Chair: Guy Sudron, data centre manager, Jisc.
The Jisc shared data centre service continues to generate a lot of interest, so much so that we will be launching a second shared data centre in the north of England in the coming months. This session will introduce our portfolio of data centres.
It is also an opportunity to hear from current tenants who will share their experiences of developing business cases, and re-architecting campus networks and services to fully utilise the Janet network, to support their strategies of using off-campus shared data centres.
Running order of talks:
11:30-11:55 - Jisc data centre service update
Speaker: Guy Sudron, Jisc.
11:55-12:20 - Data centre procurement and planning
Speaker: Steve Aldridge, University of Liverpool.
12:20-12:45 - Distributed storage platform at the Jisc shared data centre
Speaker: Professor Paul Jeffreys, Institute of Cancer Research.
2. Please switch your mobile phones to silent
19:30
No fire alarms scheduled. In the event of an
alarm, please follow directions of NCC staff
Dinner (now full)
Entrance via Goldsmith Street
16:30 -
17:30
Birds of a feather sessions
15:20 -
16:00 Lightning talks
5. Jisc’s role
»Procurement management and contract authority
»Investment to start up the datacentre
› Procurement costs
› Legal costs
› Connection of the datacentre to the Janet core
»Ongoing promotion & awareness to drive uptake
»Anchor tenant group facilitator
»The datacentres as core to a broader ecosystem of services
12/04/2017 Jisc shared datacentres
6. Shared datacentre (south)
»Provided byVIRTUS, located in Slough
»Connected to Janet at 2 x 400Gb
»Framework launched in September 2014
12/04/2017 Jisc shared datacentres
8. Current status SDC(South)
3 Jisc datahalls, c240 racks = 2.4MW
60/40 split HPC vs Enterprise
Projects: eMedlab, MedBio
12/04/2017 Jisc shared datacentres
9. Our new shared datacentre
»Conversations with universities based in the north of England who
had visited Slough led to:
› Full OJEU procurement, led by Jisc
› Working with 4 anchor tenants
› Contract signed in September 2016
12/04/2017 Jisc shared datacentres
10. Shared datacentre (north)
»Available now
»Provided by aql from their
new DC5 in Leeds
»Connected to Janet
at 2x100G initially
› Network live on 30th March
12/04/2017 Jisc shared datacentres
12. The Jisc shared datacentre model
»Jisc frameworks
› 5 + 5 +5
› Full OJEU procurement
»Designed for education and research
› Research and enterprise applications
› Flexibility and scalability
»Datacentres connected to the Janet core
12/04/2017 Jisc shared datacentres
13. The Jisc shared datacentre model (2)
»Commercial offering
› Very favourable rack pricing (regularly benchmarked)
› Contractual PUE
› Pass through price for power
12/04/2017 Jisc shared datacentres
14. Shared datacentre model (3)
»Flexible services
› Able to ‘mix and match’ racks of varying power density with no compromise
in terms of power or cooling
› Racks can be supplied and installed by the provider or by the customer
› Smarthands; storage and assembly space; office space etc.
› Flexible networking – layer3, layer2VLANs, dedicated wavelength
12/04/2017 Jisc shared datacentres
15. Lessons learnt
»Perceived barrier of location away from campus not an issue,
but does need support from senior management
»SDCs allow central IT to contribute to institution’s ‘green agenda’
and improve efficiency
»Don’t under estimate the need for re-architecting services
»Clear, transparent, and manageable costs
»Network latency not an issue, but architecture and
services need to be planned
12/04/2017 Jisc shared datacentres
16. Findings to date
Estates
DC capacity
Re develop
Release forT&L
Research
Shared
Institution
Attract funding
Datacentre
Flexibility
Back-up
Prime
Hybrid
Shared services
Savings
Volume
Connectivity
Procurement
Power
Soft benefits recognised
» Collaborative R&D
» Shared resource
» Shared learning
» Best practice (HPC)
» Desire for on-campus
research data
» Some have capacity
» Tech, people, RDM,
storage shared services
» Build v buy
» Latency/location
» Recognition that collaboration
(incl health) is important
» Shared infra = best use of public
monies
» VAT
» ‘herd’ of tenants to max savings
» Power, build and plant – similar costs
» Space NOT always at
a premium
12/04/2017 Jisc shared datacentres
19. Outline of Presentation
»WhyThe University of Liverpool supported the project
»What we were seeking from SDC North
»What did we learn from the process
»Where does SDC North fit in with UoL’s overall strategy
»What other services would we like to see in conjunction
with SDC North?
12/04/2017 Jisc Services and Their Part in Data Centre Strategy
20. Why UoL Supported the Project
»Recognised that the growth of IT services could not be met entirely
with on prem. solutions
»The original SDC in Slough had been a success and delivered a
service for sector tailored for their needs
»Important that we had input from as many northern universities as
possible
»The Jisc/Janet brand is strong and would encourage use in
the sector
12/04/2017 Jisc Services and Their Part in Data Centre Strategy
21. Project
»As part of a consortium it was clear that the group had more
bargaining powers than a single entity
»Universities have very diverse requirements and need
flexible contracts
»Having an offsite DC with solid Janet connections was a
major factor
»As a member of N8 it was important that we helped to support the
requirements of HPC as part of the bid
12/04/2017 Jisc Services and Their Part in Data Centre Strategy
22. What Did We Learn FromThe Process
»We were aware of the amount of time needed to support this and it
was quite a commitment
»Jisc’s procurement team are first class.The complexity of the
procurement would have been too much for an individual university
»We were able to get all our requirements addressed.This was no
mean achievement given the range of options
»Being able to have a contract for a single rack to a full HPC
installation was impressive.
12/04/2017 Jisc Services and Their Part in Data Centre Strategy
23. Plans
»The UoL’s data centre strategy covers local, colocation and
cloud services
»The colocation element is especially important. It is the one that
requires a robust agreement to meet university needs.
»Primarily we wanted a small presence for BC.The agreement allows
for this and we still benefit from reduced rates created by the larger
tenant groups
»The SDC is the basis for other shared services
12/04/2017 Jisc Services and Their Part in Data Centre Strategy
24. See
»The presence of Jisc in the DC allows the development of other
shared services
»Potential for UoL are archiving services via the DC, connectivity to
Cloud providers, a storage service etc
»Options to create collaboration with other Universities.
»Access to data sets held locally at the DC seen as advantageous.
»Potential development of other HPC services
12/04/2017 Jisc Services and Their Part in Data Centre Strategy
25. See
»Thank you for your attention – Any Questions?
12/04/2017 Jisc Services and Their Part in Data Centre Strategy
28. Slide 28, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Data Centres
ICR’s Distributed Storage Platform
at the Jisc Shared Data Centre
Paul W Jeffreys
12 April 2017
29. Slide 29, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
29
Our mission is
to make the
discoveries that
defeat cancer.
Our mission is
to make the discoveries
that defeat cancer
30. Slide 30, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Presentation Outline
1. ICR / RM - very much part of the story
2. Existing ICR Research Infrastructure
3. ICR Research Data Storage Programme
4. RDS Service: solution, rationale, design
5. Overview of RDS Service
31. Slide 31, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
1. Institute of Cancer Research
and
The Royal Marsden
32. Slide 32, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
ICR at a glance
33. Slide 33, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Academic successes
33
The ICR is ranked as the top academic research centre
in the UK. cCame first in the Times Higher Education
league table of university research quality compiled from
the Research Excellence Framework (REF 2014)
Together with
The Royal
Marsden we
are rated in
the top four
cancer centres
globally
measured by
citation impact
Joint top of the Times Higher Education table for
Innovation – based on worldwide citation of our
research in patents.
Joint top of the Times Higher Education table for
Innovation – based on worldwide citation of our
research in patents.
34. Slide 34, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Our substantial and diverse funding
34
2016
• Total income £162m
• HEFCE 12% based on research excellence
• Grant income 38%
• Legacies and donations 7%
• Invention income from our discoveries 16%
Total incoming resources 2016
12%
Higher Education
Funding
Council for England
38%
Research grants
16%
Royalty income
7%
Legacies and donations
3%
Investment and tuition fees
24%
Sale of part of our future
royalty stream
35. Slide 35, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Royal Marsden Hospital
35
The ICR’s unique partnership with The Royal Marsden and
“bench-to-bedside and back” approach means we make
discoveries and deliver clinical impact in a unique way
We have an outstanding record of research
achievement dating back more than 100 years
Together we are rated in the top
four comprehensive cancer
centres worldwide for impact on
cancer research and treatment
36. Slide 36, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
ICR and Royal Marsden Vision
Dynamically adapt treatment to individual patient over time and in response
Requires: data integration and Big Data analytics
37. Slide 37, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Making the discoveries
Our strategy to defeat cancer
37
The ICR and The Royal Marsden
have worked together on a joint
strategy covering the next five years.
Our vision
We will overcome the challenges
posed by cancer’s complexity,
adaptability and evolution through
scientific and clinical excellence,
innovation and partnership
38. Slide 38, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Making the discoveries
The four pillars of our strategy
38
2
3
Unravelling cancer’s complexity
We will comprehend the full complexity of cancer by harnessing the power of new
technologies and Big data
4
Innovative approaches
We will take on the challenge of cancer’s complexity, evolution and drug resistance through
the discovery of innovative new approaches to cancer treatment
Smarter, kinder treatments
We will move a step closer to cure by bringing personalised treatments into the clinic and
developing them for patients
Making it count
We will deliver better outcomes and improved quality of life for patients by establishing
innovative treatments, diagnostics and strategies for prevention as part of routine
healthcare
1
39. Slide 39, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Two sites in London:
Sutton and Chelsea
39
Chelsea
39
Sir Richard Doll
Building
Opened 2006
Centre for Cancer
Imaging
Opened 2015
Brookes Lawley
Building
Opened in 2003
Centre for
Molecular
Pathology
Opened 2012
Chester Beatty
Laboratories
Fulham Road
Sutton
40. Slide 40, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
ICR Research Divisions
Breast
Cancer
Research
Cancer
Biology
Cancer
Therapeutics
Clinical
Studies
Molecular
Pathology
Radiotherapy
and Imaging
Structural
Biology
Genetics and
Epidemiology
41. Slide 41, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
The London Cancer Hub
A global centre for cancer innovation
41
We plan to deliver an exceptional
environment for cancer research that
enhances the discovery of new treatments
and their development for patients.
We will provide state-of-the-art facilities, and be
joined by a multitude of high-tech enterprises in a
network of 10,000 researchers, clinical staff and
support staff all operating from one site.
The London Cancer Hub aims to create a world-
leading life-science campus specialising in
cancer research, diagnosis, treatment, education
and biotech innovation.
42. Slide 42, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
2. Existing ICR Research Infrastructure
43. Slide 43, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Existing Research Infrastructure:
End-of-life Data Storage Service
• NGS – 90%
HPC Cluster
• At Jisc Shared Data Centre: 1080 cores c. 38GFlops
• Similar capacity in Chelsea/Slough
• NGS – 70%
• 70% average use
Jisc Shared Data Centre:
• Most of HPC resource will be based in Jisc data centre at Slough
• No remaining space and power in London
Implementing new ICR Research Data Storage (RDS) Service
46. Slide 46, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Service User Requirements
Collected from researcher workshops:
• 6PiB* storage, expandable to growing research need
• High level redundancy ensuring robust solution
• Cost effective and competitively priced solution
• Rapid access to all levels of data
• Ability to ‘snapshot’ to protect against accidental loss of datasets
• Ability for researchers to manage data transfers between Tiers
• Staging Areas to guarantee access to storage from instruments
* PiB is a binary unit of storage equivalent to 1.13PB (the decimal version)
47. Slide 47, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Research Data Growth
“Imaging instrumentation is currently in a period of rapid technological
advance. It is therefore difficult to project the likely data acquisition volumes
beyond a relatively short predictive horizon (for example 6 – 9 months)”
48. Slide 48, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Service Decisions
New type of leading, innovative, cost-effective service for ICR
Decisions made in partnership with researchers:
• Single service for research data under active use
• Service with two tiers (best and most cost-effective)
• Choice of solution / supplier
• Replication / resilience
• Expandability / flexibility
• Disk only
• No “backup”
• Single namespace; straight forward to use
49. Slide 49, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
4. RDS Service
solution, rationale, design
50. Slide 50, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Service Schedule
24 Mar 2016: Invitation to Tender (ITT) published
29 April 2016: ITT exercise complete
• Different types of solution offered
• Best and final offers requested
31 May 2016: Preferred supplier chosen
• OCF / DDN
• GPFS and Object Store
• Over three locations
12 Aug 2016: Contract signed with OCF/DDN
7 Dec 2016: Service delivered to users (Tier 1)
51. Slide 51, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Service – Location
Slough
ChelseaSutton
Geographically distributed system to increase level of
protection for ICR data
Service located in Sutton and two different data halls
in Slough
10Gbps private network between sites
Each site also has a 10Gbps JANET connection to
the internet
Network traffic can route around the failure of any
inter-site link
52. Slide 52, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Service Network
53. Slide 53, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Service – High Level Architecture
54. Slide 54, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Service – Conceptual Design
55. Slide 55, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Service - Principles
Tiered and disk-only system:
Tier 1: mirrored, fast storage for instant access to
files by large numbers of clients
Tier 2: less performant, but more scalable object
storage for large volumes of data; less frequently
used data will mostly be on Tier 2
Able to offer direct access to both tiers from
desktops, servers and HPC
To the end-user it looks like one big “bucket” of
storage – complexity is hidden
56. Slide 56, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Tier 1 Characteristics
• Located at Jisc Shared Data Centre (Slough) and Sutton
• Two Gridscaler 7ks + GPFS Tiebreaker
• Allows failover to surviving site without disruption
• Stretched (Mirrored) GPFS parallel filesystem with single namespace
• SSD cache
• 8TB HDDs
• MEDIAScaler provides clustered NFS and SMB
• 4 NAS* gateway servers serving as GPFS nodes (2 Sutton, 2 Slough)
• 2PiB usable capacity (really 4PiB)
• No SPFs
* Growing to 8
57. Slide 57, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS WOS Bridge Characteristics
• Based on HSM rules
• In conjunction with user, over ridable rules
• Selected files and directories will be migrated to the long-term storage
component
• Bridge rules for each research team (in principle)
• WOS Bridge uses synchronization service to exchange name space and data
changes between Tier 1 and Tier 2 objects that have been migrated to the
long-term storage
• Migrated file data is protected and replicated according to the target WOS
policy of the WOS Bridge rules
58. Slide 58, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Tier 2 Characteristics
• Located at Jisc Shared Data Centre (Slough) in two data halls, and Sutton
• Based on DDN Object Storage system
• 3-site WOS cluster based on 12 x WOS9660 storage appliances
• 4 per site
• 12 x WOS9660 appliances communicate over standard TCP/IP and expose a
single global namespace
• Global ObjectAssure used (erasure coding across 3 sites)
• Hierarchical erasure-coding scheme encoding object
• Expansion factor – 1.875 times
• Designed for cost-effective resiliency
• Allows a site offline + additional 2 disk failures in the remaining appliances
without impacting data availability
• 4.3PiB usable capacity (really 8PiB)
59. Slide 59, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Service – WOS Bridge Example
Example data flow:
1. Copy to Sutton T1
2. Mirror to Slough T1
3. File not used for a Month
4. Create on object
5. Stub files
File
Object
File FileStubStub
60. Slide 60, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Current Use of Tier 1
0
100
200
300
400
500
600
700
RDS Data Storage Use (TB) by Date
Total Use
61. Slide 61, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
5. Overview of RDS Service
62. Slide 62, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
RDS Service Status
RDS
• Tier 1 - live 7 December 2016
• Tier 2 - operating from March 2017
• Bridge in process of being switched on, team by team
• Daily snapshots being taken
• Service has run without interruption
Decision to use Jisc Shared Data Centre
• No remaining room nor power in London
• Needed geographical separation
• Used data halls 1 and 3 (for Global ObjectAssure)
• Reliable, resilient, well managed data centre
Intention to grow RDS Service, add other elements of RDS programme
63. Slide 63, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Thank You
and
Questions
paul.jeffreys@icr.ac.uk
64. Slide 64, Prof P.W.Jeffreys, ICR, Networkshop, April 2017
Cancer Research Centre of Excellence
An ICR and Imperial partnership
12
• The ICR and Imperial College London have
developed a strategic partnership through the
creation of a virtual
Cancer Research Centre of Excellence.
• The partnership will harness the complementary
expertise of both partners to deliver an
enhanced and synergistically world-leading
programme of cancer research.
• Pooling resources and expertise, and working
together in strategic partnership, will enhance
the ICR’s basic and translational research.
ICR scientists will gain
access to the extensive
resources and expertise of a
multi-faculty university in new
areas, including:
• basic life sciences
• medicine
• physical sciences
• bio-engineering
Jisc need to establish an internal business case to secure the investment needed – hence we need to know that there is the requirement and the commitment from the ATs.
Investment will be clawed back through access charges for networking. c£1.2m to establish the network
In the pipeline: UCA, Reading, Goldsmiths, Exeter College, OU,
Rapid growth. Mention spread of tenants, hown etworking isn’t the issue that it is perceived to be. Mention networking solutions in place, eg Bristol, Sanger and Sussex.
No requirement yet that hasn’t been able to be met – space, power, cooling, security etc
All this since Jan 2015 when Kings were the first to move in.
It’s this rapid take up that makes it attractive to the providers when bidding, hence we need accurate and honest forecasts, and also a substantial requirement to make it worthwhile
Leeds expected to move in during summer, Sheffield shortly afterwards. Jisc will be live in there within a couple of weeks.
Why the need for long term frameworks
Timescale to procure – how does this fit in with institutions needs?
We lead the procurement on behalf of the sector – but need resource from Ats
Stress flexibility and high density – many providers in the north couldn’t meet our demands. A select pool willing to bid and take the risk.
Why ours are different from the ‘commercial’ providers – HPC, flexibility etc, Janet connectivity
Due to the initial forecasts we were able to have enough ‘clout’ to achieve the following. Stress need for core group of ATs and high demand to make it worthwhile.
Talk about need for committed anchor tenants and realistic requirements in terms of space and power to make it worthwhile, brief history of the 2 frameworks, how the tenants worked with us
Mention the initial requirements:
South – 6 ATs with a requirment of c3.5MW (low) and 5MW (high) within 5 years
North – 4 ATs with a forecast of c2MW (low) and 5MW (high) within 5 years
Enabled us to drive the procurement
Mention the powerful framework that protects the tenants, and the SLAs, eg can be in Slough or Leeds within 4 weeks of signing the MSA.
We’d ideally need an AT group of minimum of 4, with significant requirements of c2MW within a couple of years
Good morning, my name is Paul Jeffreys
There are five parts to the talk
Most important, that Jon will give, are 3 and 4
-----------------------------
< Explanation of charging mechanism, showing that reasonable >
First pillar:
Unravelling Complexity: harnessing power of new technologies and Big Data
Foundations:
“Launching a new digital programme”
“Review of all our research infrastructure requirements”
Digital ICR Review:
Digital transformation in partnership
Directed by researchers, for researchers
< Vision for RDS Programme>
This slide shows our Research Data Storage Programme
- Existing infrastructure: HPC + scratch + instruments + sources of data
- Middle box – Jon will describe; has two tiers
- Green – future –external sharing service and repository (long term guaranteed storage)
It is tricky to predict our growth in data storage; especially we cannot say what our storage would be now if we had a decent central service
Healthcare data generally is predicted to double every 18 months to 2 years
Interesting quote from internal document from institute we know well … imaging data will become important contribution
For me, the most telling observation is – Sanger are committed to increasing their storage by at least 30% every year; currently at 30PB
Those of you anticipating significant storage need – we would really like to hear from you please
New type of leading, innovative, cost-effective service for ICR - Partnership
Decisions made
Others to be made…
Please give input through the PB faculty representatives, directly to us (lunch), by email (contact details on final slide)
The proposed solution is a highly scalable clustered NAS implementation that exports the clustered file system through industry standard protocols such as NFS and CIFS. Each NAS gateway simultaneously exports the file system thereby not limiting the performance to a single file performance.
The two NAS gateway servers are GPFS nodes that interact via a POSIX interface to the underlying high performance clustered GPFS file system. GPFS filesets provide a way to partition the file system namespace to allow administrative operations at a finer finer granularity than that of the entire file system.
The live storage component uses a single GPFS file system that implement a single name space view across two geographically disperse sites by means of a synchronous file system replication of data and metadata. A secondary copy of each file system block is maintained through physical placement of individual copies to distinct disk failure groups.
Snapshots of individual fileset provide a point-in-time view of data that allows easy recovery from common problems such as accidental deletion of a file, and comparison with older versions of a file. The ".snapshot" directories will be mounted on the NAS gateway servers.
Based on HSM rules and in conjunction with user overridable rules selected files and directories will be migrated to the long-term storage component.
GRIDScaler Bridge uses synchronization service to exchange name space and data changes between federated Live and LTS GRIDScaler clusters that have been migrated to the long-term storage
Migrated file data is protected and replicated according to the target WOS policy of the GRIDScaler WOS Bridge rules.
With multiple GS7K gateway nodes, distributed users can collaborate across geographic locations using NAS protocols. Compliance Control sets additional time for retaining each deleted object in WOS.
LTS Access NFS & SMB gateway deployed at Sutton
LTS Access NFS & SMB gateway deployed at Slough DH1.
Is WOS access available immediately?
< Explain to Faculty configuration and workflow >
< Overall impression is that chosen to ensure stability, flexibility and reliability >
-----------------------------
< Explanation of charging mechanism, showing that reasonable >