How to evolve an SSD into an Edge Server

Transforming an SSD into a Cost-Effective Edge Server

Flash Memory Summit 2018Santa Clara, CA 1

Neil WerdmullerDirector Storage Solutions

Arm


• 28 years’ experience in embedded• Lead the storage solutions team• Work closely with the industry’s top storage suppliers• Previously in wireless at Texas Instruments• BSc in Computer Science from Portsmouth University (UK)• I enjoy brewing beer at home!

Who am I?

• Why compute is moving to the edge• Design structure of servers and SSD storage• Why run an SSD as an edge server?• Ecosystem support available


What will we cover today?

4

21BnArm-based

chips shipped in 2017

#1shipping

processor in storage devices

> 5Bnpeople using Arm-based

mobile phones

120BnArm-based

chips to date

Arm computing is everywhere

Flash Memory Summit 2018Santa Clara, CA

5

Why computation is moving to the edge


Bandwidth ReliabilityPower SecurityCost Latency


Compute:• Arm Cortex-A series or Intel x86

Memory:• DRAM

Storage:• SSD/HDD

Interfaces:• Ethernet + PCIe/SATA…

Edge serverDRAM

CPU(s)

PCIe

Netw

ork

Inte

rface

Storage

Basic Server:


Compute:• Arm Cortex-R series or Cortex-A series

Memory:• DRAM ~1GB for each 1TB of flash

Storage:• Today 256GB to 64TB… flash storage

Interfaces:• PCIe/SATA/SAS…

DRAM

CPU(s)PCIe

Flash

Basic Server:SSD storage


Compute:• Arm-based system-

on-chip (SoC)

Memory:• Shared DRAM

Storage:• Shared Flash

Interface• Ethernet…

Vs.

Edge SSD

DRAM

CPU(s)

Netw

ork

Inte

rface

Flash

Basic Server:

PCIe

CPU(s)

Netw

ork

Inte

rface

Classic Edge Server:

PCIe

DRAM

CPU(s)

DRAM

Flash

Compute:• Frontend: Host I/F + Flash Translation Layer

• Typically Cortex-R or Cortex-A series• Backend: Flash management

• Typically Cortex-R or Cortex-M series• Custom accelerators: encryption, LDPC,…

Memory:• DRAM typically >1GB for each 1TB of flash


DRAMCache + FTL Tables

Frontend Processor(s)

Host Control Flash Translation Layer

PCIe/ SATA/

SAS

Encr

yptio

n

Flas

h Pr

oces

sor C

ontr

ol

Cache / BufferManagement

Com

pres

sion

MVM

eCo

mm

and

Que

ues

Host

Con

trol

Flash SystemManagement

Block Management Wear Levelling, Bad

Block mapping...

Host / FlashAddress Translation

Backend Processor

Backend Processor

Garb

age

colle

ctio

n,

ECC/

LDPC

Err

orCo

rrec

tion/

Dete

ctio

n,Re

ad S

crub

bing

...

NAND

Fla

shNA

ND F

lash

NAND

Fla

shNA

ND F

lash

CLEALECEREWEWPR/BIDin

IDout


IDout


IDoutCLEALECEREWEWPR/BIDin

IDout

Garb

age

colle

ctio

n,

ECC/

LDPC

Err

orCo

rrec

tion/

Dete

ctio

n,Re

ad S

crub

bing

...

SSD SoC Functionality:

SSD controllers

Mem

ory

Cont

rol

Mem

ory

Cont

rol

Mem

ory

Cont

rol

Mem

ory

Cont

rol

Cortex-A series apps processor running:• High-level OS such as Linux• Networking protocols: Ethernet, TCP/IP, RDMA…• Open source applications and workloads• Standard administration functions

Edge SSD server:• Single PCB and product housing• Apps processor(s) added in to a SSD• Ethernet network interface• Internal interface to SSD controller SoC• PCIe or could be enhanced by chip-to-chip interface


Apps processor + SSD:

10N

etw

ork

Inte

rfac

e

PCIe

Inte

rnal

Inte

rfac

e

NV Memory

DRAM

Applications Processor

PCIe

Inte

rnal

Inte

rfac

e DRAM

Frontend Controller Processor

Backend Processor

NV Memory

Backend Processor

Backend Processor

NV Memory

NV Memory

Adding HLOS for an Edge SSD server

SSD Apps Processor Integration:

Flash Memory Summit 2018

Santa Clara, CA 11

Ne

two

rk I

nte

rfa

ce

PC

Ie I

nte

rna

l In

terf

ace

NV Memory

DRAM

Applications

Processor

PC

Ie I

nte

rna

l In

terf

ace

DRAM

Frontend

Controller

Processor

Backend

ProcessorNV

Memory

Backend

Processor

Backend

Processor

NV

Memory

NV

Memory

Ne

two

rk I

nte

rfa

ce

DRAM

Frontend

Controller

Processor

Backend

ProcessorNV

Memory

Backend

Processor

Backend

Processor

NV

Memory

NV

Memory

Applications

Processor

Ne

two

rk I

nte

rfa

ce

DRAM

Combined

Applications

and

Frontend

Processor

Backend

ProcessorNV

Memory

Backend

Processor

Backend

Processor

NV

Memory

NV

Memory

Separate Cortex-A series processor

• Enables any SSD (or HDD) to run Linux

• Wide performance range from Cortex-A5 to Cortex-A76

Single SoC for cost/latency reduction

• Lower latency by removing internal (PCIe) interface

• Separation of apps processor and the SSD processing

• Shared DRAM and other SoC resources

Combined into frontend/apps processor

• Hypervisor provides SSD frontend separation from Linux

• Lowest cost and tightest integration

• Lowest possible latency

• Highest internal bandwidth

Optimizations

An SSD SoC has flexible compute• And an Edge server requires storage anyway

Lowest cost• Adding ‘additional’ compute in SoC inexpensive• More, or larger, cores or additional clusters

• Custom acceleration easily integrated: ML, FPGA…• Single DRAM reduces requirements

Robust • Core SSD functionality unmodified• SSDs can include power loss protection


Edge SSD Server:

12N

etw

ork

Inte

rfac

e

DRAM


Backend Processor NV

Memory

Backend Processor

Backend Processor

NV Memory

NV Memory


Net

wor

k In

terf

ace

DRAM

Combined Applications

and Frontend Processor

Backend Processor

NV Memory

Backend Processor

Backend Processor

NV Memory

NV Memory

Why a SSD for an Edge server? (1 of 3)

Lowest power• Shared single DRAM• Minimal data copying and movement• Less protocol processing

High bandwidths• Ethernet to high bandwidth internal buses• Parallelism of wide backend processing• Removes bottleneck of PCIe lanes

Lowest latencies• No internal cross interface latencies• In-storage compute or inference all in DRAM


Edge SSD Server:

Net

wor

k In

terf

ace

DRAM



Memory

Backend Processor

Backend Processor

NV Memory

NV Memory


Net

wor

k In

terf

ace

DRAM



Backend Processor

NV Memory

Backend Processor

Backend Processor

NV Memory

NV Memory


Scalability of compute and storage• Single low power core to multiple clusters of high-

performance cores

Flexibility• One SoC that is suitable for: • Edge SSD, NVMe-oF, In-storage compute,…

• Arm NEON and Compute Library for ML…• Potential for additional Arm or 3rd party ML

Security• TrustZone isolates Linux and SSD functionality• Processing of data is all done on the drive• Decrypted data remains on the drive


Edge SSD Server:

Net

wor

k In

terf

ace

DRAM



Memory

Backend Processor

Backend Processor

NV Memory

NV Memory


Net

wor

k In

terf

ace

DRAM



Backend Processor

NV Memory

Backend Processor

Backend Processor

NV Memory

NV Memory


15

www.linaro.org


Linux ecosystem on Arm

https://www.linaro.org/


A few ‘Works on Arm’ partners

www.worksonarm.com

https://www.worksonarm.com/

• Edge SSD is an evolution not a revolution

• Edge SSD delivers with low-cost, low-power and lowest-latency

• High-performance, high-bandwidth and highly flexible

• Edge SSD and In-storage compute opens up many possibilities

• This presentation will be made available in FMS proceedings

• Please come to my second talk on Thursday at 8:30am for more information…

• COMP-301-1: “Bringing Intelligence to Enterprise Storage Drives”

Flash Memory Summit 2018

Santa Clara, CA 17

Conclusions for Edge SSD


For more information, visit storage.arm.com.

[email protected]/nwerdmuller

To learn more...

mailto:[email protected]

https://www.linkedin.com/in/nwerdmuller/

1919 Confidential © 2018 Arm Limited

Thank You!Danke!Merci!��!��!Gracias!Kiitos!감사합니다ध"यवाद

2020 © 2018 Arm Limited

The Arm trademarks featured in this presentation are registered trademarks or trademarks of Arm Limited (or its subsidiaries) in the US and/or elsewhere. All rights reserved. All other marks featured may be trademarks of their respective owners.

www.arm.com/company/policies/trademarks Flash Memory Summit 2018Santa Clara, CA

Documents

How to evolve an SSD into an Edge Server