These platters are usually made of some hard material (such as aluminum), and then coated with a thin magnetic layer that enables the drive to persistently store bits even when the drive is powered off. Each platter has 2 sides, each of which is called a surface Silberschatz, Galvin and Gagne 2013

12 Basic Geometry Components of a modern disk (contd.) The platters are all bound together around the spindle, which is connected to a motor that spins the platters around at a constant rate. The rate of rotation is often measured in rotations per minute (RPM), and typical modern values are in the 7,200 RPM to 15,000 RPM range Silberschatz, Galvin and Gagne 2013

13 Basic Geometry Components of a modern disk (contd.) Data is encoded on each surface in concentric circles of sectors; we call one such concentric circle a track. The process of reading (i.e., sensing the magnetic patterns on the disk) and writing (i.e., inducing a change in them) is accomplished by the disk head; there is one such head per surface of the drive. The disk head is attached to a single disk arm, which moves across the surface to position the head over the desired track Silberschatz, Galvin and Gagne 2013

14 A Simple Disk Drive Suppose a simple disk with three tracks Silberschatz, Galvin and Gagne 2013

15 Rotation Rotations and seeks (NOT transfer) are the most costly disk operations. Rotational delay Time of waiting for the desired sector to rotate under the disk head. Average rotational delay is half of the full rotational delay Silberschatz, Galvin and Gagne 2013

16 Seek Phases in a seek: First an acceleration phase as the disk arm gets moving; Then coasting as the arm is moving at full speed; Then deceleration as the arm slows down; Finally settling as the head is positioned over the correct track. The settling time is often quite significant, e.g., 0.5 to 2 ms, as the drive must be certain to find the right track. Seek Time Time of moving the disk arm to the correct track. Average disk-seek time is roughly one-third of the full seek time Silberschatz, Galvin and Gagne 2013

17 I/O Time The complete picture of I/O time: First a seek, then a remaining rotation, and finally the transfer. The rate of I/O is often more easily used for comparison between drives Silberschatz, Galvin and Gagne 2013

18 I/O Time Assume two workloads Random workload, issues small (e.g., 4KB) reads to random locations on the disk. Sequential workload, simply reads a large number of sectors consecutively from the disk, without jumping around. Let us perform the following calculation Silberschatz, Galvin and Gagne 2013

19 I/O Time A couple of modern disks from Seagate The Cheetah 15K.5, a high-performance SCSI drive The Barracuda, a SATA drive built for capacity Silberschatz, Galvin and Gagne 2013

20 Random Workload Transfer 4KB / 6 / 0.66 / /? /? / Silberschatz, Galvin and Gagne 2013

21 Sequential Workload Assume the size of the transfer is 100 MB. How to calculate / and /? T I/O for the Barracuda and Cheetah is about 800 ms and 950 ms, respectively. R I/O are very nearly the peak transfer rates of 125 MB/s and 105 MB/s, respectively Silberschatz, Galvin and Gagne 2013

22 Results There is a huge gap in drive performance between random and sequential workloads. There is a large difference in performance between high-end performance drives and low-end capacity drives Silberschatz, Galvin and Gagne 2013

23 Some Other Details Track skew: to make sure that sequential reads can be properly serviced even when crossing track boundaries Silberschatz, Galvin and Gagne 2013

24 Some Other Details Multi-zoned disk drives Outer tracks have more sectors than inner tracks. The disk is organized into multiple zones, and where a zone is consecutive set of tracks on a surface. Each zone has the same number of sectors per track, and outer zones have more sectors than inner zones. Cache (or track buffer) Some small amount of memory (usually around 8 or 16 MB) which the drive can use to hold data read from or written to the disk Silberschatz, Galvin and Gagne 2013

25 Solid-State Disks Nonvolatile memory used like a hard drive Many technology variations Can be more reliable than HDDs More expensive per MB Maybe have shorter life span Less capacity But much faster Bottleneck: Busses can be too slow -> connect directly to PCI for example No moving parts, so no seek time or rotational latency Silberschatz, Galvin and Gagne 2013

26 Magnetic Tape Was early secondary-storage medium Evolved from open spools to cartridges Relatively permanent and holds large quantities of data Access time slow Random access ~1000 times slower than disk Mainly used for backup, storage of infrequently-used data, transfer medium between systems Kept in spool and wound or rewound past read-write head Once data under head, transfer rates comparable to disk 140MB/sec and greater 200GB to 1.5TB typical storage Common technologies are LTO-{3,4,5} and T Silberschatz, Galvin and Gagne 2013

27 Chapter 10: Mass-Storage Systems Overview of Mass Storage Structure Disk Structure Disk Attachment Disk Scheduling Disk Management Swap-Space Management RAID Structure Stable-Storage Implementation Silberschatz, Galvin and Gagne 2013

28 Disk Structure Disk drives are addressed as large 1-dimensional arrays of logical blocks, where the logical block is the smallest unit of transfer Low-level formatting creates logical blocks on physical media The 1-dimensional array of logical blocks is mapped into the sectors of the disk sequentially Logical to physical address is not easy Hiding bad sectors Non-constant # of sectors per track via constant angular velocity Silberschatz, Galvin and Gagne 2013

29 Chapter 10: Mass-Storage Systems Overview of Mass Storage Structure Disk Structure Disk Attachment Disk Scheduling Disk Management Swap-Space Management RAID Structure Stable-Storage Implementation Silberschatz, Galvin and Gagne 2013

30 Disk Attachment Host-attached storage accessed through I/O ports talking to I/O busses SCSI itself is a bus, up to 16 devices on one cable, SCSI initiator requests operation and SCSI targets perform tasks Each target can have up to 8 logical units (disks attached to device controller) FC is high-speed serial architecture Can be switched fabric with 24-bit address space the basis of storage area networks (SANs) in which many hosts attach to many storage units I/O directed to bus ID, device ID, logical unit Silberschatz, Galvin and Gagne 2013

31 Network-Attached Storage Network-attached storage (NAS) is storage made available over a network rather than over a local connection (such as a bus) Remotely attaching to file systems NFS and CIFS are common protocols Implemented via remote procedure calls (RPCs) between host and storage over typically TCP or UDP on IP network iscsi protocol uses IP network to carry the SCSI protocol Remotely attaching to devices (blocks) RPCs Silberschatz, Galvin and Gagne 2013

32 Storage Area Network Common in large storage environments Multiple hosts attached to multiple storage arrays - flexible Silberschatz, Galvin and Gagne 2013

33 Chapter 10: Mass-Storage Systems Overview of Mass Storage Structure Disk Structure Disk Attachment Disk Scheduling Disk Management Swap-Space Management RAID Structure Stable-Storage Implementation Silberschatz, Galvin and Gagne 2013

34 Disk Scheduling The operating system is responsible for using hardware efficiently for the disk drives, this means having a fast access time and disk bandwidth Minimize seek time Seek time seek distance Disk bandwidth is the total number of bytes transferred, divided by the total time between the first request for service and the completion of the last transfer Silberschatz, Galvin and Gagne 2013

35 Disk Scheduling (Cont.) There are many sources of disk I/O request OS System processes Users processes I/O request includes input or output mode, disk address, memory address, number of sectors to transfer OS maintains queue of requests, per disk or device Idle disk can immediately work on I/O request, busy disk means work must queue Optimization algorithms only make sense when a queue exists We illustrate scheduling algorithms with a request queue (0-199) Head pointer 53 98, 183, 37, 122, 14, 124, 65, Silberschatz, Galvin and Gagne 2013

36 FCFS Illustration shows total head movement of 640 cylinders Silberschatz, Galvin and Gagne 2013

37 SSTF Shortest Seek Time First selects the request with the minimum seek time from the current head position SSTF scheduling is a form of SJF scheduling; may cause starvation of some requests Illustration shows total head movement of 236 cylinders Silberschatz, Galvin and Gagne 2013

38 SCAN The disk arm starts at one end of the disk, and moves toward the other end, servicing requests until it gets to the other end of the disk, where the head movement is reversed and servicing continues. SCAN algorithm Sometimes called the elevator algorithm Illustration shows total head movement of 236 cylinders But note that if requests are uniformly dense, largest density at other end of disk and those wait the longest Silberschatz, Galvin and Gagne 2013

39 SCAN (Cont.) Silberschatz, Galvin and Gagne 2013

40 C-SCAN Provides a more uniform wait time than SCAN The head moves from one end of the disk to the other, servicing requests as it goes When it reaches the other end, however, it immediately returns to the beginning of the disk, without servicing any requests on the return trip Treats the cylinders as a circular list that wraps around from the last cylinder to the first one Total number of cylinders? Silberschatz, Galvin and Gagne 2013

41 C-SCAN (Cont.) Silberschatz, Galvin and Gagne 2013

42 C-LOOK LOOK a version of SCAN, C-LOOK a version of C-SCAN Arm only goes as far as the last request in each direction, then reverses direction immediately, without first going all the way to the end of the disk Total number of cylinders? Silberschatz, Galvin and Gagne 2013

43 C-LOOK (Cont.) Silberschatz, Galvin and Gagne 2013

44 Exercise Suppose that a disk drive has 200 cylinders, numbered 0 to 199. The drive is currently serving a request at cylinder 123, and the previous request was at cylinder 175. The queue of pending requests, in FIFO order, is 86, 147, 180, 28, 95, 151, 12, 77, 30. Starting from the current head position, what is the total distance (in cylinders) that the disk arm moves to satisfy all the pending requests for each of the following diskscheduling algorithms respectively? a. SSTF b. C-SCAN c. LOOK Silberschatz, Galvin and Gagne 2013

45 Exercise Disk requests come into the disk driver for cylinders as follows: Request No. Cylinder No. Arrive time (ms) Assume that the disk has 100 cylinders (from 0 to 99). A seek takes 6 ms per cylinder moved. Compute the average seek time for the request sequence given above for First-come, First-served Shortest Seek Time First (SSTF) LOOK (with the disk-arm initially moving towards higher number cylinders from lower number cylinders) C-SCAN (with the disk-arm initially moving towards higher number cylinders from lower number cylinders) In all the cases, the arm is initially at cylinder Silberschatz, Galvin and Gagne 2013

46 Disk Scheduling Remaining problem: SCANs do not represent the best scheduling technology (they ignore rotation). Suppose the head is currently positioned over sector 30 on the inner track. The scheduler thus has to decide: should it schedule sector 16 (on the middle track) or sector 8 (on the outer track) for its next request. It depends on the relative time of seeking as compared to rotation Silberschatz, Galvin and Gagne 2013

47 Disk Scheduling SPTF: Shortest Positioning Time First (also called shortest access time first or SATF) Usually performed inside a drive Silberschatz, Galvin and Gagne 2013

48 Selecting a Disk-Scheduling Algorithm SSTF (or SPTF) is common and has a natural appeal SCAN and C-SCAN perform better for systems that place a heavy load on the disk Less starvation Performance depends on the number and types of requests Requests for disk service can be influenced by the file-allocation method And metadata layout Where is disk scheduling performed? In older systems, the OS did all the scheduling. In modern systems, the OS scheduler performs I/O merging; the disk then uses its internal knowledge of head position and detailed track layout information to service said requests in the best possible (SPTF) order. The OS has other constraints on the service order for requests Silberschatz, Galvin and Gagne 2013

49 Chapter 10: Mass-Storage Systems Overview of Mass Storage Structure Disk Structure Disk Attachment Disk Scheduling Disk Management Swap-Space Management RAID Structure Stable-Storage Implementation Silberschatz, Galvin and Gagne 2013

50 Disk Management Low-level formatting, or physical formatting Dividing a disk into sectors that the disk controller can read and write Each sector can hold header information, plus data, plus error correction code (ECC) Usually 512 bytes of data but can be selectable To use a disk to hold files, the operating system still needs to record its own data structures on the disk Partition the disk into one or more groups of cylinders, each treated as a logical disk Logical formatting or making a file system To increase efficiency most file systems group blocks into clusters Disk I/O done in blocks File I/O done in clusters Raw disk access for apps that want to do their own block management, keep OS out of the way (databases for example) Silberschatz, Galvin and Gagne 2013

51 Disk Management (Cont.) Boot block initializes system Bootstrap loader program in ROM Bootstrap program in boot blocks of boot partition Methods such as sector sparing used to handle bad blocks Silberschatz, Galvin and Gagne 2013

52 Chapter 10: Mass-Storage Systems Overview of Mass Storage Structure Disk Structure Disk Attachment Disk Scheduling Disk Management Swap-Space Management RAID Structure Stable-Storage Implementation Silberschatz, Galvin and Gagne 2013

53 Swap-Space Management Swap-space Virtual memory uses disk space as an extension of main memory Less common now due to memory capacity increases Swap-space can be carved out of the normal file system, or, more commonly, it can be in a separate disk partition (raw) Swap-space management Silberschatz, Galvin and Gagne 2013

54 Chapter 10: Mass-Storage Systems Overview of Mass Storage Structure Disk Structure Disk Attachment Disk Scheduling Disk Management Swap-Space Management RAID Structure Stable-Storage Implementation Silberschatz, Galvin and Gagne 2013

55 RAID Structure Redundant Arrays of Inexpensive Disks (RAIDs) A technique to use multiple disks in concert to build a faster, bigger, and more reliable disk system. Externally, a RAID looks like a disk: a group of blocks one can read or write. This transparency improves deploy ability of RAID. Internally, a RAID consists of multiple disks, memory (both volatile and non-), and one or more processors to manage the system. RAIDs offer various advantages Silberschatz, Galvin and Gagne 2013

56 RAID Structure Why using RAID instead of a single disk? Performance: Using multiple disks in parallel can greatly speed up I/O times. Capacity: Large data sets demand large disks. Reliability: With some form of redundancy, RAIDs can tolerate the loss of a disk and keep operating as if nothing were wrong Silberschatz, Galvin and Gagne 2013

57 RAID Structure Some calculation for redundancy Increases the mean time to failure Mean time to repair exposure time when another failure could cause data loss Mean time to data loss based on above factors If mirrored disks fail independently, consider disk with 1300,000 mean time to failure and 10 hour mean time to repair Mean time to data loss is 100, / (2 10) = hours, or 57,000 years! Frequently combined with NVRAM to improve write performance Silberschatz, Galvin and Gagne 2013

58 RAID Levels We are about to discuss the RAID designs. RAID Level 0 (striping) RAID Level 1 (mirroring) RAID Levels 4/5 (paritybased redundancy) Other RAID Levels Silberschatz, Galvin and Gagne 2013

59 Fault Model RAIDs are designed to detect and recover from certain kinds of disk faults. The fail-stop fault model: A disk can be in exactly one of two states: working or failed. With a working disk, all blocks can be read or written. In contrast, when a disk has failed, we assume it is permanently lost. When a disk has failed, it is easily detected Silberschatz, Galvin and Gagne 2013

60 How to Evaluate a RAID There are a number of different approaches to building a RAID. How to evaluate a RAID? Recall that RAID has advantages compared to a single disk: Performance Capacity Reliability Each RAID design can be evaluated along these 3 axes Silberschatz, Galvin and Gagne 2013

61 RAID Level 0: Striping The simplest form of striping will stripe blocks across the disks of the system as follows (assume here a 4-disk array): The blocks in the same row are called a stripe Silberschatz, Galvin and Gagne 2013

62 RAID Level 0: Striping RAID Level 0: is actually NOT a RAID level at all, in that there is no redundancy. serves as an excellent upper-bound on performance and capacity and thus is worth understanding Silberschatz, Galvin and Gagne 2013

63 RAID Level 0: Striping A variation Silberschatz, Galvin and Gagne 2013

64 Chunk Size Chunk size mostly affects performance of the array. A small chunk size implies that many files will get striped across many disks, thus increasing the parallelism of reads and writes to a single file. However, the positioning time to access blocks across multiple disks increases, because the positioning time for the entire request is determined by the maximum of the positioning times of the requests across all drives. Vice versa Silberschatz, Galvin and Gagne 2013

65 RAID Level 0: Striping RAID-0 Analysis Capacity: Perfect, given N disks each of size B blocks, striping delivers N B blocks of useful capacity. Reliability: Perfect but in the bad way, any disk failure will lead to data loss. Performance: Excellent, all disks are utilized, often in parallel, to service user I/O requests. More in detail Silberschatz, Galvin and Gagne 2013

66 Evaluating RAID Performance Two performance metrics Single-request latency Steady-state throughput Two types of workloads Sequential Random Two types of I/Os Read Write Silberschatz, Galvin and Gagne 2013

67 Evaluating RAID Performance We will assume that a disk can transfer data at S MB/s under a sequential workload, and R MB/s when under a random workload. In general, S is much greater than R (i.e., S R). Recall the calculation of / and / Silberschatz, Galvin and Gagne 2013

68 RAID Level 0: Striping Performance Single-request latency The latency of a single-block request should be just about identical to that of a single disk, since RAID-0 will simply redirect that request to one of its disks. Steady-state throughput Sequential N S MB/s Random N R MB/s Silberschatz, Galvin and Gagne 2013

69 RAID Level 1: Mirroring Simply make more than one copy of each block in the system; each copy should be placed on a separate disk, to tolerate disk failures Silberschatz, Galvin and Gagne 2013

70 RAID Level 1: Mirroring This common arrangement is sometimes called RAID-10 or (RAID 1+0) because it uses mirrored pairs (RAID-1) and then stripes (RAID-0) on top of them. You can imagine another arrangement, RAID-01 or RAID Silberschatz, Galvin and Gagne 2013

71 RAID (0 + 1) and (1 + 0) Silberschatz, Galvin and Gagne 2013

72 RAID Level 1: Mirroring RAID-1 Analysis Capacity: Expensive, with the mirroring level = 2, we only obtain half of our peak useful capacity. Reliability: Good, it can tolerate the failure of any one disk, and maybe more, up to N/2 disk failure Silberschatz, Galvin and Gagne 2013

73 RAID Level 1: Mirroring Performance: Single-request latency Read» The same as the latency on a single disk. Write» Two writes in parallel. Thus suffers the worst-case seek and rotational delay of the two. Steady-state throughput Sequential» NS/2 MB/s, for both writes and reads. (why?) Random» NR MB/s for reads, and NR/2 MB/s for writes Silberschatz, Galvin and Gagne 2013

74 RAID Level 4: Saving Space with Parity For each stripe of data, add a single parity block that stores the redundant information for that stripe of blocks. Parity is computed from the stripe of blocks by using a mathematical function XOR Silberschatz, Galvin and Gagne 2013

75 RAID Level 4: Saving Space with Parity You can imagine how parity information can be used to recover from a failure Silberschatz, Galvin and Gagne 2013

76 RAID Level 4: Saving Space with Parity RAID-4 Analysis Capacity: N-1, since 1 disk is used for parity information. Reliability: Tolerate 1 disk failure and no more Silberschatz, Galvin and Gagne 2013

77 RAID Level 4: Saving Space with Parity Performance: Single-request latency Read» The same as the latency on a single disk. Write» Twice as the latency on a single disk. (why?) Steady-state throughput Sequential» (N-1)S MB/s, for both reads and writes. (why?) Random» (N-1)R MB/s for reads, R/2 for writes. (why?) Silberschatz, Galvin and Gagne 2013

78 RAID Level 4: Saving Space with Parity Full-stripe writes are the most efficient way for RAID-4 to write to disk. (why?) For random writes Additive parity: To compute the value of the new parity block, read in all of the other data blocks in the stripe in parallel. Subtractive parity: use only the modified block. (where is the XOR operator): Using the subtractive method, for each write, the RAID has to perform 4 physical I/Os (two reads and two writes) Silberschatz, Galvin and Gagne 2013

79 RAID Level 4: Saving Space with Parity Now imagine there are lots of writes submitted to the RAID; how many can RAID-4 perform in parallel? Small-write problem: Even though small-writes to disk 0 and 1 can be performed in parallel, the problem that arises is with the parity disk Silberschatz, Galvin and Gagne 2013

80 RAID Level 4: Saving Space with Parity The parity disk is a bottleneck under the random workload. Even though the data disks could be accessed in parallel, the parity disk prevents any parallelism from materializing; all writes to the system will be serialized because of the parity disk. Thus we achieve (R/2) MB/s The parity disk has to perform two I/Os (one read, one write) per logical I/O. Terrible, and does not improve as you add disks to the system. Discussion: Improve RAID level 4, how to address the small-write problem Silberschatz, Galvin and Gagne 2013

81 RAID Level 5: Rotating Parity Rotate the parity block across drives Silberschatz, Galvin and Gagne 2013

82 RAID Level 5: Rotating Parity RAID-5 Analysis Capacity: N-1, since 1 disk is used for parity information. Reliability: Tolerate 1 disk failure and no more Silberschatz, Galvin and Gagne 2013

83 RAID Level 5: Rotating Parity Performance: Single-request latency Read» The same as the latency on a single disk. Write» Twice as the latency on a single disk. Steady-state throughput Sequential» (N-1)S MB/s, for both reads and writes. Random» NR MB/s for random read»? for random write Silberschatz, Galvin and Gagne 2013

84 RAID Level 5: Rotating Parity Random read performance is a little better, because we can utilize all of the disks. (NR MB/s) Random write performance improves noticeably over RAID-4, as it allows for parallelism across requests. (NR/4 MB/s) Imagine a write to block 1 and a write to block 10; this will turn into requests to disk 1 and disk 4 (for block 1 and its parity) and requests to disk 0 and disk 2 (for block 10 and its parity). Thus, they can proceed in parallel. In fact, we can generally assume that given a large number of random requests, we will be able to keep all the disks about evenly busy Silberschatz, Galvin and Gagne 2013

85 RAID Analysis Silberschatz, Galvin and Gagne 2013

86 Other RAID Levels RAID level 2 Stores Hamming codes as error-correcting code (ECC) in additional disks Silberschatz, Galvin and Gagne 2013

87 Other RAID Levels RAID Level 3 Bit-interleaved parity organization Silberschatz, Galvin and Gagne 2013

88 Other RAID Levels RAID Level 6 Extends RAID 5 by adding another parity block; thus, it uses block-level striping with two parity blocks distributed across all member disks RAID 6 implemented in software will have a more significant effect on system performance, and a hardware solution will be more complex Silberschatz, Galvin and Gagne 2013

89 Other Features Regardless of where RAID implemented, other useful features can be added Snapshot is a view of file system before a set of changes take place (i.e. at a point in time) More in Ch 12 Replication is automatic duplication of writes between separate sites For redundancy and disaster recovery Can be synchronous or asynchronous Hot spare disk is unused, automatically used by RAID production if a disk fails to replace the failed disk and rebuild the RAID set if possible Decreases mean time to repair Silberschatz, Galvin and Gagne 2013

90 Extensions RAID alone does not prevent or detect data corruption or other errors, just disk failures Solaris ZFS adds checksums of all data and metadata Checksums kept with pointer to object, to detect if object is the right one and whether it changed Can detect and correct data and metadata corruption ZFS also removes volumes, partitions Disks allocated in pools Filesystems with a pool share that pool, use and release space like malloc() and free() memory allocate / release calls Silberschatz, Galvin and Gagne 2013

91 Chapter 10: Mass-Storage Systems Overview of Mass Storage Structure Disk Structure Disk Attachment Disk Scheduling Disk Management Swap-Space Management RAID Structure Stable-Storage Implementation Silberschatz, Galvin and Gagne 2013

92 ZFS Checksums All Metadata and Data Silberschatz, Galvin and Gagne 2013

93 Traditional and Pooled Storage Silberschatz, Galvin and Gagne 2013

94 Stable-Storage Implementation Write-ahead log scheme requires stable storage Stable storage means data is never lost (due to failure, etc) To implement stable storage: Replicate information on more than one nonvolatile storage media with independent failure modes Update information in a controlled manner to ensure that we can recover the stable data after any failure during data transfer or recovery Disk write has 1 of 3 outcomes 1. Successful completion - The data were written correctly on disk 2. Partial failure - A failure occurred in the midst of transfer, so only some of the sectors were written with the new data, and the sector being written during the failure may have been corrupted 3. Total failure - The failure occurred before the disk write started, so the previous data values on the disk remain intact Silberschatz, Galvin and Gagne 2013

95 Stable-Storage Implementation (Cont.) If failure occurs during block write, recovery procedure restores block to consistent state System maintains 2 physical blocks per logical block and does the following: 1. Write to 1 st physical 2. When successful, write to 2 nd physical 3. Declare complete only after second write completes successfully Systems frequently use NVRAM as one physical to accelerate Silberschatz, Galvin and Gagne 2013

96 End of Chapter 10 Silberschatz, Galvin and Gagne 2013