pNFS Update - HPC User Forum

Download Report

Transcript pNFS Update - HPC User Forum

IDC HPC User Forum Update
APRIL 16, 2012
PANASAS PRODUCT MARKETING
PANASAS OVERVIEW
 Panasas Solutions Shipping Since 2004
• ActiveStor appliances in 4th generation, 19 patents issued, others pending
 Panasas Management
• Storage-focused executive management team
• Highly experienced technical team
−
−
Dr. Garth Gibson, founder & Chief Scientist, author of seminal “Berkeley RAID Paper”
Dr. Brent Welch, Chief Technology Officer
• Near doubling of staffing in 2011
−
Major expansion in engineering and global sales presence
 VC Funded
• Intel Capital, Mohr Davidow, Carlyle Group, Centennial
Industry Recognition
Faye Pairman
Cloud Project of
the Year
2
PANASAS GROWTH
 Strong Financial Position
Revenue & Customer Growth
• Double-digit revenue growth
 Loyal, Brand Name
Customers
• >75% repeat buyers
 Global Presence
• >400 customers
• >50 countries
Worldwide Support with Over 50 Resellers
3
WHAT IS BIG DATA?
 Name to describe a quantum shift in information technology
• Big data requires new software tools and hardware systems to capture,
store, and process data in a tolerable timeframe
 Big data is not a data type, it is a data phenomenon
• The use of data to extract value
 Data only becomes big data when it creates value
•
•
•
•
Provide predictive answers to complex questions
Faster time-to-results through simulation (vs. experimentation)
Enhanced productivity by sharing data sets
Better targeting of products
4
HOW BIG DATA VALUE IS CREATED
BIG Data Application Segments
ENGINEERING COLLABORATION
ANALYTICS
Design Optimization
Process Flow
Fluid Dynamics
3D Modeling
Predictive Modeling
Decision Processing
Demographics
Behavior Analysis
SIMULATION
DATA WAREHOUSE
Genome Sequencing
Seismic Processing
Monte Carlo
Visualization
Hosting
Digitization/archive
Backup
Web 2.0
5
BIG DATA STORAGE GROWTH
 Huge growth in unstructured data
2009–14E
CAGR
10.6
CAPACITY IN EXABYTES
23%
23%
9.8
67.4
7.5
60%
42.2
6.0
28.5
4.7
10.2
2010
60%
CAGR
16.0
2011
2012
File-based (NAS)
2013
2014
Block-based (SAN)
Source: IDC 2011
1 IP-SAN market includes iSCSI, InfiniBand, Switched SAS and Fibre Channel over Ethernet markets
6
TRADITIONAL STORAGE SYSTEMS
DATA CENTER ISSUES
EXISTING STORAGE TECHNOLOGIES
PERFORMANCE
Lag improvements in processing and
networking
SCALABILITY
Not optimized for flexible
deployments required today
IT COMPLEXITY
Lack of interoperability and
manageability
IT BUDGETS
Static budgets despite exponential
growth of data
• 20-year-old file systems do not easily or inexpensively support new
data models
• Requires next-generation architectures
7
BIG DATA STORAGE REQUIREMENTS
 Global namespace
• Consolidated view of networked storage
• Global access to files
 Scale-out storage
• Storage scales by adding more devices
• Seamless and non-disruptive
 Dynamic load balancing
• Data is automatically re-balanced in the background to ensure balanced
performance
 Parallel data access
• Direct access between compute clients and data storage
• No filer heads in the data path
• Performance scales with capacity
8
PANASAS® ACTIVESTOR™
 Scale Out Storage for HPC
 Seamless scaling from 40TB to 6PB of storage
 Compute nodes see a single, unified namespace
 Scales up to 1000 storage nodes
 Fully Parallel Data Access
• Performance scales to150GB/s and beyond
• No in-band filer heads or hardware RAID controllers
to constrain performance
 Easy to Deploy, Use, and Manage
• Set up or grow capacity in under ten minutes
• Dynamic load balancing as new storage is added
 High Reliability and Availability
• Object RAID with vertical parity and parallel RAID
reconstruction limits exposure upon drive failure
• High redundancy in hardware and software
ActiveStor
10 shelves, 600TB
9
ACTIVESTOR BLADE ARCHITECTURE
Director
Blade
Storage
Blade
ActiveStor Appliance
 CPU, cache, network
 Orchestrates system activity
 Metadata services
 CPU, cache, data storage
 Enables parallel reads/writes
 Advanced caching algorithms
Full
Rack
Switch
Module





60TB per 4U chassis
Scalable to 6 petabytes
Up to 1.5GB/s per chassis
New storage integrates seamlessly
Low Total Cost of Ownership
 10GbE networking
 InfiniBand Router 2 option
for IB connectivity
 600TB & 15GB/s per 40U rack
10
DIRECTFLOW® MAXIMIZES PERFORMANCE
 DirectFlow Client
• Standard installable file system
• Enables parallel, direct client
communication to disk
• Framework for emerging pNFS
standard
Panasas DirectFlow Data Path
 Director Blades
• Namespace of virtual volumes
• Scalable metadata
(no bottleneck)
 Storage Blades
• Wide striping for large files
• Read ahead/write behind for
small files
 NFS and CIFS Access
• Fully supported for
heterogeneous client access
11
PNFS: INDUSTRY-STANDARD PARALLELISM
 pNFS is an extension included in the Network File System v4.1
protocol standard.
 pNFS enables parallel, direct access.
• From pNFS clients to storage devices over multiple storage protocols
• Moves the NFS (metadata) server out of the data path
• Most valuable for storage systems based on parallel file systems
data
pNFS
Clients
NFSv4.1 Server
Object (OSD) /
File (NFS) /
Block (FC)
Storage
12
PNFS TIMELINE
 2003: DirectFlow starts shipping as a Panasas proprietary
precursor to pNFS
 2003: pNFS born out of discussions Panasas CTO Garth Gibson
had with Gary Grider (LANL), Lee Ward (Sandia), and Peter
Honeyman (UMich/CITI)
 2004-2008: Industry level coalition built (Panasas, NetApp, Sun,
others). Collaboration via the IETF standards body make pNFS
part of NFSv4.1.
• 2010: Final IETF RFCs for NFSv4.1 (including pNFS) published
 2009-2012: Linux NFSv4.1 client and server code developed
 2012-2013: First Linux distributions with viable pNFS support
and first storage systems with pNFS server-side support
 Next up: pNFS goes into production!
13
WHY PANASAS
 Panasas ActiveStor storage addresses the Big Data challenges
found across HPC markets
•
•
•
•
High performance
Scalable
Easy to manage
Reliable
 Panasas has been behind pNFS from the beginning, with
DirectFlow available today!
14