Embed Size (px)
Citation preview
FlashBlox: Achieving Both Performance Isolation
and Uniform Lifetime for Virtualized SSDs
Jian Huang † Anirudh Badam Laura Caulfield
Suman Nath Sudipta Sengupta Bikash Sharma Moinuddin K. Qureshi
† ‡
‡
Flash Has Changed Over the Last Decade
2
Performance
Improvement
100x lower latency
5,000x higher throughput
Flash Has Changed Over the Last Decade
2
Increased
Parallelism
Dozens of
parallel chips
Performance
Improvement
100x lower latency
5,000x higher throughput
Flash Has Changed Over the Last Decade
2
Increased
Parallelism
Dozens of
parallel chips
Became
Commodity
Less than $0.2/GB
Performance
Improvement
100x lower latency
5,000x higher throughput
Flash Has Changed Over the Last Decade
2
Increased
Parallelism
Dozens of
parallel chips
Became
Commodity
Less than $0.2/GB
Significant improvements on Flash
Performance
Improvement
100x lower latency
5,000x higher throughput
Shared Flash-Based Solid State Disk (SSD) in the Cloud
3
…….
Shared Flash-Based Solid State Disk (SSD) in the Cloud
3
…….
SSDs are virtualized and shared in data centers
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
Channel
Chip Chip
……… …
Flash Translation Layer
Performance Interference in Shared SSD
4
…….
Flash-based SSD: A Black Box
Write Read
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
Channel
Chip Chip
……… …
Flash Translation Layer
Performance Interference in Shared SSD
4
…….
Flash-based SSD: A Black Box
Write ReadRead/write interferences cause long (3x) tail latency!
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
Channel
Chip Chip
……… …
Flash Translation Layer
Performance Interference in Shared SSD
4
…….
Write Read
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
Channel
Chip Chip
……… …
Flash Translation Layer
Performance Interference in Shared SSD
4
…….
Write Read
FlashBlox: Hardware Isolation in Cloud Storage
5
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
…….
Channel
Chip Chip
……… …
Flash Translation Layer
FlashBlox: Hardware Isolation in Cloud Storage
5
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
…….
Channel
Chip Chip
……… …
Flash Translation Layer
Leveraging parallel chips for hardware isolation
Internal Parallelism Enables Hardware Isolation
6
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
Internal Parallelism Enables Hardware Isolation
6
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
Channel-Level Parallelism
Internal Parallelism Enables Hardware Isolation
6
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
Channel-Level Parallelism
Chip-Level Parallelism
Internal Parallelism Enables Hardware Isolation
6
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
Channel-Level Parallelism
Chip-Level Parallelism
Plane-Level
Parallelism
Plane-level parallelism is constrained
as each chip contains only one address buffer
Internal Parallelism Enables Hardware Isolation
6
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
Channel-Level Parallelism
Chip-Level Parallelism
Plane-Level
Parallelism
Different parallelism level provides different isolation guarantee
New Abstractions for Hardware Isolation
7
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
New Abstractions for Hardware Isolation
7
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
Virtual SSD
(Chip Level)
Virtual SSD
(Channel Level)
Virtual SSD
(Plane Level)
High Medium Low
New Abstractions for Hardware Isolation
7
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
Virtual SSD
(Chip Level)
Virtual SSD
(Channel Level)
Virtual SSD
(Plane Level)
High Medium Low
Software-based
Hardware Isolation Meets the Pay-As-You-Go Model in Cloud
8
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
vSSD (Chip)vSSD (Channel) vSSD (Software)
Hardware Isolation Meets the Pay-As-You-Go Model in Cloud
8
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
vSSD (Chip)vSSD (Channel) vSSD (Software)
Azure
DocumentDB
Azure
SQL DatabaseAmazon
DynamoDB
Hardware Isolation Meets the Pay-As-You-Go Model in Cloud
8
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
vSSD (Chip)vSSD (Channel) vSSD (Software)
Azure
DocumentDB
Azure
SQL DatabaseAmazon
DynamoDB
Throughput
Single Partition Size
Price
Hardware Isolation Meets the Pay-As-You-Go Model in Cloud
8
Channel
Chip
plane plane
Chip
plane plane
Channel
Chip
plane plane
Chip
plane plane
…………… ……… … …
Channel
Chip
plane plane
Chip
plane plane
……… …
vSSD (Chip)vSSD (Channel) vSSD (Software)
Hundreds of vSSDs can be
supported in a single server
Impact of Hardware Isolation on SSD Lifetime
9
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
…….
Channel
Chip Chip
……… …
Flash Translation Layer
Impact of Hardware Isolation on SSD Lifetime
9
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
…….
Channel
Chip Chip
……… …
Flash Translation Layer
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Ave
rage
#B
lock
s Era
sed/s
ec
The average rate at which flash blocks are erased
Impact of Hardware Isolation on SSD Lifetime
9
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
…….
Channel
Chip Chip
……… …
Flash Translation Layer
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Ave
rage
#B
lock
s Era
sed/s
ec
The average rate at which flash blocks are erased
Impact of Hardware Isolation on SSD Lifetime
9
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
…….
Channel
Chip Chip
……… …
Flash Translation Layer
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Ave
rage
#B
lock
s Era
sed/s
ec
The average rate at which flash blocks are erased
Flash blocks wear out at different rate with different workload
Impact of Hardware Isolation on SSD Lifetime
9
Channel
Chip Chip
Channel
Chip Chip
…………… ……… … …
…….
Channel
Chip Chip
……… …
Flash Translation Layer
Write Intensive
FlashBloxChallenges
10
Chip
…
Chip
…
Chip
…
AppApp App
SSD Lifetime Performance Isolation
FlashBloxChallenges
10
Chip
…
Chip
…
Chip
…
AppApp App
SSD Lifetime Performance Isolation
Chip
…
Chip
…
Chip
…
AppApp App
SSD Lifetime Performance Isolation
FlashBloxChallenges
10
Chip
…
Chip
…
Chip
…
AppApp App
SSD Lifetime Performance IsolationSSD Lifetime
FlashBlox: Swapping Channels for Wear Balance
11
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
Adjusting the wear imbalance at a more coarse time granularity
can achieve near-ideal SSD lifetime
FlashBlox: Swapping Channels for Wear Balance
11
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
The channel that has incurred the maximum wearout
The channel that has the minimum rate of wearout
FlashBlox: Swapping Channels for Wear Balance
11
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
Channel migration takes 15 minutes, once per 19 days
Overall performance drops only for 0.04% of all the time
How Frequently Should We Swap?
12
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
Imbalance = MaxWear / AvgWear
How Frequently Should We Swap?
12
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
M
Imbalance = MaxWear / AvgWear4
App
How Frequently Should We Swap?
12
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
M M
Imbalance = MaxWear / AvgWear42
App
How Frequently Should We Swap?
12
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
M M M
Imbalance = MaxWear / AvgWear424/3
App
How Frequently Should We Swap?
12
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
M M M M
Imbalance = MaxWear / AvgWear424/31
App
How Frequently Should We Swap?
12
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
M M M M
Imbalance = MaxWear / AvgWear424/31
M
8/5
M
4/3
M
8/7
M
1
App
How Frequently Should We Swap?
12
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
M M M M
Imbalance = MaxWear / AvgWear424/31
M
8/5
M
4/3
M
8/7
M
1
App
How many times should we swap within SSD lifetime?
Quantifying the Swapping Frequency
13
Quantifying the Swapping Frequency
13
after K rounds of cycling:
Quantifying the Swapping Frequency
13
after K rounds of cycling:
Wear Imbalance = (MK + M)/(MK + M/N) = (K + 1)/(K + 1/N) ≤ (1 + x)
Maximum Wearout Average Wearout
Quantifying the Swapping Frequency
13
after K rounds of cycling:
Wear Imbalance = (MK + M)/(MK + M/N) = (K + 1)/(K + 1/N) ≤ (1 + x)
K ≥ (N – 1 – x) / (Nx)
Quantifying the Swapping Frequency
13
after K rounds of cycling:
Wear Imbalance = (MK + M)/(MK + M/N) = (K + 1)/(K + 1/N) ≤ (1 + x)
K ≥ (N – 1 – x) / (Nx)
If N = 16, x = 0.1, then K = 9, which means after swap NK = 148 times,
we can guarantee the wear imbalance is bounded in 1.1
Quantifying the Swapping Frequency
13
after K rounds of cycling:
Wear Imbalance = (MK + M)/(MK + M/N) = (K + 1)/(K + 1/N) ≤ (1 + x)
K ≥ (N – 1 – x) / (Nx)
If N = 16, x = 0.1, then K = 9, which means after swap NK = 148 times,
we can guarantee the wear imbalance is bounded in 1.1
For an SSD with 5 years lifetime, swap once per 12 days
can guarantee the channels are well balanced for worst case
Adaptive Wear Leveling in Practice
14
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
M
M/3 M/20
App App App App
Adaptive Wear Leveling in Practice
14
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
M
M/3 M/20
App App
Using erase rate as the trigger condition for swapping
App App
Intra Channel Wear Leveling
15
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
Intra Channel Wear Leveling
15
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
Chips will be swapped along with the channel migration
Chip
Intra Channel Wear Leveling
15
Channel 1
Use
d E
rase
Cyc
les
Channel 2 Channel 3 Channel 4
Chips will be swapped along with the channel migration
Chip
Intra-chip wear leveling mechanisms +
FlashBlox Architecture
16
App
Channel-Level
Wear Leveling
Flash
Resource
Manager
Chip-Level
Wear Leveling
FlashBlox Architecture
16
App
Channel-Level
Wear Leveling
Flash
Resource
Manager
App
Virtual SSD
App
Virtual SSD…
Chip-Level
Wear Leveling
Isolation, Bandwidth & Capacity Requirement
(Virtual SSD to Parallel Chips Mappings)
FlashBlox Architecture
16
App
Channel-Level
Wear Leveling
Flash
Resource
Manager
App
Virtual SSD
App
Virtual SSD…
Chip-Level
Wear Leveling
Isolation, Bandwidth & Capacity Requirement
(Virtual SSD to Parallel Chips Mappings)
FlashBlox Architecture
16
App
Channel-Level
Wear Leveling
Flash
Resource
Manager
App
Virtual SSD
App
Virtual SSD…
Chip-Level
Wear Leveling
Pay-As-You-Go Model in Cloud
Inter Channel
Swapping
Isolation, Bandwidth & Capacity Requirement
(Virtual SSD to Parallel Chips Mappings)
FlashBlox Architecture
16
App
Channel-Level
Wear Leveling
Flash
Resource
Manager
App
Virtual SSD
App
Virtual SSD…
Chip-Level
Wear Leveling
Channel Channel
Intra Channel
Swapping
Intra Channel
Swapping
Other FTL
Algorithms
… … …
Inter Channel
Swapping
Isolation, Bandwidth & Capacity Requirement
(Virtual SSD to Parallel Chips Mappings)
FlashBlox Architecture
16
App
Channel-Level
Wear Leveling
Flash
Resource
Manager
App
Virtual SSD
App
Virtual SSD…
Chip-Level
Wear Leveling
Channel Channel
FlashBlox Experimental Setup
17
14 data center workloads
16 channels
4 chips
4 planes
16 KB page size
Yahoo Cloud Service Benchmark
Bing Search / Index / PageRank
Transactional Database
Azure Storage
Tail Latency Reduction with FlashBlox
18
App1 App2
Tail Latency Reduction with FlashBlox
18
App1 App2 A: Session store recording recent actions
B: Photo tagging
C: User profile cache
D: User status update
E: Threaded conversations
F: User database
Tail Latency Reduction with FlashBlox
18
0
100
200
300
400
500
600
700
A+A A+B A+C A+D A+E A+F
99th
Perc
entile
Lat
ency
(mic
rose
cons)
Yahoo Cloud Service Benchmark (YCSB)
App1-Software Isolation App1-FlashBlox
App2-Software Isolation App2-FlashBlox
Tail latency reduction: 2.6x, average latency reduction: 1.4x
App1 App2
Impact of Channel Migration on Application Performance
19
0
0.1
0.2
0.3
0.4
0.5
0.6
Lat
ency
(m
illis
eco
nds)
Time (Seconds)
Bing Search’s Performance During Channel Migration
Without MigrationWith Migration
Impact of Channel Migration on Application Performance
19
0
0.1
0.2
0.3
0.4
0.5
0.6
Lat
ency
(m
illis
eco
nds)
Time (Seconds)
Bing Search’s Performance During Channel Migration
Without MigrationWith Migration
34%
Impact of Channel Migration on Application Performance
19
Channel migration takes 15 minutes, once per 19 days
Overall performance drops only for 0.04% of all the time
0
0.1
0.2
0.3
0.4
0.5
0.6
Lat
ency
(m
illis
eco
nds)
Time (Seconds)
Bing Search’s Performance During Channel Migration
Without MigrationWith Migration
34%
FlashBlox Summary
20
2.6x reduction on tail latency
Near-ideal SSD lifetime
Swap once per 19 days
21
Thanks!Jian Huang†
Anirudh Badam Laura Caulfield Suman Nath
Sudipta Sengupta Bikash Sharma Moinuddin K. Qureshi
†
Q&A
‡
‡
