Re-think Data Management Software Design Upon the Arrival of Storage - - PowerPoint PPT Presentation

▶
re think data management software design upon the arrival
SMART_READER_LITE
LIVE PREVIEW

Re-think Data Management Software Design Upon the Arrival of Storage - - PowerPoint PPT Presentation

Dec 17, 2022 43 likes •163 views

Re-think Data Management Software Design Upon the Arrival of Storage Hardware with Built-in Transparent Compression 07/2020 1 The Rise of Computational Storage Homogeneous Computing Heterogenous Computing Compute Networking Storage

slide-1
SLIDE 1

1

Re-think Data Management Software Design Upon the Arrival of Storage Hardware with Built-in Transparent Compression

07/2020

slide-2
SLIDE 2

2

The Rise of Computational Storage

Domain Specific Compute

Compute

FPGA/GPU/TPU End of Moore’s Law

Networking

SmartNICs 10 → 100-400Gb/s

Storage

Computational Storage

Fast & Big Data Growth

Homogeneous Computing Heterogenous Computing

slide-3
SLIDE 3

3

The Rise of Computational Storage

§ CPU & Memory I/O bottlenecks § Limited FPGAs, specific sockets required § Massive data movement § No compute parallelism

FPGA

DRAM

CPU

DRAM

… …

Controller

SSD

Flash Controller

SSD

Flash Controller

SSD

Flash Controller

SSD

Flash

to Data-Driven

§ Balanced compute & storage I/O § Multiple FPGAs, easily plug-in via storage § Minimize data movement § Maximum compute parallelism

CPU

DRAM

DRAM

… …

FPGA Flash

CSD

FPGA Flash

CSD

FPGA Flash

CSD

FPGA Flash

CSD

CSD: Computational Storage Drive

from Processor-Driven…

slide-4
SLIDE 4

4

Computational Storage: A Simple Idea

q End of Moore’s Law è heterogeneous computing

Flash Control NAND Flash FPGA

In-line per-4KB zlib compression & decompression

HW → ←SW

Driver

Low-hanging fruits

FPGA/GPU/TPU SmartNICs Computational Storage

Computational Storage Drive (CSD) with Data Path Transparent Compression

slide-5
SLIDE 5

5

ScaleFlux Computational Storage Drive

ü Complete, validated solution

ü Pre-Programmed FPGA ü Hardware ü Software ü Firmware

ü No FPGA knowledge or coding ü Field upgradeable ü Standard U.2 & AIC form factors

Multiple, discrete components for Compute and SSD Functions

SSD

CPU

FPGA

Flash Controller Flash Flash

FPGA

FC Flash Flash

CSD

CPU

Single FPGA combines Compute and SSD Functions

slide-6
SLIDE 6

6

CSD 2000: Data Path Compression/Decompression

50 100 150 200 250

1 % 9 % 8 % 7 % 6 % 5 % 4 % 3 % 2 % 1 % %

100% Reads

100 200 300 400 500 600 700

1 % 9 % 8 % 7 % 6 % 5 % 4 % 3 % 2 % 1 % %

Increasing Mix R/W Performance Advantage with Larger Block Sizes

IOPS (k) Better 170%

70/30 R/W

100% Writes

FIO: 4K Random R/W IOPS FIO: 16K Random R/W IOPS

230%

100% Write

50 100 150 200 250 300 350 400 450

1 % 9 % 8 % 7 % 6 % 5 % 4 % 3 % 2 % 1 % %

FIO: 8K Random R/W IOPS

100% Reads 100% Writes

200%

70/30 R/W

220%

100% Write

220%

70/30 R/W

100% Reads 100% Writes

220%

100% Write

2.5:1 Compressible Data, 8 jobs, 32 QD, steady state after preconditioning 2.5:1 Compressible Data, 8 jobs, 32 QD, steady state after preconditioning 2.5:1 Compressible Data, 8 jobs, 32 QD, steady state after preconditioning

IOPS (k) Better IOPS (k) Better

CSD 2000 Vendor-A NVMe

slide-7
SLIDE 7

7

Open a Door for System-level Innovations

Logical storage space utilization efficiency Physical storage space utilization efficiency

OS/Applications can purposely waste logical storage space to gain performance benefits

Data path compression

Data 0’s 4KB

Transparent compression

Compressed data Unnecessary to completely fill each 4KB sector with user data Transparent compression NAND Flash (e.g., 4TB) Expanded LBA space (e.g., 32TB) Unnecessary to use all the LBAs

slide-8
SLIDE 8

8

PostgreSQL

Normalized Performance Physical storage usage

600GB 1,200GB 100% 150% 300GB

Data 8KB/page Fillfactor (FF) Reserved for future update FF Performance Storage space Data 0’s 8KB/page

Data path compression

Compressed data

Commodity NVMe SFX NVMe FF=100 FF=50

slide-9
SLIDE 9

9

PostgreSQL (Sysbench-TPCC)

60.0% 70.0% 80.0% 90.0% 100.0% 110.0% 120.0% 130.0% 140.0%

FF100 (740GB)

Vendor-A

Normalized TPS

CSD 2000

FF100

(178GB)

FF75

(905GB)

60.0% 70.0% 80.0% 90.0% 100.0% 110.0% 120.0% 130.0% 140.0% 150.0% FF75

(189GB) FF100 (1,433GB) FF100 (342GB) FF75 (1,762GB) FF75 (365GB)

Vendor-A CSD 2000

Fillfactor Drive Logical size (GB) Physical size (GB) Comp Ratio 100 Vendor-A 740 740 1.00 CSD 2000 178 4.12 75 Vendor-A 905 905 1.00 CSD 2000 189 4.75

740GB 1.4TB

Fillfactor Drive Logical size (GB) Physical size (GB) Comp Ratio 100 Vendor-A 1,433 1,433 1.00 CSD 2000 342 4.19 75 Vendor-A 1,762 1,762 1.00 CSD 2000 365 4.82 Normalized TPS

slide-10
SLIDE 10

10

Table-less Hash-based Key-Value Store

Key space K

. . .

Hash function fKàL

. . .

4KB Key space K

. . .

Hash table

. . . . . .

4KB LBA space L LBA space L KV pairs are tightly packed in L KV pairs are loosely packed in L

Hash function fKà T

Unoccupied space

KV store purposely under-utilizes logical storage space to eliminate hash table without sacrificing physical storage utilization

slide-11
SLIDE 11

11

Table-less Hash-based Key-Value Store

ü Simple code base & high operational concurrency ü Very small memory footprint ü Absence of frequent background operations (e.g., GC and compaction) è low and consistent CPU usage

Key space K

. . .

Hash function fKàL

. . .

4KB LBA space L KV pairs are loosely packed in L Unoccupied space

We will open source and are looking for collaborations to together grow the community! Compared with RocksDB ² >2x ops/s improvement ² >2x less average CPU usage

slide-12
SLIDE 12

12

Summary

www.scaleflux.com tong.zhang@scaleflux.com

Logical storage space utilization efficiency Physical storage space utilization efficiency

Transparent compression

Unique opportunities to re-think the data management software design