Storage Virtualisation – Why? A Case Study Steve O’Donnell Global Head of Data Centres and Bridge Operations BT plc
BT Global Data Centre Capability 52 Data Centres Globally –26 Data Centres in the UK –5 Data Centres in the US –13 Data Centres in Asia Pacific –10 Data Centres in Europe 750,000 ft2 of technical space 150MW of electrical power – Same again for cooling £100M electricity bill Largest Operational Bridge in Europe Full capability from simple element monitoring to full service view Manage internal and customer systems simultaneously 34,653 elements under management 982 applications Multi vendor – MVS to Linux
Background We have a global operation driven by unrelenting pressures to reduce costs and improve effectiveness Early last year we decided to pursue a global programme of data centre consolidation to –Reduce costs –Improve effectiveness and agility –Build global centres of excellence –Deliver Disaster Recovery Capability This year we bought Radienz and have even more Data Centres As part of the programme we determined to deliver the consolidated environment into Virtualised Data Centres We wanted to virtualise our infrastructure as much as possible in order to –Maximise flexibility –Increase consolidation ratios –Take out people costs
Data Centre Infrastructure Design
Guiding Principles Keep the design simple Minimise the number of different configurations that are supported Create tools to deliver and maintain the standard build and configurations Simplify and consolidate shared infrastructure (DNS, LDAP etc..) Leverage the power of shared Enterprise Storage Use the same underlying infrastructure to support Mainframe, Mid-Range and Intel environments Use the network to simultaneously deliver high performance and security Use state of the art technologies to deliver Data Centre Virtualisation
Why deploy Enterprise Storage? A Centralised Storage Architecture is very much easier to manage than dispersed and directly connected disk subsystems –As much as 50% of an mid-range support professional’s time can be spent manipulating storage –Even with Enterprise Storage organising a re-boot when storage needs increased is time consuming and adds risk Average utilisation tends to be much higher in ES systems than in directly attached systems Building and managing a Disaster Recovery platform is much easier with Enterprise Storage We determined to use only Enterprise Storage for our new data center design –This gave us a complexity problem to address (We do allow local disks for swap space) –We could choose to deploy a single type of storage and either pay a performance or cost penalty We determined to use Storage Virtualisation to hide the underlying complexity
The Connectivity Matrix with example solutions Fibre ChanneliSCSINAS Tier 1Database with DR NT / Linux App + DR App Exes with DR Tier 2Database no DR NT / Linux App no DR UNIX home dir Tier 3ArchivePST files HSM NT home dir
Data Centre Virtualisation – Building in Flexibility We built in virtualisation by leveraging the following technologies: Massive network capacity between the Data Centres – multiple Gigabit connections – this delivers the possibility of a single data centre design Server Virtualisation for the Intel Estate –We can layer multiple virtual servers of each physical servers –We have Blade technology with virtualises network and storage connectivity into a single rack –We have VMotion which allows virtual servers to be migrated around the blade farm Storage Virtualisation with SVC – where we can deliver a single joined up storage service Automation and Orchestration Single Backup Environment with powerful inter-site capability
Consolidation A number of standard consolidation environments have been constructed to leverage powerful hardware and drive up utilisation –Oracle Platform –Microsoft SQL Server –Microsoft IIS –Windows Intel Applications using VMWare –Linux Applications –Firewall Complex – external and internal POP We are working on over individual servers and plan to reduce by 40%
Storage Design Enterprise Storage is central to the design of the new data centres – all systems will use ONLY Enterprise Storage – local storage can only be used for Swap and Page Files Enterprise Storage is provided in three flavours –Tier 1 – High Performance storage with remote protection –Tier 2 – High Performance storage with local protection –Tier 3 – low cost SATA storage Connectivity is provided in three ways –Network Attached Storage – via NetApp Filers providing centralised NFS and CIFS storage –Fibre Channel attached Storage –iSCSI attached Storage All storage is virtualised using IBM’s SVC to simplify provisioning and take out people costs
SVC Logical Configuration
Vendor Independent Storage Management We wanted to ensure best value for money going forwards for our own use As an application hosting firm we wanted to productise our own storage systems and offer them to clients We wanted the flexibility to procure the most cost effective storage arrays and not be locked into a single vendor We wanted to be able to utilise the most varied set of hosts at the front end
The Benefits of SVC Separation between the storage front end and back end allows data to be migrated transparently We can deploy low cost storage (SATA) for all applications and only migrate them to higher performance storage (FC) if the performance is unacceptable – we can migrate them transparently with no outage We can remove out of date or unreliable storage arrays with no down time We can increase the capacity of storage for a host without any down-time We can totally change the design of the back end without impacting the hosts We can integrate multi vendor storage arrays with no host compatibility impact The performance of SATA disk arrays is noticeably improved over direct attachment We can snapshot between FC and SATA disks allowing Integration and User Acceptance testing to be performed at low cost We can provide inter-site synchronous replication without expensive array based software such as SRDF
Extra Benefits of SVC – Siebel Estate Our Siebel estate was enormous with five fully separate environments to enable a structured development and testing process By making the decision only to support Enterprise Storage and using SVC we have been able to reduce the number of servers required to deliver application environment this by over 60% We keep all of the development and test environments on SATA disk ready to deploy at will The Siebel developers themselves use scripts to mount the appropriate filesystems depending on the current stage in the deployment process SVC is used to create local snapshots of the production environment which can then be deployed back into development as required For performance and other non-functional tests the SATA based data can be migrated onto faster FC based disk arrays
Points to Remember Despite SVC having a good connectivity matrix for hosts – be sure that your versions of OS integrate well with SVC –Very old versions may have issues –Understand IBM’s development roadmap for SVC Watch out for compatibility of multi-pathing and Volume Manager software with SVC –This is a warning to lab test your solution before deployment –Ensure you have a working build before getting into production roll-out –Read the compatibility guide carefully Design in resilience and the ability to take components out for upgrade –Use the SVC cluster design and multi-path connectivity to ensure that your design will survive a failure and can have firmware or software upgrades without outage –SVC is a relatively new product and upgrade cycles are fast and furious Spread the load over multiple SVC clusters in a very large environment
Summary All servers in our strategic data centres use enterprise storage through SVC It has dramatically simplified our technical architecture and taken out cost and complexity We have reduced the numbers of people managing storage and servers in the environment and productivity has increased –Recently we were able to deliver a completely new and fully configured server within an hour of the request being issued – in the past this would have taken weeks Support requests are simplified and we are looking at turning many of these into automated self service processes We are now focussing on how we can deliver this value to our customers