PlanetLab VINI: A Vi rtualized N etwork I nfrastructure Marc E. - - PowerPoint PPT Presentation
PlanetLab VINI: A Vi rtualized N etwork I nfrastructure Marc E. Fiuczynski, Ph.D. Princeton UniversityResearch Scholar PlanetLab ConsortiumR&D Staff Member (mef@cs.princeton.edu) What is PlanetLab? Consortium : Academic,
Princeton University—Research Scholar PlanetLab Consortium—R&D Staff Member (mef@cs.princeton.edu)
2
– Formally formed January 2004, hosted by Princeton U.
– HP and Intel as founding Industrial members
– United States Government funded (NSF and DARPA)
3
– 800+ Linux-based servers at 400+ sites in 40+ countries
4
5
– Researchers can get a set of “virtual machines” across these servers (SLICE) – In a SLICE researchers can deploy & evaluate … – … distributed systems services and applications “The next Internet will be created as an overlay in the current one” – … network architectures and protocols “The new Internet will be created in parallel next to the current one”
6
and transmit ~4TB of data per day
7
– Software, called MyPLC, is Open Source – Manages a set of (remote) machines – Manages distributed virtualization (SLICES) across machines
8
– Intel (Wireless PlanetLab) – NYU (Medical PlanetLab) – U. Melbourne (e-science PlanetLab) – EPFL Switzerland (p2p PlanetLab) – … others we don’t know about …
9
serving large files to TP broadband customers
Web Portals:
10
11
Collaborative Overlay Research Environment
Overlay test-bed based on “Private PlanetLab” Provision resources for mission critical services
Features we would like to have…
Custom hardware to optimize overlay forwarding PoP/Core collocation (nodes “inside” network) Custom hardware to optimize overlay forwarding
Target overlay research
Not just on distributed system apps More on network core architectures
Utilize both private & public environments
Local v.s. Global / Provisioned v.s. Best-Effort
Kyutech
Sendai Tsukuba Tokyo Nagano Kanazawa Nagoya OsakaKeihanna Kochi Okayama Kitakyushu Fukuoka Sapporo
Hiroshima U. Kochi-tech Osaka U. NICT Koganei NICT Otemachi
Tohoku U. Sapporo Medical U. NII
12
– Software Platform: based on PlanetLab
– Facility: Private PlanetLab with its own nodes on NLR and “new” I2 s c
BGP BGP BGP BGP
13
… continually reinventing networking architectures and protocols for the new Internet
ideas in real networks
–Implement layer 3, etc. in your slice
14
– known vulnerabilities lurking in the Internet
– addressing security comes at a significant cost
– e-Commerce increasingly depends on fragile Internet
9’s)
– an issue of ease-of-use for everyday users
15
– the whole world is becoming networked
processors – assumptions about edge devices (hosts) no longer hold
– scientists have significant bandwidth requirements
– purpose-built solutions are not cost-effective
16
– apply point-solutions to the current architecture
– replace the Internet with a new network architecture
– point-solutions result in increased complexity
– architectural limits may lead to a dead-end
17
– the Internet originally viewed network traffic as fundamentally cooperative, but should view it as adversarial
– the Internet was originally developed independent of any commercial considerations, but today the network architecture must take competition and economic incentives into account
– the Internet originally assumed host computers were connected to the edges of the network, but host-centric assumptions are not appropriate in a world with an increasing number of sensors and mobile devices
18
– the Internet originally did not expose information about its internal configuration, but there is value to both users and network administrators in making the network more transparent
– the Internet originally provided only a best-effort packet delivery service, but there is value in making processing capability and storage capacity available in the middle of the network
– the Internet originally drew a sharp line between the network and the underlying transport facility, but allowing bandwidth aggregation and traffic engineering to be first-class abstractions has the potential to improve efficiency and performance
19
– no competitive advantage to architectural change – no obvious deployment path
– simulation models too simplistic – little or no real-world experimental evaluation
– production testbeds: real users but incremental change – research testbeds: radical change but no real users
20
It is time for the research community, federal governments, and commercial sector to jointly pursue the second path. This involves experimentally validating new network architectures, and doing so in a sustainable way that fosters wide-spread deployment.
21
– scores of architectural proposals – ready to step up to the challenge of making it real
– OS virtualization and interposition mechanisms – overlay networks are maturing – high-speed data pipes in the core – fast network processors and FPGAs
– PlanetLab (as a starting point) – High-speed, geographically dispersed networks serving “real” users
22
– support experimental validation of new architectures
– provide plausible deployment path
– virtualization
– opt-in on a per-user / per-application basis
23
– PlanetLab provides “access network” with global reach
– NLR/I2 provides high-speed backbone in the United States
– each architecture (service) runs in its own slice – two modes of use
24
Slice Manager (SM) Virtualization Software x86 Server Hardware Slice Slice Slice Slice
25
26
27
28
29
30
31
Client Server NAT
wireless
32
XORP
(routing protocols)
vif1 vif2 vif0
IPv4 Fwd table
User Kernel
Filters, shapers PlanetLab VM
E-GRE tunnels
– Adds routes to copy of kernel IPv4 forwarding table – Kernel forwards packets between virtual interfaces
– Add delay and loss, constrain bandwidth
– Appear as Ethernet devices in a slice – Reduce MTU for tunneling
– Hack standard GRE tunnels to preserve MAC headers
33
– ~20Kpps IPv4 packets forwarded through Click in user-space – Approx 200Mbps max bandwidth
– ~500Kpps IPv4 packets forwarded thru per-slice Network Container – Achieves 1Gbps link rate (with plenty of CPU to spare) – Comparison point: Native Linux ~750Kpps forwarding rate
– Get: big performance improvement over original IIAS – Give up: ability to modify the data plane
34
XORP
(routing protocols)
vif1 vif2 vif0
Click
User Kernel
Filters, shapers PlanetLab VM
E-GRE tunnels
– Only implements forwarding and data plane changes – Interact with virtual devices via pcap, raw sockets
– Better performance from pcap than UDP sockets?
35
– worm and virus containment, DDoS prevention,…
– managability, fault & anomaly diagnosis, reliability,…
– functionality, evolvability, reliability, heterogeneity,…
– mobility, ease-of-use, reliability, evolvability,…
– scalability, heterogeneity, mobility,…
– performance, managability, ease-of-use,…
– performance, evolvability,…
36
– convergence of multiple architectural visions – approach to deployment succeeds – ready for commercialization
– multiple architectures co-exist – create a climate of continual re-invention
– ideas retro-fitted into today’s architecture – pursuing second path improves the odds of first path succeeding
37
technologies into the Internet
enable deployment studies of new networking ideas in real networks
– PlanetLab: http://www.planet-lab.org
– VINI: http://www.vini-veritas.net