Storage Virtualization Architectures 2002 Cisco Systems Inc

Storage Virtualization Architectures © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 1

Objective After you complete this lesson, you will be able to describe the different options available for implementing SAN-based virtualization and data-sharing. © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 2

Outline • What is SAN-Based Virtualization? • How Does Virtualization Work? • Symmetric and Asymmetric Virtualization • Virtualization Implementations • Tape Virtualization • What is SAN Data Sharing? • The Direct Access File System • Symmetric and Asymmetric Data Sharing © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 3

What is SAN-Based Virtualization? The SNIA Shared Storage Model Application File/Record Subsystem File Management Storage Domain Virtualization Database File System Host-based Block Aggregation Network-based Device-based Storage Devices (disks, arrays) Block Subsystem © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 4

What is SAN-Based Virtualization? (cont. ) File/Record Layer File Servers Host with LVM Can implement storage network, block aggregation inside the box FC Host-Based Aggregation Network-Based Aggregation Device-Based Aggregation dehcatt. A-NAS egarot. S kcol. B CF © 2002, Cisco Systems, Inc. All rights reserved. CF Block Layer ESAN ACCELERATE Block 3— 5 ? NAS

How Does Virtualization Work? Manual provisioning Fixed allocation Coarse granularity Automatic provisioning Dynamic allocation Finer granularity Subsystem and OS LVMs SAN-Based Virtualization © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 6

How Does Virtualization Work? SAN-Based Virtualization Features Virtualization layer © 2002, Cisco Systems, Inc. All rights reserved. • Create and expand virtual disks on the fly • Dynamically reconfigure storage with no loss of availability • Ideal for heterogeneous server environments • Maximize performance by intelligently distributing data across spindles • Administrators can focus on applications’ storage requirements instead of managing physical disks ESAN ACCELERATE Block 3— 7

Symmetric and Asymmetric Virtualization Symmetric architecture: • Control data (metadata) uses same path as data • In-band Asymmetric architecture: • Metadata flows across LAN • Data flows across SAN • Out-of-band © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 8

Symmetric and Asymmetric Virtualization (cont. ) Symmetric: • Data and control info use same path • Data I/O routed through storage manager: LAN FC FC FC Data request SAN Storage manager Data FC FC © 2002, Cisco Systems, Inc. All rights reserved. FC – Appliance – Router – Switch • Servers might not need proprietary interface • Performance challenges ESAN ACCELERATE Block 3— 9

Symmetric and Asymmetric Virtualization (cont. ) Asymmetric: • Data and control info use different paths LAN Data request • Data I/O direct through SAN FC FC FC • Metadata through LAN or SAN Metadata manager SAN • Servers need interface to access metadata manager Data FC FC © 2002, Cisco Systems, Inc. All rights reserved. FC ESAN ACCELERATE Block 3— 10

Virtualization Implementations: Array-Based Array-based: LAN • Virtualization manager implemented in array firmware FC FC FC • Array controllers cooperate to virtualize storage • Typically symmetric • Good performance • Easy to implement and manage FC FC FC Virtualization manager • Does not support low-end RAID or JBOD storage • Does not support multivendor environments © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 11

Virtualization Implementations: Host-Based Host-based: LAN Virtualization manager FC FC FC • Virtualization engine implemented as distributed software installed on each host • Typically asymmetric • Low-cost solution • Scalability and performance issues FC FC © 2002, Cisco Systems, Inc. All rights reserved. FC • Supports multivendor environments ESAN ACCELERATE Block 3— 12

Virtualization Implementations: Appliance-Based Appliance-based: LAN • Dedicated server appliance manages virtualization process FC FC • Symmetric or asymmetric FC • Performs better than host-based Virtualization manager FC FC © 2002, Cisco Systems, Inc. All rights reserved. • Requires host agents • Might support multivendor environments FC ESAN ACCELERATE Block 3— 13

Virtualization Implementations: Switch-Based Router or switch-based: LAN • Module in fabric switch implements virtualization engine FC FC • Symmetric FC • More scalable and robust than appliance- or host-based solutions Virtualization manager FC FC © 2002, Cisco Systems, Inc. All rights reserved. • Likely to support multivendor environments FC • Switch-based implementations do not yet exist ESAN ACCELERATE Block 3— 14

Tape Virtualization In a standard backup architecture, tape drives are associated with hosts: • Limits utilization of tape drives (time and capacity) • Increases TCO of backup solution • Reliability issues Tape Library Backup server SAN © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 15

Tape Virtualization (cont. ) Tape virtualization: • Pools tape resources • Presents logical tape drives to hosts • Managed by backup software Tape Library Backup server SAN © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 16

Tape Virtualization (cont. ) • Improves tape and library utilization • Leverages investment in tape resources • Allows heterogeneous platforms to share tape resources • Enables tape drive failover • Can associate multiple physical drives with a single logical drive—tape multicasting © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 17

What is SAN Data Sharing? Linux Windows SAN • SANs support shared devices • What’s needed: shared data Logical volumes © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 18

What is SAN Data Sharing? (cont. ) Data-sharing techniques: • Database server: – Limited applications – Scalability and performance issues • Read-only snapshots: – Timeliness of data is an issue – Good for backup, batch processes, and data mining • Server clusters: – General-purpose and application-specific – Most clusters do not permit true data-sharing • True SAN data-sharing: – One copy of data accessed by multiple hosts – Supports heterogeneous platforms © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 19

What is SAN Data Sharing? (cont. ) Synchronization Linux Windows ? Application compatibility Common file system and file-level protocols SAN Distributed file and record locking © 2002, Cisco Systems, Inc. All rights reserved. Security ESAN ACCELERATE Block 3— 20

The Direct Access File System DAFS is a “SAN-aware” file system: • Optimized for use in shared-storage environments • Requires low-latency, high-throughput networks like FC, Gig. E, and Infini. Band • Built on the Virtual Interface (VI) architecture—RDMA Key features: • • • Performance and reliability in SAN environments Concurrent access to shared storage volumes Security Platform-independence Low network overhead, latency, and client CPU load © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 21

The Direct Access File System (cont. ) User Local file system Buffers DAFS Application NFS Application Buffers DAFS API Kernel FS switch Buffer cache File system Buffer cache VIPL NFS Control Data SCSI driver HBA driver Hardware FS switch HBA © 2002, Cisco Systems, Inc. All rights reserved. TCP/IP Packet buffers NIC driver NIC VI NIC driver VI-enabled NIC ESAN ACCELERATE Block 3— 22

Symmetric and Asymmetric Data Sharing Symmetric (in-band) Asymmetric (out-of-band) Request Metadata FC FC FC FS controller FC FC Request Data FS SAN controller FC Data FC FC File system © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 23

Asymmetric Data Sharing Example— Tivoli SANergy Example—IBM Tivoli SANergy: • Host-based, out-of band data sharing • Allows heterogeneous hosts to share a single volume or file on SAN-attached storage • Supports Windows, UNIX, Linux, AIX, Mac. OS • Transparently supports Oracle, DB 2, MS SQL, Exchange © 2002, Cisco Systems, Inc. All rights reserved. SANergy ESAN ACCELERATE Block 3— 24

Asymmetric Data Sharing Example— Tivoli SANergy (cont. ) Metadata Controller (MDC)— controls the file system: • MDC mounts the file system locally LAN • Uses MDC’s native file system and disk format (NTFS, UFS, EXT 2 FS) • Controls authentication & physical layout SAN • Ensures consistency of stored and cached data © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 25

Asymmetric Data Sharing Example— Tivoli SANergy (cont. ) Other hosts mount and use the volume as a normal network drive using: LAN • SMB/CIFS • NFS SAN © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 26

Asymmetric Data Sharing Example— Tivoli SANergy (cont. ) 1. Host requests access from MDC 2. MDC authenticates, returns handle 3. MDC returns metadata containing physical location of file 4. Data payload transferred directly between host and disk over SAN © 2002, Cisco Systems, Inc. All rights reserved. LAN 1 2 3 SAN 4 ESAN ACCELERATE Block 3— 27

Asymmetric Data Sharing Example— Tivoli SANergy (cont. ) SANergy features: • Heterogeneous sharing of NTFS, UFS, QFS, and EXT 2 file systems • High-availability option: – Automated SANergy NT/W 2 K/UNIX fail-over – Automated client-mapped drive fail-over • Best suited for large files (>100 KB) : – Opens up server bottleneck – Reduces CPU loading – Provides high throughput to the client © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 28

Lesson Review 1. What is different between standard RAID and SANbased virtualization? 2. How is tape virtualization implemented? 3. What type of information does a file system controller send to a host in an asymmetric SAN data sharing implementation? © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 29

Lesson Review (cont. ) 4. A large bank has implemented an AVVID infrastructure. They plan to implement a collapsed SAN fabric design next year and want to have SAN management software that can handle heterogeneous hosts and storage devices. What kind of virtualization implementation do you recommend for this customer? © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 30

Summary • Virtualization is a block-level function that can be implemented in the storage device, the host, or the network • SAN-based virtualization engines dynamically and nondisruptively partition and allocate storage media. • Symmetric virtualization is an in-band implementation where data and metadata share the same SAN data path • Asymmetric virtualization uses an out-of-band virtualization manager • Virtualization engines can be implemented in the array, the host, a dedicated appliance, or a switch or router • Backup software can support tape virtualization better allocation of tape resources and multicasting © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 31

Summary (cont. ) • Data sharing is a file-level function (e. g. DAFS), and can be implemented in-band or out-of-band • In symmetric data sharing, there is no direct communication between the server and the storage; all communication goes through the metadata controller • In asymmetric data sharing, the metadata controller authorizes communication, and the server and storage then exchange data directly © 2002, Cisco Systems, Inc. All rights reserved. ESAN ACCELERATE Block 3— 32