Beyond the cloud! a small overview of cloud challenges. Credits: NASA

Beyond the cloud a small overview of cloud challenges Credits: NASA Adrien Lebre / Ascola Project Team Cumulo NumBio - June 3rd, 2015

Looking back xxx Computing Meta / Cluster / Grid / Desktop / Hive / Cloud / Sky xxx as Utility Computing A common objective: provide computing resources (both hardware and software) in a flexible, transparent, secure, reliable,... way Challenges Software/Hardware heterogeneity Security (Isolation between applications,...) Reliability / Resiliency Data Sharing... 2/33

Looking back Network of Workstations 1990 / 20xx working node Frontend (Resource Management System) 3 /33

Looking back Network of Workstations 1990 / 20xx working node Frontend (Resource Management System) Storage nodes (Distributed File System) Compute nodes 3 /33

Looking back Network of Workstations 1990 / 20xx Frontend (Resource Management System) working node Storage nodes (Distributed File System) Compute nodes 3 /33

Looking back Network of Workstations 1990 / 20xx Frontend (Resource Management System) working node Network File System Compute nodes 3 /33

Looking back Network of Workstations 1990 / 20xx Frontend (Resource Management System) working node Network File System Compute nodes Map/Reduce framework (leverage attached storage facilities) 3 /33

Looking back Network of Workstations 1990 / 20xx Frontend (Resource Management System) working node Network File System Compute nodes Long term Storage Map/Reduce framework (leverage attached storage facilities) 3 /33

Looking back Network of Workstations 1990 / 20xx The Grid 1997 / 201x working node Wide Area Network working node working node 4 /33

Looking back Network of Workstations 1990 / 20xx The Grid 1997 / 201x working node Grid File System Wide Area Network working node working node 4 /33

Looking back System virtualization: One to multiple OSes on a physical node thanks to a hypervisor (an operating system of OSes) Virtual Machines (VMs) Virtual Machine Monitor A virtual machine (VM) provides a faithful implementation of a physical processor s hardware running in a protected and isolated environment. Virtual machines are created by a software layer called the virtual machine monitor (VMM) that runs as a privileged task on a physical processor. Hypervisor Physical Machine (PM) 5 /33

Looking back System virtualization: a great sandbox VM 1 VM 2 VM 3 Isolation ( security between each VM) Hypervisor 6 /33

VM 1 VM 2 VM 3 Looking back System virtualization: a great sandbox Virus / Invasion / Crash Isolation ( security between each VM) Hypervisor 6 /33

VM 1 VM 2 VM 3 Looking back System virtualization: a great sandbox Virus / Invasion / Crash Isolation ( security between each VM) Hypervisor Snapshotting (a VM can be easily resume from its latest consistent state) 6 /33

Looking back System virtualization: a great sandbox VM 1 VM 2 VM 3 Hypervisor Isolation ( security between each VM) Snapshotting (a VM can be easily resume from its latest consistent state) 6 /33

Looking back System virtualization: a great sandbox VM 1 VM 2 VM 3 Hypervisor Isolation ( security between each VM) Snapshotting (a VM can be easily resume from its latest consistent state) VM 1 VM 2 VM 3 Suspend/Resume Hypervisor 6 /33

Looking back System virtualization: a great sandbox VM 1 VM 2 VM 3 Hypervisor Isolation ( security between each VM) Snapshotting (a VM can be easily resume from its latest consistent state) VM 1 VM 2 VM 3 Suspend/Resume Live migration (negligible downtime ~ 60 ms) Post/Pre Copy Hypervisor 6 /33

Looking back System virtualization: a great sandbox VM 1 VM 2 VM 3 Hypervisor Isolation ( security between each VM) Snapshotting (a VM can be easily resume from its latest consistent state) VM 1 VM 2 VM 3 Suspend/Resume Live migration (negligible downtime ~ 60 ms) Post/Pre Copy Hypervisor Hypervisor 6 /33

Looking back System virtualization: a great sandbox VM 1 VM 2 VM 3 Hypervisor Isolation ( security between each VM) Snapshotting (a VM can be easily resume from its latest consistent state) VM 1 VM 2 VM 3 Suspend/Resume Live migration (negligible downtime ~ 60 ms) Post/Pre Copy Hypervisor 6 /33

Looking back Network of Workstations 1990 / 20xx The Grid 1997 / 201x The Cloud 2005 / 201x working node working node working node Infrastructure as a Service model 7 /33

Looking back Network of Workstations 1990 / 20xx The Grid 1997 / 201x The Cloud 2005 / 201x working node working node WAN working node Infrastructure as a Service model 7 /33

Looking back Network of Workstations 1990 / 20xx The Grid 1997 / 201x The Cloud 2005 / 201x working node working node WAN Is The Cloud the ultimate UC model? Are there remaining concerns? working node Infrastructure as a Service model 7 /33

Where we are? IaaS challenges 8 /33

Where we are? IaaS challenges Scalability / Energy 8 /33

Where we are? IaaS challenges Scalability / Energy Reliability 8 /33

Where we are? IaaS challenges Scalability / Energy Reliability nothing really new 8 /33

Where we are? IaaS challenges Scalability / Energy Reliability nothing really new Hypervisor 8 /33

Where we are? IaaS challenges Scalability / Energy Reliability nothing really new Virtualize IT impacts performances (difficulty to guarantee performances, SLAs) Hypervisor 8 /33

Virtualization and Performance 9

Viable and Non Viable Configurations Fine management of resources (efficiency and energy constraints) Find the right mapping between needs of VMs and resources provided by PMs credits: F. Hermenier, OSDI poster session 2008 10

Fluctuations of VM Requirements 0:00 12:00 24:00 0:00 12:00 24:00 VM 1 VM 2 0:00 12:00 24:00 Static placement policies (as delivered by most of the popular Cloud Computing management systems) Simple but prevent CC providers to maximize the usage of CC resources while guaranteeing VM resource requirements Used advanced dynamic placement strategies to relocate VMs according to the scheduler objectives / available resources / waiting queue /... 0:00 12:00 24:00 VM 3 VM 4 11

Dynamic VM Placement Policies Generale idea: leverage VM capabilities to manipulate VEs in a similar way of usual processes on a laptop (a VE is a users working environment, possibly composed of several interconnected VMs) Each VE is in a particular state Perform VE context switches (a set of VM context switches) to reschedule/rebalance the LUC infrastructure 12

Placement constraints (btrplace) Current Status Correct Status Find the right mapping between needs of VMs, their constraints and resources provided by PMs Non-viable manipulations cost: 3 cost: 2 credits: F. Hermenier, Sophia Antipolis University, www.btrplace.org 13 /33

/33 14 a Small Example 110 30 90 30 30 30 30 20 50 20 90 30 30 30 30 60 90 110 110 50 80 130 10 10 50 90 30 30 30 30 120 30 30 30 30 10 10 ban ban spread spread fence 40 40 80 110

a Small Example Only CPU is considered in this simple example spread spread fence 30 30 50 30 30 30 30 30 30 80 30 80 30 30 40 10 30 10 30 30 20 50 30 30 30 10 40 90 10 110 90 20 110 130 90 90 110 60 120 ban ban 14 /33

Another with Map/Reduce Frontend Storage nodes (Distributed File System) Compute nodes Long term Storage 15/33

Another with Map/Reduce Frontend Storage nodes (Distributed File System) Compute nodes Long term Storage Map/Reduce framework (leverage attached storage facilities) 15/33

Another with Map/Reduce Frontend Storage nodes (Distributed File System) Hosting nodes Long term Storage 16/33

Virtualization and Performance Virtualization Contextualization / portability / isolation Hard to guarantee (reproductible) performances Scheduling Mainly static lead to energy wastes Dynamic scheduling strategies Good achievements but still food for researchers Other approaches : application / nested virtualisation 17

VMs make the control of performance harder but clouds are so good. Anything else? 18

The Current Trend: Large off shore DCs To cope with the increasing UC demand while handling energy concerns but credits: datacentertalk.com - Microsoft DC, Quincy, WA state 19

The Current Trend: Large off shore DCs To cope with the increasing UC demand while handling energy concerns but credits: google map - Quincy 19

The Current Trend: Large off shore DCs To cope with the increasing UC demand while handling energy concerns but credits: coloandcloud.com 19

Inherent limitations of current solutions Large off shore DCs to cope with the increasing UC demand while handling energy concerns but Denis 1. Externalization of private applications/data (jurisdiction concerns, PRISM NSA scandal, Patriot Act) 20

Inherent limitations of current solutions Large off shore DCs to cope with the increasing UC demand while handling energy concerns but Denis 1. Externalization of private applications/data (jurisdiction concerns, PRISM NSA scandal, Patriot Act) 2. Overhead implied by the unavoidable use of the Internet to reach distant platforms Internet backbone 20

Inherent limitations of current solutions Large off shore DCs to cope with the increasing UC demand while handling energy concerns but Denis 1. Externalization of private applications/data (jurisdiction concerns, PRISM NSA scandal, Patriot Act) 2. Overhead implied by the unavoidable use of the Internet to reach distant platforms Internet backbone 3. The connectivity to the application/data cannot be ensured by centralized dedicated centers (disaster recovery) 20

Inherent limitations of current solutions Large off shore DCs to cope with the increasing UC demand while handling energy concerns but Denis 1. Externalization of private applications/data (jurisdiction concerns, PRISM NSA scandal, Patriot Act) 2. Overhead implied by the unavoidable use of the Internet to reach distant platforms Internet backbone 3. The connectivity to the application/data cannot be ensured by centralized dedicated centers (disaster recovery) Hybrid platforms: a promising approach It depends how you are going to extend the private one... 20

Inherent limitations of current solutions Large off shore DCs to cope with the increasing UC demand while handling energy concerns but Denis 1. Externalization of private applications/data (jurisdiction concerns, PRISM NSA scandal, Patriot Act) 2. Overhead implied by the unavoidable use of the Internet to reach distant platforms Internet backbone 3. The connectivity to the application/data cannot be ensured by centralized dedicated centers (disaster recovery) Hybrid platforms: a promising approach It depends how you are going to extend the private one... Is there a way to address these concerns all in one? 20

µdc at the edge How and where the µdc concept can be deployed? Locality Based Utility Computing Toward LUC Infrastructures 21

Beyond the Clouds, the DISCOVERY Initiative Locality-based UC infrastructures A promising way to deliver highly efficient and sustainable UC services is to provide UC platforms as close as possible to the end-users. http://www.renater.fr/raccourci?lang=fr 22

Beyond the Clouds, the DISCOVERY Initiative Locality-based UC infrastructures A promising way to deliver highly efficient and sustainable UC services is to provide UC platforms as close as possible to the end-users. 22

Beyond the Cloud, the DISCOVERY Initiative Locality-based UC infrastructures A promising way to deliver highly efficient and sustainable UC services is to provide UC platforms as close as possible to the end-users. Leveraging network backbones Extend any point of presence of network backbones with UC servers (from network hubs up to major DSLAMs that are operated by telecom companies and network institutions). Operating such widely distributed resources requires the definition of a fully distributed system 23

The DISCOVERY Proposal DIStributed and COoperative framework to manage Virtual EnviRonments autonomously Internet backbone 24

The DISCOVERY Proposal DIStributed and COoperative framework to manage Virtual EnviRonments autonomously The LUC OS A fully distributed IaaS system and not a distributed system of IaaS systems. We want to/must go further than high level cloud APIs (cross-cutting concerns such as energy/security) Leverage P2P algorithms and self-* approaches lots of scientific/technical challenges Cost of the network? partial view of the system? Impact on the others VMs?, management of VM images? Which software abstractions to make the development easier and more reliable (distributed event programming)? How to take into account locality aspects? DISCOVERY Internet Backbone 24

The DISCOVERY Proposal DIStributed and COoperative framework to manage Virtual EnviRonments autonomously The LUC OS A fully distributed IaaS system and not a distributed system of IaaS systems. We want to/must go further than high level cloud APIs (cross-cutting concerns such as energy/security) Leverage P2P algorithms and self-* approaches lots of scientific/technical challenges Cost of the network? partial view of the system? Impact on the others VMs?, management of VM images? Which software abstractions to make the development easier and more reliable (distributed event programming)? How to take into account locality aspects??? A distributed version of the EGI Core that directly manipulates resources http://www.egi.eu/infrastructure/cloud/?? DISCOVERY Internet Backbone 24

Why not a broker? federation of clouds (sky computing) Sporadic (hybrid computing/cloud bursting) almost ready for production While standards are coming (OCCI, OVF,.), current brokers are rather limited Denis 25/33

Why not a broker? federation of clouds (sky computing) Sporadic (hybrid computing/cloud bursting) almost ready for production While standards are coming (OCCI, OVF,.), current brokers are rather limited Advanced brokers must reimplement standard IaaS mechanisms while facing the API limitation Denis 25/33

The Discovery Initiative Pros/Cons Pros Locality (jurisdiction concerns, latency-aware apps, minimize network overhead) Reliability/redundancy (no critical point/location/center) The infrastructure is naturally distributed throughout multiple areas Lead time to delivery Leverage current PoPs and extend them according to UC demands Energy footprint (on-going investigations with RENATER) Bring back part of the revenue to NRENs/Telcos Cons Security concerns (in terms of who can access to the PoPs) Operate a fully IaaS in a unified but distributed manner at WAN level Not suited for all kinds of applications : Large tightly coupled HPC workloads 50 nodes/1000 cores, 200 nodes / 4000 cores (5 racks), so1000 nodes in one PoP does not look realistic Peering agreement / economic model between network operators 26

The cloud from end-users Dan Duke Sam Paula Rob 27

The cloud in reality Dan Duke Internet backbone Sam Paula Rob 28

The Discovery Initiative Dan Duke DISCOVERY Network DISCOVERY Network DISCOVERY Network DISCOVERY Network DISCOVERY Network DISCOVERY Network DISCOVERY Network DISCOVERY Sam Paula DISC Tom Rob 29

One Step Further Dan Duke DISCOVERY Network DISCOVERY Network DISCOVERY Network DISCOVERY Network DISCOVERY Network DISCOVERY Network DISCOVERY Network DISCOVERY Sam - - Paula DISC Tom Rob 30

One Step Further Duke DISCOVERY Network - - 30

Radio Base Stations Duke DISCOVERY Network backbone Pam Paula 31

Radio Base Stations DISCOVERY Network Duke DISCOVERY Network DISCOVERY Network backbone DISCOVERY Network DISCOVERY Network Pam Paula 31

Beyond Discovery From sustainable data centers to a new source of energy A promising way to deliver highly efficient and sustainable UC services is to provide UC platforms as close as possible to the end-users and to... Leverage green energy (solar, wind turbines...) Transfer the green micro/nano DCs concept to the network PoP Take the advantage of the geographical distribution Leveraging the data furnaces concept Deploy UC servers in medium and large institutions and use them as sources of heat inside public buildings such as hospitals or universities http://parasol.cs.rutgers.edu https://www.aoterra.de 32

Beyond the Clouds Clouds give the illusion that the infrastructure disappeared but it did not Thank you / Questions? Several researchers, engineers, stakeholders of important EU institutions and SMEs have been taking part to numerous brainstorming sessions (BSC, CRS4, Unine, EPFL, PSNC, Interoute, Orange Labs, Peerialism, TBS Group, XLAB, ) http://beyondtheclouds.github.io/ adrien.lebre@inria.fr 33