The Data Center of the Future Steve Duplessie Founder & Senior Analyst

The Data Center of the FutureSteve DuplessieFounder & Senior AnalystEnterprise Strategy Group, Inc.

Infrastructure services

The data center of the future

Application services

Compute services

Network services

Data services



Our reason for being The tools our clients use to do their jobs The ONLY thing our client cares about (in

terms of IT)


Application services (2) In the data center of the future,

application services are modules added, deleted and modified completely independently of the underlying infrastructure

Application services are assigned a “business value” – which tells the infrastructure the attributes necessary to comply with that service’s SLA

Application services (3) The application services layer is

COMPLETELY abstracted (and virtualized) from the infrastructure layers

Applications interface to the physical via APIs to call function – not to provide function outside of their intended use. We don’t even use volume manager anymore.




Compute services

Compute services A collection of processors – blades most

likely Interconnected through the network

services layer through both high- and low-speed interconnects – IB and Ethernet (IP) near term, who knows (or cares) long term

The “grid” is capable of looking like anything – a single machine that looks like a lot of machines or a single machine comprised of many “component” machines - or any combination

Compute services (2) Blades are: a) Disposable, b) Hot pluggable,

and c) Never obsolete As long as they are operable, they are part of a

pool of compute resources used by the “grid manager”

The grid manager controls the inventory of compute assets, assimilates virtual compute instances for periods of time, dictated by SLA requirements of the “application services manager”

The grid manager controls the liquidity of the server farm – how the server farm presents itself to the application services layer




Compute services

Network services

Network services Physical connectivity layer connecting everything to everything

• Multiple current and future technologies (IP, IB, FC)

• Multiple concurrent protocol support (FC, SCSI, iSCSI, FICON, IB)

• Smarts to know what’s what and why• Smarts to utilize, provision, and provide true

QoS (by individual application service)• Intelligence lives here

Network services (2)Intelligence in the network

• Smarts exist both here and at the data services layer, with the heavy lifting happening at the network layer

Network services (3)The network will control:

• Macro data placement (what goes on what, when)• Volume management (who sees what)• Data migration (routing)• Replication• Quality of Service to the application services layer• ILM/DLM policy execution• Backup/Recovery/DR• Security management

ESG research shows users want network-based services

14%

11%

26%

24%

11%

18%

17%

18%

18%

35%

34%

35%

36%

39%

37%

34%

45%

47%

0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%

Archiving

Storage Tiering

Backup

Snapshot

Quality of Service

Replication

Data Migration

Provisioning

Volume Management

Percent of users that wantto run in the network

Percent of users that wantto run on host or array

Source: ESG Research Report, “The Future of Network-Based Storage Intelligence,” September 2004

Network services will enable ILMDo you believe that intelligent storage networks are necessary to

implement an effective Information Lifecycle Management solution?

Don't know, 11%No, you do not need

intelligent storage networks to have an

effective ILM solution, 7%

Somewhat, intelligent storage networks would

facilitate an effective ILM solution but are not necessary, 42%

Yes, you cannot have an effective

ILM solution without intelligent storage networks,

40%

Source: ESG Research Report, “The Future of Network-Based Storage Intelligence,” September 2004




Compute services

Network services

Data services

Data services This is the physical storage layer, and the

associated resource management, movement and protection

This is where ILM/DLM is real, not fantasy Big issues:

• Change management• Data valuation• Asset classification• SLAs by application class (not infrastructure

class)• Information security

Data services – Storage tiers Tier 1: Enterprise disk (50TB)

Tier 2: FC disk (NAS and SAN) (50TB)

Tier 3: iSCSI/FC ATA/SATA/Other cheap disk (150TB)

Tier 4: Massive disk archive (1PB)

Tier 5: Deep archive tape & optical (multiple PBs)

All tiers tied together via multiple protocol, super-intelligent switching

platforms with high-end QoS capabilities (network services layer)

Common factors among tiers All tiers fault-tolerant All systems future proofed – never get

old - just get newly “tiered” Tiers may have overlapped components Access to each tier will include block,

file AND object

Tier 1 High-performance (perhaps not the

highest) Super-scale in 3 dimensions (capacity,

I/O, throughput) Mainframe AND open connectivity

Tier 2 High-performance (probably the

highest) Will need to present single system

image from small to huge – single box to many

Flexible capacity entry points

Tier 3 Low-cost, idiot-proof, automatic add,

management “free” – self-actualizing storage

ISCSI mandatory Thousands of server connections

Tier 4 Scale to multi-petabytes Massive density – super low acquisition

cost AND operating cost (power) Kills the big tape library market

Tier 5 Deep archive tape & optical – smaller

libraries, bigger fatter cheaper media Performance is irrelevant (it almost is

already) 50TB/cartridge plus Will have to have object indexing offload

right in the media – which will require standards


Grid manager

Example

App 1 App 2 App 3

Data services manager

Tier 2 Tier 3

Tier 4 Tier 5

Tier 1Tier 1 Tier 2

Network manager

Storage manager skill requirementsFor the next 5 years:

Control the baseline Build the tiered infrastructure Connect everything to everything Master physical management Create classifications of

infrastructure/data services Create the cross-functional committee to

determine application service “values” Understand costs per class

Storage manager skill requirements (2)After 5 years:

All infrastructure/data services WILL be automated – traditional storage administrations services will have gone the way of the dodo. Any manual labor job is no longer in existence. There are no more tape guys. There are no more “sys admins” either. The box is smarter at the mundane than the human.

Storage manager skill requirements (3)After 5 years:

The “Data Services Manager” (DSM) will be responsible for adding/removing/changing and integrating new technologies into the Liquid Infrastructure

The DSM will “monitor” SLAs, given to the application services layer – and make changes as needed. The DSM spends his/her time on creating and implementing POLICY changes.

Storage manager skill requirements (4)After 5 years: The upside

The upside for the DSM is the network services layer. The critical smarts WILL execute in the network – so the DSM who speaks the networking language is in the driver’s seat – and owns ALL of the strategic responsibilities of the infrastructure – which is much more valuable than owning the tactical responsibilities.

THE LESSON: Get your networking act together. Now.

This has been done before!This has been done before!

Infrastructure Services

Application Services

Compute Services

Network Services

Data Services

Questions?

Thank [email protected]

508-482-0188

The Data Center of the Future Steve Duplessie Founder & Senior Analyst

Documents