Netapp
@
10th TF-Storage Meeting
Wojciech Janusz, Netapp Poland Bogusz Błaszkiewicz, Netapp Poland
Agenda
Data Ontap Cluster-Mode
pNFS
FAS Market Share Trend – Revenue
NetApp 13.8% EMC 29.4% Dell 8.8% IBM 12.1% HP 10.7%432TB 144 Drives 408TB 136Drives 4,320TB 1,440 Drives 6TB Flash Cache 4,320TB 1,440 Drives 8TB Flash Cache 1,800TB 600 Drives 1TB Flash Cache 3,600TB 1,200 Drives 3TB Flash Cache 2,880TB 960 Drives 2TB Flash Cache 720TB 240 Drives 512GB Flash Cache
New Entry Systems Added to Line
Truly unified
Most efficient
Extremely flexible
FAS/V3210 FAS2040 FAS/V6240 FAS/V6280 FAS/V3240 FAS/V6210 FAS/V3270 FAS2240 NAS and SAN
Familiar Ethernet and FC infrastructures
No special client or host code
FAS/V62x0, FAS60x0,
FAS/V32x0, FAS31x0
FAS3070, FAS3040, FAS2040*
SATA, SAS, FC, and SSD storage
Third-party arrays with V-Series
Flash Cache
Data ONTAP 8.1 Cluster-Mode System
Multiprotocol Access (NFS/CIFS/FC/iSCSI/FCoE)
Virtualized Storage and Network
Data ONTAP® 8.1 Cluster-Mode System
Namespace is unchanged
as volumes move
No client code
Easy to manage
Easy to change
Seamlessly scales to
many petabytes
Access using a single NFS
mount or CIFS share
B A2A3 A1 B1 B2 A R C A4
Single NFS mount/CIFS share
Data ONTAP 8 Cluster-Mode
VS1
Data ONTAP 8.1 Cluster-Mode
Virtual Server
Logical, flexible, secure resource pool for a NAS namespace and LUNs
All data access is through a Vserver, which supports one or more
protocols
Includes FlexVol® volumes, LUNs, and logical network interfaces (LIFs)
Minimum of one Vserver required: can support 100s
SAN hosts/NAS clients
Integrated Shared Architecture
Logical Interfaces
FlexVol volumes
Vserver
LIF2 LIF1
VS2
Data ONTAP 8.1 Cluster-Mode
Multi-Tenancy
SAN hosts/NAS clients
HA
LIF2 LIF3
LIF4 LIF2 LIF1 LIF1
Workload Isolation
VS1
Vservers enable multiple storage domains sharing a common resource pool
Maintain logical separation
– A Vserver defines a domain for
volumes, LIFs, and access protocols Secure, delegated administration
100s of Vservers supported
HA
Data ONTAP 8.1 Cluster-Mode
Cluster Expansion
SAN hosts/NAS clients
Transparent Operational Flexibility
LIF2 LIF1 LIF3
LIF4
VS1
Nondisruptively grow and redistribute resources
Vserver adjusts as the cluster is seamlessly expanded
Mix and match controllers
Mix and match drive types
− SATA, SAS, FC, SSD
Third-party arrays with V-Series
Hosts 1000s of volumes
PB-sized namespace
Cluster-Mode Transparent Volume Move
B C A2 A3 C1 A1 B1 B2 A R C2 LUN LUN LUN A B C A1 A2 A3 B1 B2 C1 R LUN LUN Uninterrupted AccessContinuous data access by clients and hosts
Uses Snapshot™ technology to copy data to a new aggregate
in the background
Nondisruptively move volumes between ANY aggregates
anywhere in the cluster
Storage space savings, mirror relationships, and Snapshot copies are unchanged
NFS / CIFS / iSCSI /FC / FCoE
Cluster-Mode:
On-Demand Flexibility
The Results
Seamlessly add capacity
Rebalance resources
Rapidly deploy new system
The Challenges “Disk full” errors
Over-provisioning in anticipation of future capacity needs
Managing access to new storage
The Benefits
Nondisruptive volume movement is transparent to clients and hosts
Namespace and LUN mapping unchanged Shared storage infrastructure
Cluster-Mode:
Operational Efficiency
The Results
Virtualized tiered services
Integrated unified system
Match business priorities
The Challenges
Changing workload demands
Critical projects need appropriate resources
The Benefits
Nondisruptive volume movement is transparent to clients and hosts
Mix controllers and disk types in the same cluster
On-demand mobility for critical projects Adapt resources to meet business demand
Cluster-Mode:
Operational Lifecycle
The Results
ZERO downtime
ZERO processing interruptions
ZERO client changes
The Challenges
Upgrade an entire storage system 24x7 operation during move
The Execution
Identify affected volumes and LUNs Nondisruptively move volumes Perform technology refresh
Cluster-Mode Networking Overview
SAN / NAS Mgmt Cluster Interconnect 10GbE HA HAMassively Scalable NAS Platform
One Namespace or Many, Classes of Service
Heterogeneous cluster
– Mix of controller types in single cluster per workload needs
– Entry, midrange, and high-end platforms – Native and third-party storage
(FAS, V-Series) – Multiprotocol
Tiered storage
– Match data to disk price/ performance
– Manage multiple tiers in the same namespace or many – Examples Reference data DR mirror destination Scalable archives B C A2 A3 C1 C2 Projects A B C A1 A2 A3 B1 B2 C1 C2 C3 A1 B1 High-Speed Storage (Highest Performance Across All Workloads)
High-Capacity Storage (Lower $/GB)
Personality Semantics Authentication Identification Permissions Transport Caching Data Locality Stateful
Supports UNIX & WINDOWS Strong (Kerberos)
Stringbased([email protected]) WINDOWS like access
TCP Only File Delegations Referrals NFSv4 Stateless UNIX Only Weak(AUTH_SYS) 32 bit UID/GID UNIX based UDP & TCP Ad-hoc No Remote Access NFSv3 Stateful
Supports UNIX & WINDOWS Strong (Kerberos)
Stringbased ([email protected]) More WINDOWS friendly
TCP Only
Files & Directory Delegations* Referrals NFSv4.1
Comparison of NFSv3, NFSv4 and NFSv4.1
NFSv4.1
•
Is a minor release
of NFSv4
•
Does not modify any NFSv4
features
•
Is only in DOT8.1 c-mode
•
Bug fixes
•
Supports pNFS
pNFS
pNFS Protocol
− Minor version of NFSv4.1
− Transparent to applications
Storage Access Protocol
− files (NFSv4.1)
− blocks (FC, iSCSI, FCoE)
− objects (OSD2)
Control Protocol
− Outside of the
pNFS standard
User Applications NFSv4.1 Generic pNFS layout File layout OSD layout Block layout SUN RPC iSCS I SCSI TCP FCP RDM A pNFS Protocol Control Protocol MetadataServer Data Servers
Client
Regular NFS Vs. pNFS
pNFS with Data Ontap 8.1
Meta-data, Control, Data use the same path
NFS
Single Stack
NFSv4.1 Sessions
RHEL 6.0/ RHEL 6.1
SLES 11 SP1 (Untested)
Important fixes missing
pNFS Files
RHEL 6.2 (File support Only) – kernel 2.6.40
SLES 11 SP 2* (Files & objects)
pNFS Files and Objects
Fedora 16 (Linux-3.0
)* Internal testing of RHEL Early Access bits under way
Client NFSv4.1 and pNFS in Linux Distributions
NetApp Sets Record NFS Performance
NetApp® delivers better performance and
efficiency than Isilon
35%
more IOPS
40%
faster response time
50%
less disks
80%
fewer controller nodes
Leading Performance
FAS6240 24-node cluster with Data ONTAP
®8.1
SPECsfs2008 NFS benchmark
Big Data Market Requirements
Distributed compute farms with standard software
Simpler systems, faster to deploy
Optimized balance between compute & storage Reliable data (metadata) Very high streaming bandwidth Dense configurations - GB/s/Rack Unit
Large data sets (containers)
Simpler system configurations
Boundaryless containers (10’s PBs)
Simple access to dynamic datasets
New access
technology(CDMI) with legacy application
interfaces
E-Series Controllers
5400 2600 Host/SAN connectivity • (4) 40Gb IB • (8) 8Gb FC • (16) 8Gb FC Host/SAN connectivity • (8) 6Gb SAS • (8) 8Gb FC and (4) SAS • (8) 1Gb iSCSI and (4) SAS • (4) 10Gb iSCSI and (4) SASDrive channels / types / max
• (2) 6Gb SAS ports • SSD/SAS/SED/NL-SAS • Up to 384 drives
Drive channels / types / max
• (2) 6Gb SAS ports • SSD/SAS/SED/NL-SAS • Up to 192 drives Enclosures • (60) drive DE6600 • (24) drive DE5600 • (12) drive DE1600 Enclosures • (60) drive DE6600 • (24) drive DE5600 • (12) drive DE1600 Cache • 12 / 24 / 48 GB Cache • 2 / 4 GB Performance • 6,000 MB/s – disk reads • 3,100 MB/s – CME disk writes • 900,000 IOPS – cache reads • 350,000 IOPS – SSD reads • 150,000 IOPS – disk reads
Performance
E-series disk shelves
DE5600
2U / 24 / 2.5”
– SAS, SAS SSD drives
DE1600
– 2U / 12 / 3.5” – SAS, SAS SSD
DE6600
– 4U / 60 / 3.5”
5400 + DE6600
5400 + DE5600
2600 + DE1600
5460
5424
2612
Any controller to ANY enclosure
Insert 2 E-Series
Controller(s) into enclosure to create 1 storage system (RBOD)
E Series Power
• 1.8PB in 1 rack
• High density 4U /60
• up to 60GB/s reads
StorageGRID
®E-Series Building Blocks
E2600/5400 + DE6600 120/180TB Array E2600 + DE1660 24TB Arrays 2 x Control Nodes 2 x Storage Nodes 2 x Gateway Nodes 2 x Storage Nodes StorageGRID Software on Generic x86 ServersNetApp and StorNext
Designed for
NetApp E-Series storage
– High performance – High capacity
– Highly reliable foundation
StorNext File System
– Heterogeneous data sharing across Fibre Channel SAN clients
Mac, Win, Linux, UNIX
– Simultaneously serve LAN clients with StorNext LAN Gateway & LAN clients
StorNext Storage Manager (Optional)
– Intelligent archiving
– Data always stored on most cost-effective storage tier
NetApp Auto Support
Correlate disk latency (hot) with disk type
– 24 billion records – 4 weeks to run query
– After Hadoop implementation -> 10.5 hours
Bug detection through pattern matching
– 240 billion records – Too large to run – After Hadoop implementation = 18 hours