1 / 11

Control Mechanisms for the clustered OBS

Control Mechanisms for the clustered OBS. Motivation. Losses due to contention plague OBS performance*. Losses exponential with contention points. Objective of clustering is to avoid contention within few first and last nodes.

irina
Download Presentation

Control Mechanisms for the clustered OBS

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. Control Mechanisms for the clustered OBS

  2. Motivation • Losses due to contention plague OBS performance*. Losses exponential with contention points. Objective of clustering is to avoid contention within few first and last nodes. • Exploit ring topology and reservation-based access to nullify burst loss inside a cluster where round trip delay not prohibitive • Use either static burst schedules or “classic” OBS between clusters (still only few remaining contention points and hence loss probability much reduced) • Fixed size slots alleviate heavy control needed for lossless 2-way slot reservations and then inter-cluster bursts are trains of slots creating an also fixed-size frame for slower, cheaper switching among clusters

  3. MN OBS Header λ0 λi OBS Header λ0 λj MN MN λ0 OBS Header λy λ0 Header Slot λi λj MN Concept of operation

  4. Burst creation as trains of slots ·Using the control wave length λ0, nodes place in the Slot Reservation Map their queue length towards each other cluster implying a request to send to that cluster. ·  The MN collects all requests and on the basis of total waiting traffic it issues the destination cluster of the next burst frame and an Slot Allocation Map which directs nodes where to place their payload slots (which λ and which slots in the frame). · The nodes also add to the Slot Destination Map the specific Dest. Node inside the Dest. Cluster to be used at the receiving cluster side.

  5. Slot Reservation Map … Cluster C Cluster 0 Cluster1 … DestCluster Nodeaddr NodeAddr SD EC EC QL QL SD=Start Delimiter Format of slot reservation

  6. Node 1 Node 0 Node i Total Node N Cluster 0 Cluster 1 Cluster j Cluster C Slot Reservation Matrix Reservations express queue length in slots minus current allocations

  7. Frame Allocation Algorithm • · The reservation total towards a cluster is used to decide next destination cluster • ·   This occurs when size can fill a frame or timer expired • ·  Individual allocations are per request after checking for possible SLA violations. • If not enough slots for all nodes allocation “pro rata” according to SLA calculated weights • Fairness checks and policing are possible • · Immediate access mode also possible

  8. 100000 bits SAP SAM SAM=Slot Allocation Map SAP=Slot Alignment Pattern DCA=Destination Cluster Address SA=Source (node) Address DA=Destination (node) Address … λ1 λL λ2 8bits 12bits 12bits 8bits 8bits … DCA EndSlot EndSlot EndSlot StartSlot StartSlot StartSlot EC EC EC SA DA SA DA SA DA W Y 2 5 7 Z N 254 255 1 X 0 4 Slot 255 Slot 0 Slot 254 Slot 4 Slot 7 Slot 5 Time lead ofcontrol header … Node1 Node 2 NodeN λ1 … Payload Frames Node 3 Node7 Node K λ2 … … Node 5 Node 11 Node L λL Format of Control Header slot

  9. Inter-cluster burst exchanges • Bursts are sent between clusters only via MN nodes • The burst header once it reaches the MN has information relating to OBS routing added and slot allocations (only relevant to intra-cluster communication) removed • The destination node map is kept in the OBS header through the travel and it is then inserted into the destination cluster control channel to allow nodes to identify slots addressed to them (all three headers precede actual payload by enough slot times to allow processing) • Resulting burst can also be fixed size for cheaper switching (efficiency penalty) or variable but still integer multiple of slots whenever cheaper faster switching available (better fill-levels)

  10. Version B: Variable burst size • The choice of fixed-size bursts is made only to allow slower cheaper switching technology to be used among MNs • Otherwise, the described control system can easily support variable size bursts as trains of integer number of slots. • The burst allocation logic is modified to: burst allocated to cluster that has reached max burst size or timer expired (in which case size decided by reserved slot number or min size, whichever larger). • Obviously variable bursts will improve efficiency because the empty part of burst is cut off also improving chances of burst not colliding at MN switching.

  11. Conclusions • By exploiting 2-way MAC-like slot reservations within a cluster, burst loss is avoided in the first and last hops where round trip delay not prohibitive. “Classic” OBS used between clusters where distances prohibit demand driven reservation due to intolerant delays. • The burst control header is used in all three steps ahead of the payload (as always in OBS) albeit with slightly modified functionality. • However the key to this scheme is the reservation slot which allows the system to respond to actual fluctuating demand. • And loss reduction due to fewer contention points. • Also traffic aggregation among several clustered sources and destinations improves burstification and hence efficiency • Bottom line: Good trade-off between loss-burdened OBS, low-fill levels of lightpath reservations and inefficiency of static SDH • Key: responsive to bursty traffic at required timescales*

More Related