Storage Technologies - 3 - COMP 25212 - Lecture...

transcript

Storage Technologies - 3COMP 25212 - Lecture 10

Antoniu Popantoniu.pop@manchester.ac.uk

1 March 2019

Antoniu Pop – Storage Technologies - 3 1 / 20

Learning Objectives - Storage 3

I Understand characteristics of Solid-State Drives (SSD)

I Compare SSD and Hard DisksI Understand Logical Volume Management (LVM)I Understand Storage Area Networks (SAN)I Relate LVM and SAN to a modern File System implementation

I Understand characteristics of Solid-State Drives (SSD)I Compare SSD and Hard Disks

I Understand Logical Volume Management (LVM)I Understand Storage Area Networks (SAN)I Relate LVM and SAN to a modern File System implementation

I Understand characteristics of Solid-State Drives (SSD)I Compare SSD and Hard DisksI Understand Logical Volume Management (LVM)

I Understand Storage Area Networks (SAN)I Relate LVM and SAN to a modern File System implementation

I Understand characteristics of Solid-State Drives (SSD)I Compare SSD and Hard DisksI Understand Logical Volume Management (LVM)I Understand Storage Area Networks (SAN)

I Relate LVM and SAN to a modern File System implementation

I Understand characteristics of Solid-State Drives (SSD)I Compare SSD and Hard DisksI Understand Logical Volume Management (LVM)I Understand Storage Area Networks (SAN)I Relate LVM and SAN to a modern File System implementation

Solid-State Drive/Disk (SSD)

Floating Gate

Control Gate

DrainSource

Flash Memory:I Floating Gate Field Effect Transistor

I Charge stored on the floating gateI No electrical connectionI Conceptually like a switch: on (0) / off (1)I Possibly multi-level (4 states - 2 bits)

Floating Gate

Control Gate

DrainSource

Flash Memory:I Floating Gate Field Effect TransistorI Charge stored on the floating gate

I No electrical connectionI Conceptually like a switch: on (0) / off (1)I Possibly multi-level (4 states - 2 bits)

Floating Gate

Control Gate

DrainSource

Flash Memory:I Floating Gate Field Effect TransistorI Charge stored on the floating gateI No electrical connection

I Conceptually like a switch: on (0) / off (1)I Possibly multi-level (4 states - 2 bits)

Floating Gate

Control Gate

DrainSource

Flash Memory:I Floating Gate Field Effect TransistorI Charge stored on the floating gateI No electrical connectionI Conceptually like a switch: on (0) / off (1)

I Possibly multi-level (4 states - 2 bits)

Floating Gate

Control Gate

DrainSource

Flash Memory:I Floating Gate Field Effect TransistorI Charge stored on the floating gateI No electrical connectionI Conceptually like a switch: on (0) / off (1)I Possibly multi-level (4 states - 2 bits)

Flash Controller

Issues:I Data retention (10 year?)

I Wear-out with write cycles (aka P/E cycles)I Performance degradation with wearout

Implement:I Error Correcting CodesI (Bad) block remappingI Wear-levelling:

I remap Logical Block Addresses (LBA) to physical addressesI avoid wearing out specific blocks

Flash Controller

Issues:I Data retention (10 year?)I Wear-out with write cycles (aka P/E cycles)

I Performance degradation with wearoutImplement:I Error Correcting CodesI (Bad) block remappingI Wear-levelling:

Flash Controller

Issues:I Data retention (10 year?)I Wear-out with write cycles (aka P/E cycles)I Performance degradation with wearout

Flash Controller

Implement:I Error Correcting Codes

I (Bad) block remappingI Wear-levelling:

Flash Controller

Implement:I Error Correcting CodesI (Bad) block remapping

I Wear-levelling:I remap Logical Block Addresses (LBA) to physical addressesI avoid wearing out specific blocks

Flash Controller

I remap Logical Block Addresses (LBA) to physical addresses

I avoid wearing out specific blocks

Flash Controller

Disk access example (recap)Rotation

I Host initiates readsends a list of blocks to readI Block schedule requested...I ... may not be optimalI and leads to extra revolutionsI HDD internal processoroptimizes the scheduleI No direct mapping fromblock numbers to thesector/track/cylinder position(high-level interfaces like ATA / SCSI)

Hard Disk Performance (recap)Seek time Time for the head to reach the target track.Search time Time for the target sector to arrive under the head. Alsocalled rotational latency.Transfer rate Amount of data that can be read / written per unit oftime. Dependent on access patterns.Aka. “sustained transfer rate” in contrast to “interface transfer rate”

Disk access time = seek time + search time + transfer time

Note: all values are average as they depend on many factors.Antoniu Pop – Storage Technologies - 3 6 / 20

Hard Disks are too slow (recap)

Slow because of:I High seek time

I Reduce the number of times the head must moveI Multiple platters =⇒more tracks/sectors per cylinder

I High search time (aka. rotational latency)I Increase the rotation speed (e.g., server disks up to 15000 RPM)

I Low sustained transfer rateI Increase rotation speed (physical limitations)I Increase the recording density (physical limitations)I Apply cache and prefetch principlesI “Stripe” file system across multiple disks

SSD vs. HDD

HDD SSD

Hitachi 7k4000 Samsung SSD 840

SSD vs. HDD

HDD SSDStreaming Reads 205 MB/s 530 MB/s

SSD vs. HDD

HDD SSDStreaming Reads 205 MB/s 530 MB/sStreaming Writes 205 MB/s 240 MB/s

SSD vs. HDD

HDD SSDStreaming Reads 205 MB/s 530 MB/sStreaming Writes 205 MB/s 240 MB/sRandom 4kB Read 15.5 ms 11 µs

SSD vs. HDD

HDD SSDStreaming Reads 205 MB/s 530 MB/sStreaming Writes 205 MB/s 240 MB/sRandom 4kB Read 15.5 ms 11 µsRandom 4kB Write 6.4 ms 23 µs

SSD vs. HDD

HDD SSDStreaming Reads 205 MB/s 530 MB/sStreaming Writes 205 MB/s 240 MB/sRandom 4kB Read 15.5 ms 11 µsRandom 4kB Write 6.4 ms 23 µsPower 4/6/8 W 0.3/4.2 W

SSD vs. HDD

HDD SSDStreaming Reads 205 MB/s 530 MB/sStreaming Writes 205 MB/s 240 MB/sRandom 4kB Read 15.5 ms 11 µsRandom 4kB Write 6.4 ms 23 µsPower 4/6/8 W 0.3/4.2 WCapacity – Price 4 TB – £140 250 GB – £125

SSD vs. HDD

HDD SSDStreaming Reads 205 MB/s 530 MB/sStreaming Writes 205 MB/s 240 MB/sRandom 4kB Read 15.5 ms 11 µsRandom 4kB Write 6.4 ms 23 µsPower 4/6/8 W 0.3/4.2 WCapacity – Price 4 TB – £140 250 GB – £125Price per GB £0.035/GB £0.5/GBHitachi 7k4000 Samsung SSD 840

Example: disk access time (1 - recap)How long would it take on average to read / write a 512 byte sector onthis disk?Disk access time = seek time + search time + transfer time

seek time: 8.5 ms

search time: the disk must, on average, complete a half rotation7200 RPM =⇒ 0.5 rotations · 60 sec

min7200 RPM = 4.16 ms

transfer time: 512 B177·106 B/sec = 2.89 µs

access time = 8.5 + 4.16 + 2.89 · 10−3 = 12.66 ms

Example: disk access time (2 - recap)How long would it take on average to read / write 512 MB on thisdisk? (assuming sectors are “contiguous”)Disk access time = seek time + search time + transfer time

seek time: 8.5 ms

search time: the disk must, on average, complete a half rotation7200 RPM =⇒ 0.5 rotations · 60 sec

min7200 RPM = 4.16 ms

transfer time: 512 ·106 B177·106 B/sec = 2.89 s

access time = 8.5 · 10−3 + 4.16 · 10−3 + 2.89 = 2.9 s

Storage Virtualization

File System

Device Drivers

File System (classical)I FS to HDD partition mappingI FS does not span multiple drives

RAID changes thisI E.g., Striping or Mirroring

Storage Virtualization:I break the FS/HDD mapping

File System

Device Drivers

File System

Device Drivers

Logical Volume Management

I Virtual mapping between file system code and physical device

I Similar (but not identical!) to virtual memory addressingI FTSE: one more level of indirection

I “Volume Group”: set of drives in a poolI Storage space in “Volume Group” divided into “Physical Extents”

I usually all same sizeI “Logical Volume” is a set of “Physical Extents”

I Virtual mapping between file system code and physical deviceI Similar (but not identical!) to virtual memory addressing

I FTSE: one more level of indirectionI “Volume Group”: set of drives in a poolI Storage space in “Volume Group” divided into “Physical Extents”

I FTSE: one more level of indirection

I “Volume Group”: set of drives in a poolI Storage space in “Volume Group” divided into “Physical Extents”

I FTSE: one more level of indirectionI “Volume Group”: set of drives in a pool

I Storage space in “Volume Group” divided into “Physical Extents”I usually all same size

I “Logical Volume” is a set of “Physical Extents”

I usually all same size

I “Logical Volume” is a set of “Physical Extents”

Device Drivers

File System

LVM layer

Mirror/Stripe/RAIDI within the LVM layer

Resize the File SystemI add physical extentsI extend the partition

“Snapshot” a live filesystem

Device Drivers

File System

LVM layer

Device Drivers

File System

LVM layer

Example: the Linux File System

I / – mostly read: want fast seeks, high read transfer ratesI swap – read / write: want high bandwidth, data loss (?)I /opt – infrequent accessI /var – huge, infrequent access

I Mirror /I Stripe swapI spare space to /opt and /var

I / – mostly read: want fast seeks, high read transfer rates

I swap – read / write: want high bandwidth, data loss (?)I /opt – infrequent accessI /var – huge, infrequent access

I / – mostly read: want fast seeks, high read transfer ratesI swap – read / write: want high bandwidth, data loss (?)

I /opt – infrequent accessI /var – huge, infrequent access

I / – mostly read: want fast seeks, high read transfer ratesI swap – read / write: want high bandwidth, data loss (?)I /opt – infrequent access

I /var – huge, infrequent access

I Mirror /

I Stripe swapI spare space to /opt and /var

I Mirror /I Stripe swap

I spare space to /opt and /var

LVM Example

Efficient / flexible resource management

I / is mirrored across two disksI swap is striped across two other disksI /opt uses space on one diskI /var takes the remaining space across other 3 disks

LVM Example

Efficient / flexible resource managementI / is mirrored across two disks

I swap is striped across two other disksI /opt uses space on one diskI /var takes the remaining space across other 3 disks

LVM Example

Efficient / flexible resource managementI / is mirrored across two disksI swap is striped across two other disks

I /opt uses space on one diskI /var takes the remaining space across other 3 disks

LVM Example

Efficient / flexible resource managementI / is mirrored across two disksI swap is striped across two other disksI /opt uses space on one disk

I /var takes the remaining space across other 3 disks

LVM Example

Efficient / flexible resource managementI / is mirrored across two disksI swap is striped across two other disksI /opt uses space on one diskI /var takes the remaining space across other 3 disks

Storage Area Networks

I Implement LVM features in a separate storage controller

I Connect multiple servers to storage controllerI via SCSI, or FibreChannel, or Infiniband, or...I SAN over Ethernet, aka Networked Attached Storage (NAS)

I Share disk resources across multiple serversI Rapid migration of disk images

I Implement LVM features in a separate storage controllerI Connect multiple servers to storage controller

I via SCSI, or FibreChannel, or Infiniband, or...

I SAN over Ethernet, aka Networked Attached Storage (NAS)I Share disk resources across multiple serversI Rapid migration of disk images

I via SCSI, or FibreChannel, or Infiniband, or...I SAN over Ethernet, aka Networked Attached Storage (NAS)

I Share disk resources across multiple servers

I Rapid migration of disk images

SAN Controller

SANController

"Blade servers"

Decouple compute servers from storage servers.Connect through network:I Bandwidth ?I Latency ?

SAN Controller

SANController

"Blade servers"

Decouple compute servers from storage servers.

Connect through network:I Bandwidth ?I Latency ?

SAN Controller

SANController

"Blade servers"

Decouple compute servers from storage servers.Connect through network:

I Bandwidth ?I Latency ?

SAN Controller

SANController

"Blade servers"

Decouple compute servers from storage servers.Connect through network:I Bandwidth ?

I Latency ?

SAN Controller

SANController

"Blade servers"

Decouple compute servers from storage servers.Connect through network:I Bandwidth ?I Latency ?

SAN Key Features

I FunctionalityI Key element of System Virtualization

I Migrating virtual machinesI “De-duping” – share common subsets of file systems (thinkVirtual Machine images!)

I Management:I Manage storage separately from server physical resourcesI Maximize flexibility of storage provisioning

SAN Key Features

I FunctionalityI Key element of System VirtualizationI Migrating virtual machines

I “De-duping” – share common subsets of file systems (thinkVirtual Machine images!)I Management:

I Manage storage separately from server physical resourcesI Maximize flexibility of storage provisioning

SAN Key Features

I FunctionalityI Key element of System VirtualizationI Migrating virtual machinesI “De-duping” – share common subsets of file systems (thinkVirtual Machine images!)

SAN Key Features

I FunctionalityI Key element of System VirtualizationI Migrating virtual machinesI “De-duping” – share common subsets of file systems (thinkVirtual Machine images!)

ZFS – Volume Aware File System

Marketing claims:

I Lost a file?I Run out of space?I Difficult disk upgrade?I Want to grow/shrink?I Data Corruption?

Marketing claims:I Lost a file?

I Run out of space?I Difficult disk upgrade?I Want to grow/shrink?I Data Corruption?

Marketing claims:I Lost a file?I Run out of space?

I Difficult disk upgrade?I Want to grow/shrink?I Data Corruption?

Marketing claims:I Lost a file?I Run out of space?I Difficult disk upgrade?

I Want to grow/shrink?I Data Corruption?

Marketing claims:I Lost a file?I Run out of space?I Difficult disk upgrade?I Want to grow/shrink?

I Data Corruption?

Marketing claims:I Lost a file?I Run out of space?I Difficult disk upgrade?I Want to grow/shrink?I Data Corruption?

ZFS SolutionsI Lost a file?

I Copy-on-Write (CoW)I simple rollback/recoveryI (indirect wear-leveling)

I Run out of space / difficult disk upgrade?I Add new storage to live systemsI Self-checking, self-healing

I Want to grow / shrink?I Data Corruption?

I end-to-end sumcheckingZFS combines File System and Logical Volume Management

I Copy-on-Write (CoW)

I simple rollback/recoveryI (indirect wear-leveling)

I Copy-on-Write (CoW)I simple rollback/recovery

I (indirect wear-leveling)I Run out of space / difficult disk upgrade?

I Add new storage to live systemsI Self-checking, self-healing

I Run out of space / difficult disk upgrade?

I Add new storage to live systemsI Self-checking, self-healing

I Want to grow / shrink?

I Data Corruption?I end-to-end sumchecking

ZFS combines File System and Logical Volume Management

I end-to-end sumchecking

ZFS combines File System and Logical Volume Management

Storage Technologies - 3 - COMP 25212 - Lecture...

Documents