1 / 19

On-Demand e-Science Grids based on Adaptive-Mesh Network Architecture

On-Demand e-Science Grids based on Adaptive-Mesh Network Architecture. NEP Virtual Organization: - Canadian Light Source Inc. (CLS) - Optimum Communications Services, Inc. (OCS) - TRLabs - CYBERA Mark Sandstrom, President, OCS mark@optimumzone.net.

totie
Download Presentation

On-Demand e-Science Grids based on Adaptive-Mesh Network Architecture

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. On-Demand e-Science Grids based on Adaptive-Mesh Network Architecture NEP Virtual Organization: - Canadian Light Source Inc. (CLS) - Optimum Communications Services, Inc. (OCS) - TRLabs - CYBERA Mark Sandstrom, President, OCS mark@optimumzone.net

  2. Adaptive-Mesh for On-Demand E-Science Grids • The goal: To enable dynamic, high-performance and cost-efficient (i.e. practical) location-independent scientific collaboration • Pilot applications to connect CLS’ users, CYBERA computing facilities and TRLabs’ member universities via a realtime self-optimizing, self-organizing Adaptive-Mesh grid network • Examples of current CLS’ projects demanding bursty, yet high-performance network connectivity: • Protein crystallography (CMCF); • Materials sciences (VESPERS) • Need to dynamically integrate cross-disciplinary virtual labs over WANs Need For On-Demand E-Science Grid Networks

  3. The Platform • Applications: • CLS, CLS’ users, TRLabs member universities • Innovative grid technologies: • OCS’ Adaptive-Mesh • E.g. UCLP based way to form Adaptive-Mesh groups on-demand • Network and computing facilities: • TRnet • CYBERA (super computing and data storage)

  4. Proof of Concept: Adaptive-Mesh over TRnet Adaptive-Mesh networks can extend to campus networks and beyond to frequent (int’l) collaborators Saskatoon: - CLS Synchrotron facility Edmonton - Univ. of Alberta: nanotech research -OCS: Adaptive-Mesh remote management - TRLabs: TRnet remote management Adaptive-Mesh grid for remote, on-demand scientific collaboration Winnipeg - Univ. of Manitoba: medical research with CLS using high bandwidth remote imaging Calgary - TRLabs / Univ. of Calgary: e.g. environmental, geological research done with CLS - CYBERA: super computing facilities Winnipeg - TRLabs / Univ. of Regina: e.g. high bandwidth remote imaging (new digital media development)

  5. CLS Remote User Network Requirements • General characteristics (per CLS): • Secure data transfer • Real-time performance – Guaranteed QoS  Public Internet not a feasible network solution • Users located at research universities, institutes and customer corporate facilities across Canada and internationally  Purpose-built networks too inflexible and costly  Need for a secure, deterministic, flexible and cost-efficient E-Science Grid network

  6. Saskatoon MPLSrouter 1.A MPLSrouter 1.B Assume each site has a pair of mutually protecting MPLS edge routers Requirements for the network grid: 1) Provide maximum possible amount of non-blocking, high-availability, direct light path like inter-site connectivity among the five sites using the single 10Gbps wavelength of TRnet. 2) 10Gbps BW on-demand between any two sites. 3) Network connectivity to be provided as a transparent managed service, and must be administration-free for the users! Rationale: The network must enable transparent, on demand scientific collaboration, instead of requiring the users to spend resources on network administration Edmonton Winnipeg MPLSrouter 5.B MPLSrouter 2.A MPLSrouter 5.B MPLSrouter 2.B Calgary Regina MPLSrouter 4.B MPLSrouter 4.A MPLSrouter 3.B MPLSrouter 3.A

  7. Network Implementation Approaches: • Packet ring (routers/switches directly on the 10Gbps ring): • Hop count between access sites up to 4 • Unrelated traffic streams competing for the same capacity  Traffic that will not even get through on time can block other traffic  Non-deterministic performance, potentially high latencies and packet loss rate, causing low actual efficiency • Hub and spoke: • Requires core L2/3 switches/routers, which need to be configured by a 3rd party service provider, and are not transparent  Not an administration-free, transparent network as required • Requires 10 SONET ADMs in addition to the 2 core L2/L3 switches/routers  in practice also cost prohibitive • Adaptive-Mesh: • 4xOC-48 access per site, totaling 50Gbps, possible with 10 OCS’ ITN nodes • No core routers/switches and no ADMs • Transparent packet forwarding controlled directly by customer’s edge routers

  8. Preferred Network Implementation • Packet ring or hub-and-spoke cannot even in theory provide any more access capacity than Adaptive-Mesh (i.e. 50Gbps on the 10Gbps) ring • The architectural advantages of Adaptive-Mesh, i.e., maximized throughput with deterministic performance and built-in security, will become even more compelling when building larger on-demand e-science grids  Proposal to trial Adaptive-Mesh on TRnet, to dynamically connect TRLabs’ member universities and CLS as a pilot project of the architecture  After such a successful pilot, Adaptive-Mesh should be considered also for wider deployment on research network backbones  These concepts of flexible, cost-efficient, high-performance and secure on-demand grids (and grids of grids) can form the basis of the Future Internet

  9. Adaptive-Mesh Grid over TRnet Implementation Architecture Saskatoon MPLSrouter 1.A MPLSrouter 1.B ITN IF Modules IM “1.A” IM “1.B” ITN Adaptive-Mesh provides direct Layer 1 circuits (of adaptive bandwidth) between each pair of “A” IMs, as well as “B” IMs, with only one instance of packet-forwarding between the MPLS router interfaces See next slide for network ring capacity requirements of an Adaptive-Mesh, e.g. between the “A” IMs of this network  Edmonton Winnipeg MPLSrouter 5.B IM “5.B” IM “2.A” MPLSrouter 2.A MPLSrouter 5.A IM “5.A” IM “2.B” MPLSrouter 2.B Single 10G wavelength ring sufficient for the twenty 2.5G (OC-48c) access points i.e. 50 Gbps of MPLS packet-switched access capacity IM “4.B” IM “3.A” MPLS over OC-48c IM “4.A” IM “3.B” Calgary Regina MPLSrouter 4.B MPLSrouter 4.A MPLSrouter 3.B MPLSrouter 3.A

  10. IM 1 All ITN IF Modules (IMs) able to map packets on all AMBs passing by them; an AMB carries packets to its destination IM from all source IMs along them. Each Adaptive-Concatenation Mux Bus (AMB) capacity MxSTS-1 WDM/SONET cloud West AMB to IM 1 East AMB to IM 1 West AMB to IM 2 East AMB to IM 5 IM 5 IM 2 As many AMBs as there are IMs in the local ring-half, i.e. (N-1)/2 AMBs for N-node full-mesh, needed in each ring direction. In this case, N=5 and thus (5-1)/2 = 2 AMBs needed per ring direction. Thus, protected, non-blocking full-mesh using AMBs among N nodes with MxSTS-1 worth of IF capacity to the network per node, requires [(N-1)/2]xSTS-M of ring capacity. For example, [(5-1)/2]xSTS-48=2xSTS-48  two 5-node 2xSTS-48 rate A-Ms possible over STS-192 ring. East AMB to IM 2 West AMB to IM 5 Each IM interfaces with a customer router, or IM of another Adaptive-Mesh, over a pair of PPP links, each with nominal capacity of MxSTS-1, e.g. OC-48c for M=48. West AMB to IM 3 East AMB to IM 4 IM 3 East AMB to IM 3 West AMB to IM 4 MPLS LER (site 4) IM 4

  11. Adaptive-Mesh Demo The World’s First, Realtime Self-Optimizing, Self-Organizing Network In Action... Geoff Kliza, VP Operations, OCS geoff@optimumzone.net

  12. Double AMB Test Network Configuration • 5 ITN nodes on OC-48 ring • Test requires 2 x OC-12 AMBs: • Nodes # 1, 2, 3 and 4 are sources to Node #5 • The AMB under test carries traffic generated by Port 1 and Port 2 of Agilent N2X tester

  13. Test Setup • Since Adaptive-Mesh architecture of ITN is formed of multiple similar, though independently operating AMBs, only a single AMB needs to be tested to fully characterize ITN Adaptive-Mesh • Agilent N2X MPLS router tester provides priority and bulk traffic of random packet sizes from 64-1500 bytes • Each of the four source-destination traffic streams periodically peaks up to 100% of the OC-12c link capacity • AMB bandwidth allocation among the sources continuously optimized to maintain maximum usage of destination egress capacity • Direct circuit like QoS is maintained even at 100% traffic loads

  14. Adaptive-Mesh for On-Demand E-Science Grids... and the Future Internet • Network to enable, instead of restrict, on-demand scientific collaboration: • Dynamic access and sharing of scientific, computing, storage etc. facilities  The more broadly Adaptive-Mesh is deployed in network backbones, the more practical and cost-efficient worldwide e-Science becomes • A need to dynamically configure Adaptive-Mesh groups among sites with higher mutual network throughput requirements, e.g. based on UCLP • The longer-term vision: Internet like flexibility and reach, with direct light path like performance

  15. We look forward to working with you! Optimum Communications Services, Inc. www.optimumzone.net

  16. Context - Connectivity Requirements for Core Network Content site / Data center SAN All MPLS backbone network access points interconnected as if through a dedicated point-to-point L1/0 connections to the doubled core MPLS switches dedicated for the mobile operator . . . . . . PSTN LER LER LER/ MGW Internet LER/ BGR PSTN . . . Core MPLS switch (A) Core MPLS switch (B) Virtual core routers/switches transparently interconnecting all the LERs MGW Internet LER BGR . . . Wireless access LER Doubled MPLS over OC-Nc or 1/10GbE per site LER Wireless access LER . . . Wireless access LER LER LER LER LER LER Wireless access Wireless access Wireless access MPLS routers at the customer switching centers/content sites/Internet/PSTN access points etc. BGR = Border Gateway Router (with firewalls) LER = Label Edge Router MGW = Media (PSTN<>VoIP) Gateway

  17. Implementation of Connectivity Requirements Using OCS’ITN Adaptive-Mesh Wireless access Wireless access Content site / Data center SAN Wireless access . . . LER LER . . . . . . PSTN LER LER LER . . . IM IM LER/ MGW 7-site 2xSTS-N-rate Adaptive-Mesh over OC-4N ring IM IM IM IM IM Internet Doubled 5-site STS-N-rate Adaptive-Mesh over OC-4N ring; interconnects the four regional Adaptive-Mesh networks and the PSTN/Internet access points See next slide  LER/ BGR IM IM IM IM PSTN . . . MGW IM IM IM Internet IM 1+1 Protected, Link-aggregated MPLS over STS-Nc PPP links LER LER BGR IM IM IM 7-site 2xSTS-N inter-site rate Adaptive-Mesh over OC-4N ring . . . IM IM 7-site 2xSTS-N inter-site rate Adaptive-Mesh over OC-4N ring Wireless access IM . . . LER LER IM IM Doubled MPLS over OC-Nc or 1/10GbE per site IM IM IM IM LER LER IM IM Wireless access LER . . . Wireless access LER LER LER LER LER LER Wireless access Wireless access Wireless access MPLS routers at the customer switching centers/content sites/Internet/PSTN access points etc. BGR = Border Gateway Router (with firewalls) LER = Label Edge Router MGW = Media (PSTN<>VoIP) Gateway IM = OCS’ Intelligent Transport Network™ IF module

  18. Adaptive-Mesh for On-Demand E-Science Grids Summary: • Customer controllable, packet-layer transparent network service • Packet transport using circuits of dynamically optimized bandwidth • Maximized data throughput based on realtime traffic load patterns Adaptive-Mesh Network Service: User groups’ private multi-service backbone -- just without the need for the users to spend capital or resources on deploying or operating their backbonenetwork

More Related