NREN strategies for Data-‐Intensive Science in a Carbon Constrained
World
Bill St. Arnaud
Unless otherwise noted all material in this slide deck may be reproduced, modified or distributed without prior permission of the author
Theme of this talk • We have already lost the baIle to save the planet from extreme
climate change. Rather than focusing on reducing energy consumpKon, (MiKgaKon) we now need to focus on surviving climate change (AdaptaKon)
• Explosion of data and energy consumpKon by computers and networks is contribuKng to energy demand and CO2 emissions
• But big data and science will be criKcal as move to focus on adapKng to climate change
• How can Internet and IT help us build NRENs and support science
and educaKon that can adapt to global warming?
Changing NREN networking environment
• Global Virtual Research CommuniKes
• Increasing co-‐operaKon between public and private researchers
• Rapidly changing users demands
• Increasing potenKal of commercial ICT-‐service providers
• EducaKon: any Kme, any place, any device
• CiKzen Science and M2M communicaKons and sensors
• The disappearing campus IT & diminishing experKse in ICT centres of connected insKtuKons
Although there is less news coverage global warming has not disappeared
Half of US experienced record droughts or deluges in 2011
2010 warmest year ever – we are only at the start of the curve of the hockey s7ck. The worst is yet to come
Blame it on Canada How warming in the ArcKc affects weather in Louisiana
• Warming ArcKc slowing down jet stream • Basic Thermodynamics -‐ polar temperatures
drive the jet stream, – There’s been a 20 percent drop in the zonal
wind speeds. • As get stream slows down, it leads to those
bigger kinks in the jet stream. – That amplificaKon is associated with
persistent weather paIerns that lead to “extremes” like drought, flooding and heat waves.
• Those slow-‐moving, persistent waves of weather energy may have played a role in the big snows that hammered parts of the West last winter, as well as some of the extreme winter weather that hit South West US and Europe
• hIp://summitcountyvoice.com/2012/01/14/global-‐warming-‐revenge-‐of-‐the-‐atmosphere/
6
Climate Forecasts
MIT
• MIT report predicts median temperature forecast of 5.2°C – 11°C increase in Northern Canada
& Europe – hIp://globalchange.mit.edu/pubs/
abstract.php?publicaKon_id=990
• Last Ice age average global temperature was 5-‐6°C cooler than today – Most of Canada & Europe was
under 2-‐3 km ice
• Nearly 90 per cent of new scienKfic findings reveal global climate disrupKon to be worse, and progressing more rapidly, than expected. • hIp://www.skepKcalscience.com/pics/
Freudenburg_2010_ASC.pdf
Future Droughts • Palmer Drought Severity Index, or PDSI.
• The most severe drought in recent history, in the Sahel region of western Africa in the 1970s, had a PDSI of -‐3 or -‐4.
• By 2030 Western USA could see -‐4 to -‐6. Drought in Texas clearly caused by global warming: hIp://goo.gl/QjHRS
• By 2100 some parts of the U.S. and LaKn America could see -‐8 to -‐10 PDSI, while Mediterranean areas could see drought in the -‐15 or -‐20 range.
hIp://www.msnbc.msn.com/id/39741525/ns/
us_news-‐environment/
DramaKc changes in precipitaKon
• Every conKnent has suffered record rainfalls • Rains submerged one-‐fioh of Pakistan, a
thousand-‐year deluge swamped Nashville and storms just north of Rio caused the deadliest landslides Brazil has ever seen.
• Observed increase in precipitaKon in the last few decades has been due in large part to a disproporKonate increase in heavy and extreme precipitaKon rates which are exceeding predicKons made in models
New Challenge: Climate AdaptaKon
• Obama’s NaKonal Science Advisor John Holdren “MiKgaKon alone won’t work, because the climate is already changing, we’re already experiencing impacts….A miKgaKon only strategy would be insanity,”
• Equal emphasis given to adaptaKon – avoiding the unmanageable, and adaptaKon – managing the unavoidable.”
• Obama’s Climate AdaptaKon ExecuKve Order
– hIp://www.stumbleupon.com/su/1tU8go/www.good.is/post/obama-‐s-‐secret-‐climate-‐adaptaKon-‐plan/
Climate Change Impact on Internet and NRENs
• UK Government study Climate Change could ruin the Internet – hIp://www.grist.org/list/2011-‐05-‐09-‐climate-‐change-‐could-‐ruin-‐the-‐internet
• California aims to have 30% renewable power
– Impact on reliability of power systems
• Last year Nuclear power plants in France were forced to shut down because cooling water was too warm
• Germany is commiIed to shuvng down all of its nuclear plants • Droughts will restrict producKon of hydro-‐electric power
• Energy shortages and disrupKons are predicted to increase in the coming years
Impact on ICT sector
www.smart2020.org
According to IEA ICT will represent 40% of all energy consump7on by 2030
ICT represent 8% of global electricity consumpKon Future Broadband-‐ Internet alone is expected to consume 5% of all electricity hIp://www.ee.unimelb.edu.au/people/rst/talks/files/Tucker_Green_Plenary.pdf
R&E biggest consumer!!
Australian Computer Society Study hIp://www.acs.org.au/aIachments/ICFACSV4100412.pdf
Per employee Per sector
The ICT energy consumpKon in higher-‐ ed
• Campus compuKng 20-‐40% electrical energy consumpKon on most campuses – Studies in UK and The Netherlands – hIp://goo.gl/k9Kib
• Closet clusters represent up to 15% of electrical consumpKon
– hIp://isis.sauder.ubc.ca/research/clean-‐technology-‐and-‐energy/green-‐it/
• Campus data center alone represents 8-‐20% of electrical consumpKon – hIp://www.iisd.org/publicaKons/pub.aspx?pno=1341
• IISD study demonstrated that moving Canadian research to cloud would
pay for itself in energy savings and CO2 reducKon – hIp://www.iisd.org/publicaKons/pub.aspx?pno=1341
The real cost of campus compuKng
• Land -‐ 2% • Core and shell costs – 9% • Architectural – 7% • Mechanical/Electrical – 82%
– 16% increase/year since 2004 Source: ChrisKan Belady
Belady, C., “In the Data Center, Power and Cooling Costs More than IT Equipment it Supports”, Electronics Cooling Magazine (February 2007)
The Data Deluge
Genomic sequencing output x2 every 9 month
Climate model intercomparison project (CMIP) of the IPCC
2004: 36 TB
2012: 2,300 TB
1330 molec. bio databases Nucleic Acids Research (96 in Jan 2001)
MACHO et al.: 1 TB Palomar: 3 TB 2MASS: 10 TB GALEX: 30 TB Sloan: 40 TB Pan-‐STARRS:
40,000 TB
Source: Ian Foster, UoChicago
Big science has achieved big successes
LIGO: 1 PB data in last science run, distributed worldwide
ESG: 1.2 PB climate data delivered to 23,000 users; 600+ pubs
OSG: 1.4M CPU-‐hours/day, >90 sites, >3000 users, >260 pubs in 2010
Robust producKon soluKons SubstanKal teams and expense Sustained, mulK-‐year effort ApplicaKon-‐specific soluKons, built on common technology
Source: Ian Foster, UoChicago
But small science is struggling
More data, more complex data Ad-‐hoc soluKons Inadequate sooware, hardware Data plan mandates
Source: Ian Foster, UoChicago
Growth in sensor networks and CiKzen Science
19
Real Time Health Monitoring
Glacier Tracking
Smart Trash
THE CHALLENGE We need soluKons to address climate change, data deluge, needs of scienKsts, global collaboraKon, the evolving network of any Kme, any place, any device and yet addresses the challenge of disappearing IT on campus while sKll providing a leadership role in next generaKon Internet and broadband, and find ways to pay for it all in an era of severe fiscal constraint.
THE SOLUTION
1. Brokered Green Clouds and off site campus IT 2. Sooware Defined Networks (OpenFlow) 3. NREN naKonal wireless network 4. Global Interconnected Dynamic OpKcal Networks 5. eScience Pla|orms with next gen IdM 6. Community anchor IXPs with CDN and M2M hosKng 7. New billion dollar revolving green energy funds at many
universiKes
21
1. Brokered Green Clouds and off site CAMPUS IT
22
UniversiKes moving to eliminate IT departments
• Already many primary funcKons of IT department are being outsourced to the cloud – E-‐mail, web, DNS, research compuKng, etc – University of Western Australia has outsourced virtually all campus servers to
an external private cloud
• Even rouKng, network and firewall funcKons being outsourced to NREN – AARnet, SUnet and other NRENs offering border gateway rouKng services with
collapsed IP backbones – Sooware Defined Networks makes it easy to configure outsourced LAN – Network faciliKes can be located
• Increasingly most traffic is in/out of campus, instead of within – Social networking, P2P, Clouds, Kuali, Blackboard – Future of Campus IT – high speed opKcal network connected to WiFi/5G hot
spots with tablets – No servers, no LAN
MIT to build zero carbon data center in Holyoke MA
• The data center will be managed and funded by the four main partners in the facility: the MassachuseIs InsKtute of Technology, Cisco Systems, the University of MassachuseIs and EMC.
• It will be a high-‐performance compuKng environment that will help expand the research and development capabiliKes of the companies and schools in Holyoke – hIp://www.greenercompuKng.com/news/2009/06/11/
cisco-‐emc-‐team-‐mit-‐launch-‐100m-‐green-‐data-‐center
NREN Brokered Cloud for IT departments and Researchers
• Internet 2 Net + – Provisioning of mulK vendor cloud services leveraging the Internet2
Network and InCommon Federated AuthenKcaKon – Interoperable marketplace for services where individual insKtuKons
might procure services from a wide range of cloud services providers.
• HEFCE and JISC to Deliver Cloud-‐Based Services for UK Research
– Besides providing brokered cloud services they are also providing cloud “soluKons” for IT departments and researchers
– hIp://www.hpcinthecloud.com/hpccloud/2011-‐06-‐27/hefce_and_jisc_to_deliver_cloud-‐based_services_for_uk_research.html?utm_medium=twiIer&utm_source=twiIerfeed
• SURFnet: Community Cloud Models and the Role of the R&E network as a
broker for cloud services – hIp://www.slideshare.net/haroldteunissen/community-‐clouds-‐shared-‐infrastructure-‐as-‐a-‐service
2. So\ware Defined networks
26
GreenStar Network World’s First Zero Carbon Cloud/Internet
OpenFlow Follow the wind/Follow the sun
Cloud Manager
Host Resource
Cloud Manager
Network Manager
VM
Mantychore2
Host Resource
Canadian GSN Domain
European GSN Domain
Dynamically Configure IP Tunnel
• Shudown VM • Copy Image • Update VM Context
• Start VM
Export VM
VM VM
Internet
NoKfy EU Cloud Manager
Cloud Proxy Host Cloud Proxy
Lightpath
OpKcal switch OpKcal switch
Shared storage
Shared storage
Host
OpenFlow-‐based cloud
Open Virtual Switch (OVS)
Host
eth0 eth1
Ethernet Switch
Open Virtual Switch (OVS)
Host
eth0 eth1 Open Virtual Switch (OVS)
Host
VM VM
eth0 eth1 Open Virtual Switch (OVS)
Host
VM
eth0 eth1
VM
OpenFlow Control plane OpenFlow Data plane
OF Controller
OpenFlow Network B
OVS
eth0 eth1
Internet
eth0 eth1
OVS
VM VM VM VM
OpenFlow Network A
Green Clouds InternaKonal • GreenLight explores how researchers can take advantage of data centers linked by
high-‐speed networking in an era of carbon-‐thrioy compuKng • Recent studies migraKng virtual machines to green energy sites indicate that 100
Gb/s networks are far superior to 10 Gb/s to make this transparent. • SURFnet 7 lightpath connecKon to GreenQCloud in Iceland
SURFconecxt control of lightpath to GreenQCloud in Iceland
Future Global Network of Green Clouds interconnected by GLIF
Science Cloud CommunicaKon Services Network
• Enterprise clouds use commodity internet; computaKonal clouds for data-‐intensive science require dynamic cloud provisioning integrated with dynamic high performance.
• TransCloud: example of dynamic networking & dynamic cloud provisioning
Example of working in the TransCloud [1] Build trans-‐con7nental applica7ons spanning clouds: • Distributed query applica7on based on Hadoop/Pig • Store archived Network trace data using HDFS • Query data using Pig over Hadoop clusters [2] Perform distributed query on TransCloud, which currently spans the following sites: • HP OpenCirrus • Northwestern OpenCloud • UC San Diego • Kaiserslautern
Source: Maxine Brown
3.0 NREN Na7onal Wireless Network
32
Building a NREN wireless network • Vision: to allow students, researchers and employees to collaborate,
research, learn anyKme and anywhere they seem fit!
• Also Internet of Things – Machine to Machine communicaKons • ExisKng 3G and 4G networks cannot handle data load
– Roaming gateways prevent global seamless access – Voice centric architectures
• New mobile networks seamlessly integrate with WiFi on campus
– New Wifi 2.0 standards 802.11u allow for data handoff from 3G networks
– Eduroam can be the global authorizaKon tool – OpenFlow can be used to architect integrated soluKons from wireless
node across opKcal network
Impact of NREN wireless networks
34
• The phone is a also a sensor pla|orm • Processing is done in real Kme in the cloud
– Allowing processing that can’t be done on the device – Big data analysis
• New campus or hot spot centric architectures integraKng LTE and Wifi – See SURFnet pilot
hIp://www.surfnet.nl/en/nieuws/Pages/BackgroundarKcle.aspx
• WiFi nodes can be powered by renewable sources such as roof top solar panel over 400Hz power systems or ethernet power
The Regulatory Challenge • Today’s SIM-‐card locks user to the network • If NREN becomes a MVNO with own SIM-‐cards, users could
roam seamlessly around the globe • Only public service providers have access to IMSI-‐numbers for
SIM-‐cards
• One opKon is to lobby regulators to give R&E networks access to IMSI-‐numbers
4. Global Interconnected Dynamic Op7cal Networks
36
GLIF
37
GOLE
e-‐Research Scenario
38 Source: SURFnet
Importance of GOLEs • Increasingly more research and educaKon is internaKonal collaboraKon
– Cornell-‐ Technion announcement – US overseas university campuses in UK and elsewhere – GOLES enable direct peering of regional networks or even insKtuKons
• Many researchers need access to commercial clouds and data specialists – AUP issues ooen prevent NRENs from directly connecKng up these insKtuKons – Genomics and bio-‐informaKcs processing and climate modeling
• Many commercial research insKtuKons need access to lightpaths – GOLES provide neutral access points for interconnect to AUP free lightpaths
• Enables new services – Sooware Defined Network using Switched lambdas
39
5. eScience Plaiorms with next gen IdM
40
Towards “research IT as a service” Scientific data management as a service
GO-CatalogGO-Compute
GO-Store GO-Galaxy
GO-Team
GO-TransferGO-Collaborate
GO-User
Source: Ian Foster, UoChicago 41
SaaS services in acKon: The XSEDE vision
Globus Online: Hosted persistent services
User Team Compute
InCommon
XSEDE service providerCommercial
provider
Catalog ...
Data provider
Open Science
Grid
Transfer
...
Academic institution
2
= Standard interface
XUAS
42
Virtual OrganisaKons
NBIC
Group
Netherlands BioInformatics Centre (NBIC)
Apps.NB
IC.nl
Supporting Services • SURFfederatie • SURFteams • OpenSocial
M
y Ex
perim
ent
Pu
bMed
Grid
res.
Publ
ishe
r
N=6 N=10 N=30 Guests N=20
N=66
Virtual IdP Network
Services Storage Services
Compute
Services
Network Broker
Storage Broker
Compute Broker
Instrument
Services
Instrument Broker
CollaboraKon Infrastructure (SURFconext)
Generic Broker
AAI provisioning
groups
A6rib.
mgmt
…
Source: SURFnet
6. eScience and Big Data for CiKzen Science and Community
44
Extending science and educaKon to the community
• Community anchor Internet Exchange Points help clear the boIleneck of content peering – Co-‐hosKng of CDN caching boxes – Managed by NREN – Examples include KAREN (New Zealand), BCnet and UNINETT (Norway)
• Minimize tromboning of R&E traffic to homes and schools • Can support extension of Eduroam to community WiFi spots and/or
community last mile networks • Allows for M2M traffic and anywhere, anyKme traffic to propagate through
the community
45
Community IXP managed by NREN
7. How to pay for it all
46
$1 billion funding program • Green revolving funds are either part of a university endowment program or publicly traded
enKKes. – hIp://www.sustainablebusiness.com/index.cfm/go/news.display/id/23028
• They make investments in energy efficiency and GHG reducKon iniKaKves. Payback typically
32% • ICT can represent up to 40% of the electrical energy consumpKon at university and growing
• The obvious low hanging fruit is to move, as much as possible the closet clusters and campus data center faciliKes to commercial clouds. Next is network infrastructure such as rouKng and servers
• Other obvious money saving pracKces are to power laptop and cell phone charging staKons with roof top solar panels or micro windmills, deploy solar/wind powered WiFi nodes, and use on the move electric charging for campus uKlity vehicles, etc
• Campus IT folk and NRENs need to educate managers of such funds the IT and networking can play a much more significant role in reducing energy consumpKon and GHG emissions then tradiKonal faciliKes based soluKons
47
Cyber-‐infrastructure in a Carbon Constrained World
hIp://net.educause.edu/ir/library/pdf/ERM0960.pdf
Let’s Keep The ConversaKon Going
Blogspot
Bill St. Arnaud hIp://green-‐broadband.blogspot.com
TwiIer
hIp://twiIer.com/BillStArnaud
E-‐mail list