Dynamic Provisioning via a Stateful PCE with ... - IEEE Xplore

2 downloads 0 Views 356KB Size Report
(3) KDDI R&D Laboratories Inc., 2-1-15 Ohara Fujimino-shi, Saitama, Japan. Abstract We report the implementation and performance evaluation of an active ...
Dynamic Provisioning via a Stateful PCE with Instantiation Capabilities in GMPLS-Controlled Flexi-grid DWDM Networks R. Casellas(1), R. Martínez(1), R. Muñoz(1), L. Liu(2), T. Tsuritani(3), I. Morita(3) (1)

CTTC, Av. Carl Friedrich Gauss n7, Castelldefels, Barcelona, Spain, [email protected] (2) University of California, Davis. One Shields Ave, Davis, CA 95616, USA. (3) KDDI R&D Laboratories Inc., 2-1-15 Ohara Fujimino-shi, Saitama, Japan.

Abstract We report the implementation and performance evaluation of an active stateful PCE that relies on a GMPLS control plane for the actual provisioning of elastic connections in a flexi-grid DWDM network. It is based on experimental extensions to the PCEP protocol and enables more advanced and concurrent path computations. Introduction and Motivation A stateful PCE1-3 is a PCE that is able to consider, for the purposes of path computation, not only the network status in terms of links and nodes (referred to as the Traffic Engineering Database or TED) but also the status of active connections (computed paths and reserved resources known as the database of Label Switched Paths or LSPDB). Although the IETF is still considering the taxonomy of stateful PCEs, a stateful PCE is commonly said to be active if it is able to recommend or modify/affect the state of existing connections. Moreover, an active, stateful PCE may also have instantiation capabilities4, i.e., the ability to trigger, upon request or autonomously, the establishment and release of connections. Active stateful PCEs are attracting a lot of attention, in view of their potential role and new opportunities tied to the challenges within the socalled Software Defined Networks (SDN), as well as given their associated performance and operational benefits. In the particular scope of flexi-grid networks, the internal book-keeping and knowledge of the path and allocated frequency slot on a per connection basis enables the deployment of more efficient algorithms for provisioning, adaptive network planning and global concurrent optimization re-routing and defragmentation, since a major advantage of an active PCE is the ability to re-route an existing connection to route a new one. A lab deployment and the experimental evaluation of a stateful PCE for flexi-grid DWDM networks assuming an OpenFlow controller to perform the connection provisioning has been recently reported5. Given the nature of the PCE architecture and the real decoupling of the path computation function, an active PCE can be equally deployed relying on an underlying GMPLS control plane, triggering signaling events by means of PCEP extensions. In this later case, the stateful PCE is able to delegate the actual establishment of lightpaths to the underlying GMPLS control plane, rather than

programming the forwarding behavior (i.e., establishing the optical cross-connects) directly through OpenFlow protocol5. The combination of an active PCE with a GMPLS control plane in clearly justified: the GMPLS architecture relies on a set of mature, well-tested protocols for link management, topology dissemination and signaling, has been extended to cover WSON and will address flexigrid extensions6. In this setting, a node can delegate the control of an LSP to the PCE, but state is also present at the corresponding head end nodes (GMPLS controllers). In case of a PCE failure, LSPs may be re-synchronized between both using a mechanism that is based on PCRpt messages2. Architecture and Control Plane procedures The implemented architecture is based on ongoing work at the IETF1-3. The two main functional entities are the (assumed unique) active stateful PCE and a Path Computation Client (PCC) component located at each endpoint that can instantiate LSPs (GMPLS controller).

Fig. 1: Deployed stateful PCE architecture over a GMPLS control plane for flexi-grid optical networks

The PCC component of the controller has an interface with the connection controller (CC) and a persistent PCEP session with the stateful PCE. Upon connection, the PCC announces the node endpoint it represents, allowing the PCE to

identify the appropriate connection for a given source/destination request. The procedure is as follows (cfr. Fig. 2): the implemented stateful PCE features a northbound interface that enables connectivity service requests (e.g., from a Network Management System or NMS). This interface is based on PCEP with extensions to indicate whether an instantiation is requested. This choice is based on the fact that PCEP allows indicating the endpoints in a request, heterogeneous traffic parameters, allows synchronized computation requests and was extended for routing and spectrum / frequency slot assignment (RSA)7. After the path and frequency slots have been computed, the PCE sends a PCEP Create (PCCreate) message to the node head-end PCC component containing the endpoints and the path. Upon reception of a PCCreate message, the PCC parses each entry in the instantiation request list and triggers the establishment of the LSP via the connection controller (CC). For this, it examines the PCEP ENDPOINTS, route (ERO), traffic parameters (BANDWIDTH) and attributes (LSPA) objects and allocates an identifier, local to the scope of the PCEP session, bound to the unique symbolic name found in the instantiation request.A state report PCEP message (PCRpt) is then sent to the PCE. If the setup succeeds the final path attributes are reported, including the binding local id-symbolic name as well as LSP identifiers allocated with the SESSION and SENDER TEMPLATE objects.

request list and proceeds accordingly. A particular case is the release of the connection. Implementation considerations The interface between the PCC and the CC is encoded using XML, allowing flexibility when requesting the establishment, re-route and release of LSPs. At the PCE, the LSPDB is implemented as a container maintaining multiple indices (multiply indexed relational table). Upon request from the northbound interface, a temporary entry is allocated in the LSPDB, indexed by the unique symbolic-name, along with its instantiation timestamp. The entry is updated upon reception of the PCRpt message, completing its PCC assigned local identifier, record route, and effective frequency slot. A created timestamp is used to compute the setup delay as seen by the PCE. For robustness, the PCE pro-actively updates its TED modifying the status of the nominal central frequencies upon reception of PCRpt messages, avoiding assigning the same slot to almost concurrent requests. Nonetheless the actual state is also updated upon parsing the OSPF-TE LSAs (also extended to include the status of the central frequencies), since there may be other changes not bound to an instantiation. Access to the PCE TED is synchronized via mutual exclusion. Performance Evaluation The system is evaluated in a 14-node testbed as shown in Fig.3. DWDM links have 128 nominal central frequencies (slices). The stateful PCE is able to perform Routing and Spectrum Assignment (RSA) using a distance-adaptive, iterative, two-phase approach, combining off-line path characterization with dynamic spatial path computation assuming CO-OFDM transmission stored in pre-cached tables, depending on requested bitrate, and the computed path hop count and distance7, requiring for the considered client data rates 1, 3, 6, 7 or 14 slices of 6.25 GHz. A sample message flow can be seen in Fig.4 and detailed in Fig. 5.

Fig. 2: PCE based Instantiation and release procedures for unidirectional connection

In case of a provisioning failure, an error is included (RSVP ERROR_SPEC TLV), attaching the symbolic name, local identifier and error code/value, an LSP object is also attached indicating that the LSP is no longer operational and that it has been removed. When a PCE wishes to modify the state of an LSP, it sends a PCEP update message (PCUpd) to the PCC/head end node. The PCC parses the update

Fig. 3: 14-node Japanese topology for the performance evaluation of the stateful PCE system.

Fig. 4: Wireshark capture of the PCEP and RSVP-TE message sequence at head end node 10.1.1.140 with router id 10.0.50.3. Setup delay corresponds to 13ms for a connection duration of 9.3s.

We evaluate the blocking probability (BP) in function of the offered traffic load considering a dynamic stochastic model of requests: the arrival process is Poisson with avg. inter-arrival set to 1s; the holding time follows a negative exponential distribution. Requests are randomly selected between distinct node pairs, with a random bandwidth profile mapped to CO-OFDM transmission.

count, both tests are stable around 2.4 and 2.5 hops, with a local avg. hop count maximum at 60 and 80 Er. respectively. At a higher offered traffic successful connections tend to use shorter paths. Avg. setup delay is ~8ms, with min and max ranging, typically, from ~3 ms for a single-hop connection to ~20 ms max values (cfr. Fig.7 for T2 at 20 Er. with 20k requests).

Fig. 5: Wireshark capture PCCreate msg (left) and PCRpt - PCUpd (right).

Conclusions We have demonstrated the feasibility and applicability of a stateful PCE, which relies on an extended GMPLS control plane for the dynamic control of flexi-grid networks. Quantitative performance indicators have been obtained, such as the BP or the setup delay. The combination of stateful and active capabilities renders the PCE a very promising functional element in SDN and GMPLS based networks.

Fig. 7: Setup delay histogram CDF for T2 at 20 Er.

In the first test (T1), client data rates are uniformly random selected [1-100] Gbps, mapped to 10, 40 or 100 Gbps. In the second test (T2) client data rates are fixed to 100 Gbps. 1..100Gbps

100Gbps

100

10

Acknowledgements

1 20

40

60

80

100

120

140

160

0,1

0,01

Work was partially funded by EC’s FP7/2007-2013 IP IDEALIST project (317999), and the Spanish MINECO through the project FARO (TEC2012-38119).

References [1] X. Zhang, draft-zhang-pce-stateful-pce-app, 2013.

0,001

Fig. 6: Obtained Blocking Probability for T1 and T2 (BP in %, offered traffic x in Er)

Fig. 6 shows the obtained blocking probability in both cases, showing a higher value for the highest profile. Regarding average route hop

[2] E. Crabbe, draft-ietf-pce-stateful-pce, 2013. [3] X.Zhang, draft-zhang-pce-pcep-stateful-gmpls,ID [4] E. Crabbe, draft-crabbe-pce-initiated-lsp, 2012. [5] R. Casellas, OW4G.2, OFC/NFOEC 2013. [6] O. González, R. Casellas, draft-ogrcetal-ccamp-flexi-gridfwk-02, 2013. [7] R. Casellas. JOCN. V4, N10, B1-B10, Nov. 2012