storage performance through standards - hpc advisory council

22
Storage Performance Through Standards Bjorn Andersson HPC Advisory Workshop, March 21-23 2011 Lugano, Switzerland

Upload: others

Post on 03-Feb-2022

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Storage Performance Through Standards - HPC Advisory Council

Storage Performance Through Standards

Bjorn Andersson

HPC Advisory Workshop, March 21-23 2011

Lugano, Switzerland

Page 2: Storage Performance Through Standards - HPC Advisory Council

Who is BlueArc

Patented Hardware Accelerated Architecture

Proven in the most demanding HPC environments

Headquarters in San Jose, CA

Highest Performing NAS in the Industry

6th Generation Product

Private Company, founded in 1998

Doubled performance and scale with each platform release

© 2010 BlueArc Corporation

Page 3: Storage Performance Through Standards - HPC Advisory Council

3

Select HPC Research Customers

Atomic Weapons Establishment

Brookhaven National Laboratory

Baylor College of Medicine

Cambridge University

Chevron

Columbia University

Cold Spring Harbor Laboratory

Commissariat a L’Energie Atomique

Cray, Inc.

Duke University

European Bioinformatics Institute

Fermi National Accelerator Laboratory

Fred Hutchinson Cancer Research Center

Genentech

Georgia Institute of Technology

HECToR Partners

Idaho National Laboratory

Jet Propulsion Laboratory

Johns Hopkins University

Lawrence Berkeley National Laboratory

Lawrence Livermore National Laboratory

Merck & Co.

Massachusetts Institute of Technology

NASA

National Cancer Institute

National Heart, Lung, & Blood Institute

National Oceanic and Atmospheric Administration

National Renewable Energy Laboratory

Oak Ridge National Labs

Ontario Institute for Cancer Research

Penn State University

Princeton University

Purdue University

Renaissance Computing Institute

RWTH Aachen

Sandia National Laboratories

Sanger Institute

Stanford University

Tokyo Tech

University of California Los Angeles

University of Michigan

University of Minnesota

Vanderbilt University

Washington University in St. Louis

© 2010 BlueArc Corp. Proprietary and Confidential

Page 4: Storage Performance Through Standards - HPC Advisory Council

4

Trend: The Rise of Unstructured Data

By 2012, over 80% of total disk storage system capacity shipped will be to support file-level data

- IDC: Worldwide Disk Storage Systems 2008.2012 Forecast

0

5000

10000

15000

20000

25000

30000

35000

40000

45000

50000

2006 2007 2008 2009 2010 2011 2012

Block Based File Based

Source: IDC, 2007

Worldwide File and Block Disk

Storage Systems, 2005-2011

(PB

)

IDC Storage

Shipment Forecast

2006 to 2012

© 2010 BlueArc Corporation

Page 5: Storage Performance Through Standards - HPC Advisory Council

The Essence of BlueArc

Build for multi-petabyte scale and HPC performance requirements

while using standards and adding features more typically found in a

general purpose file systems

5 © 2011 BlueArc Corp. Proprietary and Conf idential

Page 6: Storage Performance Through Standards - HPC Advisory Council

File Systems…

6 © 2011 BlueArc Corp. Proprietary and Conf idential

HFS+

NTFS

FAT32

UFS

ISO 9660 ext2

BFFS

AFS

HPFS

FFS

XFS CXFS

EFS

NFS DFS VxFS JFS

hfs

CacheFS

CIFS

SMB

PFS

NSS QFS

Lustre GPFS

GFS

ZFS

NAS:

NFS (standard)

Standards are good!

(… and CIFS)

Page 7: Storage Performance Through Standards - HPC Advisory Council

Getting the Most Out Of a Standard Network File System

© 2010 BlueArc Corporation

Bigger pipes

High Availability Hardware Acceleration (bandwidth&iops)

Millions of files

Advanced Tiered Architecture,

Open backend

Robust,

Scalable

Performance

Optimized

Operation

At Scale

Choice with Multiple,

Competing Implementations

Standard clients

Easy

Capacity

Scaling

User Benefits:

Simplicity and familiarity

Page 8: Storage Performance Through Standards - HPC Advisory Council

8

How is this Architected?

No compromise design:

Parallel processing of core file system

(FPGA offload engine)

Simultaneous access to CIFS, NFS

and iSCSI (pNFS mid 2011)

CPU runs data management functions

not already burdened by file system

Object File System Architecture for

Intelligent, flexible data management

(SiliconFS)

Storage virtualization and thin

provisioning

8

• Offload engine

for high speed

data transfer

• CPUs for

unburdened data

management

• Intelligent file

objects for

high efficiency

Metadata

User data

Index Map

File

8 © 2011 BlueArc Corp. Proprietary and Conf idential

Page 9: Storage Performance Through Standards - HPC Advisory Council

© 2010 BlueArc, Corp. Proprietary and Conf idential.

A Proven Approach

General

Purpose Server

Server Running

Routing Software

Dedicated File Server

Appliance with Custom OS

Server Running

CIFS/NFS Software

Software

Appliance

Dedicated Routing

Appliance with Custom OS

So

ftw

are

Pe

rfo

rma

nce

Lim

it

So

ftw

are

Pe

rfo

rma

nce

Lim

it

Hardware

Appliance

Dedicated Routing Appliance

with Routing in Hardware

Dedicated File Server

Appliance with File System in Hardware

Page 10: Storage Performance Through Standards - HPC Advisory Council

Concurrent connections or # of clients

Scaling Storage or Turning on Features

Sustained Performance at Line Rates

As concurrent connections or # of clients scale

– Performance level increases linearly

– No fall off due to CPU utilization

– Much higher level of maximum performance

– If pushed to maximum sustain 100% indefinitely

Benefits

– More users per filer

– More functions per filer

– Fewer filers & licenses

– Simplified management

Line Rate Speed

Pe

rfo

rma

nc

e

Blu

eA

rc A

dva

nta

ge

© 2010 BlueArc Corporation

Software Solution

Page 11: Storage Performance Through Standards - HPC Advisory Council

Robust NFS Performance In

dep

en

den

t o

f R

ead

/Wri

te m

ix

Scales In

dep

en

den

t o

f b

lock s

izes

© 2010 BlueArc Corporation

Page 12: Storage Performance Through Standards - HPC Advisory Council

SPECsfs®2008 Performance

Source: www.spec.org

© 2010 BlueArc Corporation

Page 13: Storage Performance Through Standards - HPC Advisory Council

Strong Nodes BlueArc File Server Platforms

Mercury 55 Mercury 110 Titan 3100

Product Class Lower Mid-range Mid-range Mid-range

Cluster Nodes 2 Up to 4 Up to 8

Max Storage Capacity 4 PB 8 PB 8 PB

NFS Throughput 700 MB/s 1100 MB/s 1200 MB/s

Performance (specSFS IOPS) 60,000 100,000 100,000

Software / File Services All software and filesystem options (NFS, CIFS, iSCSI) available

Titan 3200

High End

Up to 8

16 PB

200,000

1900 MB/s

Storage Options All BlueArc storage array options are available with each platform

© 2010 BlueArc Corporation

Page 14: Storage Performance Through Standards - HPC Advisory Council

Transparent Data Mobility That Really Works! Tiered Storage for Persistent Data

SAN

Network Storage Cluster

Automatic and transparent data migration between tiers

Rules-based policy engine reduces manual intervention

Third-party or external storage devices as an integrated tier

Reduced dependence on high performance tier for peak demands

The seamless migration of data across storage tiers within a single namespace

Ease data management and reduce costs

Back-end Storage SAN

Tier 1

• High

Performance

Tier 2

• High Capacity

Tier 0

• Solid State

Cache

Tier 3

• Deduplication

• Encryption

• Compression

• Existing NAS

Users

© 2010 BlueArc Corporation

Page 15: Storage Performance Through Standards - HPC Advisory Council

NAS Topology Comparison

External or internal Direct Attached Storage

2-way or N-way clustering

No automated storage tiering

Most Competitors

External shared switched Fabric SAN

N-way clustering

Multiple storage tiers

BlueArc

15 © 2011 BlueArc Corp. Proprietary and Conf idential

Page 16: Storage Performance Through Standards - HPC Advisory Council

Adds up to

random and

impossible to predict

workload

Example: Genome Sequencing Aggregated Workload

Instr

um

en

ts

Researchers

Applications

In-house codes

Page 17: Storage Performance Through Standards - HPC Advisory Council

Back to the Traditional Network File System

Lots of Goodness

• Proven architecture

• Enterprise features

• Open, standard protocols

• Open storage philosophy

• No Throughput Aggregation

Beyond Line Rate

BUT

© 2010 BlueArc Corporation

Page 18: Storage Performance Through Standards - HPC Advisory Council

How Does pNFS Change This?

pNFS adds parallel I/O to the NFS protocol

– Eliminates the file server bottleneck

– Provides parallel data paths, even for a single file

pNFS is part of the NFSv4.1 standard

– Approved by IETF Dec, 2008

– RFCs completed editorial review Oct, 2009

– RFC numbers issued Jan, 2010

Multiple implementations are in development

– Client software is expected to be embedded in leading OS distributions

The Only Industry Standard Parallel File System

© 2010 BlueArc Corporation

Page 19: Storage Performance Through Standards - HPC Advisory Council

BlueArc pNFS Architecture Leveraging our Technology Portfolio

© 2010 BlueArc Corporation

Page 20: Storage Performance Through Standards - HPC Advisory Council

File Systems

20 © 2011 BlueArc Corp. Proprietary and Conf idential

HFS+

NTFS

FAT32

UFS

ISO 9660 ext2

BFFS

AFS

HPFS

FFS

XFS CXFS

EFS

NFS DFS VxFS JFS

hfs

CacheFS

CIFS

SMB

PFS

NSS QFS

Lustre GPFS

GFS

ZFS

Parallel FS: Lustre, GPFS

pNFS (standard)

Page 21: Storage Performance Through Standards - HPC Advisory Council

Scale-Right Storage for Mixed Environments

Scali

ng

Up

Scaling Out

Home

Directories

Shared

Repositories

Temporary

Workspace

(Scratch)

Virtualization

Cloud

Storage

Streaming

© 2010 BlueArc Corporation

Compute

Cluster

Page 22: Storage Performance Through Standards - HPC Advisory Council

Thank You!