Patent Application
20110194854
The invention described herein applies to the area of tele-communications, specifically to the Passive Optical Networks (PONS) domain. Preferred embodiments of the invention apply to both dominant technologies of the aforementioned area:
1 Gbit/s Ethernet PON (EPON), as specified in the IEEE 802.3ah (and to the next-generation 10 Gbit/s EPON, to be standardized in IEEE 802.3av).
Gigabit capable PONS (GPONs), as specified in the G.984 ITU recommendation, operating at 1.25 and 2.5 Gbit/s (and probably to next-generation higher rate GPON implementations).
It should be understood, however, that the present invention is applicable to all packet-based networks wherein a medium is shared between a plurality of subscribers or network units in a time division multiplex scheme.
The invention aims at improving the ratio between useful user data (usually termed “payload”) and the overall data transmitted over a shared communications medium. Ideally, all data transmitted over the medium would be payload, however, data required for the management of the data transmission such as source and destination addresses, data type, data length, and other protocol-related data have to be transmitted alongside the payload. This additional data is also referred to as “overhead”. Since channel bandwidth is fixed, it is generally desirable to maximise the ratio between the payload and the overall data (payload plus overhead) which implies that the overhead should be reduced as much as possible.
As indicated above, the invention is preferably applied to PONS wherein a plurality of Optical Network Units (ONUs) located at the subscribers' premises communicate with a single Optical Line Terminal (OLT) located at a Central Office (CO). Such PONS are generally deployed using a passive tree-and-branch or bus topology which implies that the physical medium, the optical fibre, is shared by the ONUs. In the down-stream direction, the OLT broadcasts the Ethernet frames (also called “packets” herein) in a point-to-multipoint scheme. The Ethernet frames are extracted by the ONUs identified by a Logical Link IDentifier (LLID) transmitted within the frame which then forward the frames to the end-user identified by the frame destination Medium Access Control (MAC) address assigned to the end-user's network terminal. In the upstream direction, the ONUs must contend for access (in a multipoint-to-point regime) which usually is managed by Dynamic Bandwidth Allocation (DBA) algorithms dynamically assigning Time-Slots (TS) to the ONUs depending on the current network load originating from the ONUs.
Although less scalable and with no flexibility, a Fixed Slot Allocation (FSA) scheme can also be applied to distribute the upstream access, forcing each ONU to wait for its assigned time-slot to then transmit the frames that arrived from the end-users connected to it in the meantime. The time-slots are grouped in cycles so that every ONU has the possibility to transmit packets with reduced waiting time.
To enable the bandwidth management control of the upstream channel, the Multi-Point Control Protocol (MPCP) was specified in the EPON standard. This protocol has several important tasks, such as ONU synchronization and monitoring, and DBA information exchange. To grant an upstream TS, the OLT should be aware of the current ONU load. This task is achieved by a reporting mechanism where each ONU informs the OLT by sending a MPCP Report message indicating the current state of its buffers. Upon receiving the Report, the OLT grants a TS for the ONU by sending a MPCP Gate message, which includes the TS start time and duration. Using this scheduling scheme, each ONU is polled by the OLT to be allowed to transmit in the upstream channel. Since the Report/Gate message exchange can take some useful transmission time, an interleaved polling method is usually applied.
Once an upstream TS is granted, a virtual point-to-point connection is established between OLT and a given ONU. This means that during the TS duration, all the frames arriving at the OLT are transmitted by the same ONU. Since each ONU is only awarded a TS once in each cycle, there is a waiting period between every two cycles, logically and superiorly limited by the cycle size. This burst-mode transmission implies that the frames must be kept in the ONU buffers until the next TS start time is reached to start the upstream transmission.
The most prominent standards in the field of Passive Optical Networks are EPON and GPON:
The data transported by the EPON is encapsulated in 1 Gbit/s Ethernet standard frames, as depicted in
The total amount of Ethernet frame headers transmitted in the EPON introduces inefficiency to the system. Currently, since the Ethernet frame payload (where the client data is transmitted) is inferiorly and superiorly limited to a size of 46 and 1500 bytes, respectively, this inefficiency can typically achieve 9% of the channel capacity. The total amount of user data that the network is able to support plays a significant influence in the number of customers (ONUs) that can be supported. Thus, the more the inefficiency is reduced, the economically more attractive a network becomes because more ONUs and thus more customers can be provided network services via a single PON.
Unlike EPONs, GPONs do not transmit standard Ethernet frames over the optical medium. The data coming from the non-optical interface ports (usually Ethernet frames) is encapsulated in GPON Encapsulation Method (GEM) frames before being sent to the passive part of the network. The Ethernet frames are stripped from the preamble and sent without Inter Frame Gaps (IFGs). This procedure lowers some of the overhead caused by Ethernet headers, since the aforementioned encapsulation substitutes a total of 20 bytes (12 bytes+8 bytes) by a GEM header of 5 bytes (see
All the abovementioned EPON DBA mechanisms encounter their equivalent in any GPON implementation: the reports are sent in the upstream channel with the difference of being sent before each burst transmission (in the Dynamic Bandwidth Report upstream (DBRu) messages) and GATEs are included in the headers of each downstream p-frame. Logical identifiers are also found on such networks, serving the exact same purposes they serve in EPONs. The burst-and-wait feature is also present in GPONs.
ONUs are typically termed Optical Network Terminals (ONTs) in the GPON slang, nevertheless, in the remaining part of this document, as the invention applies to both types of systems, the term ONU will be used to refer to the two optical devices indifferently, unless something is said to the contrary.
The invention includes the insight that the wait-and-burst feature typical in PONS can be exploited in order to heavily reduce the inefficiency affecting the EPON upstream channel. It also relies on the fact that the header information of a plurality of the received packets is repeated. The efficiency of an optional payload compression is enhanced by a certain degree of similarity within the contents of the data units that can be expected due to correlation of a sequence of packets from the same communication stream.
The invention will now be explained referring to a set of figures wherein
One important idea of the invention is to improve the efficiency of the upstream channel of a PON by reducing the overhead imposed by the Ethernet headers while taking advantage of the fact that the frames that arrive at a given ONU are being stored (and delayed) in queues waiting for their TS to come. According to the invention, instead of storing the complete frames, all the payloads belonging to a given communication flow (same source and destination MAC addresses, and same LLID) are gathered and a larger frame is constructed which is transmitted with a single MAC header and a global Frame Check Sequence (FCS). The OLT will then receive these aggregated frames (composite frames) and de-multiplex them into the common Ethernet frames before forwarding them.
Another important aspect of the invention is the frame format that will carry the frames of a given communication flow identified by the pair of source/destination addresses. The proposed frame format, embodiments of which suitable for the EPON and GPON standards are depicted in
As can be seen in
A second aspect of the invention concerns aggregation and demultiplexing procedures. The aggregation procedure is usually carried out by the ONUs, and the de-multiplexing procedure by the OLT. Of course, the aggregation procedure can be advantageously applied to other network elements gathering data packets while waiting for a TS during which the collected data packets will be forwarded. In the aggregation process the device identifies two or more frames belonging to the same flow and constructs (or adds frames to) the composite frame by isolating the first Ethernet frame header and by stripping all the following from their FCS and addressing fields (see
If no second incoming frame is detected for the same specific source/destination pair until the arrival of the TS, the unchanged single frame can be sent instead of a composite frame only comprising the single aggregated frame. Since in this case, both standard frames and composite frames are sent over the same communications link, it is useful to provide a type indicator which declares which type of frame a received frame belongs to. For compatibility reasons it is preferable to not mark the standard frames and only add a type indicator to the composite frames. Entries in the frame's preamble which are unused in standard frames could be used or a “magic”, i.e. a specific predetermined number which is very unlikely to occur in a normal data stream, could be prepended (prefixed) to the aggregated blocks. In this manner it also becomes possible to selectively apply aggregation/de-multiplexing of frames to frames having a relatively low priority. This is advantageous because in this way high-priority data such as data belonging to real-time services (e.g. VoIP) will not be delayed by the additional processing that would be required for aggregation and de-multiplexing.
If a size limit for the composite frame exists, the procedure also checks whether the addition of the newly received block surpasses the maximum value (Step 130). If this is not the case, the block consisting of the new Ethernet frame payload plus type/length and block size field can be added to the composite frame (Steps 171 and 172), provided that a check for sufficient space in the TS is positive (Step 160); otherwise the filling of the TS finishes in Step 190. When the frame is added to the composite frame, the global FCS value, if present, should be also updated and the Total Length field incremented (Step 173).
If adding the frame to the composite frame surpasses the maximum value in Step 130, the FCS field is attached to the composite frame (Step 140) and a new composite frame is created to which the new frame may be added (Step 150). If a FCS is present, it is now appended to the end of the composite frame.
After the aggregation of the frame it is checked whether the TS is completely full (Step 180). If so, the procedure terminates in Step 190. If not, the procedure continues by branching back to Step 110 where a new frame is selected from a waiting queue.
An embodiment of this algorithm is schematised in
It must be kept in mind that, in the case of EPONs, the logical addressing scheme (supported by the LLIDs) should be enforced by assuring the aggregation is made taking them into account also (especially in Step 120). In other words, a given composite frame should only carry individual frames that, additionally to the fact of sharing the same source/destination MAC addresses, also share the same origin LLID. In EPON implementations where an LLID is assigned exclusively to each ONU, the aggregation will be made on a per ONU basis, whereas in implementations where an LLID is assigned to individual queues or users, the aggregation will be made on a per queue or per user basis.
Once the composite frame is received at the OLT side, a demultiplexing algorithm has to be applied in order to re-obtain the individual system compliant frames. This procedure is basically the inverse of the previously described one: first of all, the FCS of the composite frame has to be checked out (if present) in Step 210; if valid, the MAC destination address (DA) and source address (SA) can be read and stored in memory and the frame can be further processed (Step 220). The frame is discarded, else, (Step 215) and the TS finished (Step 310). The procedure gets all the blocks in the composite frame payload and appends to them the IFG plus Preamble (in the case of the EPON) or the GEM header in the case of the GPON and calculate the FCS for each newly reconstructed frame and adds it to the end of the data unit. A flow chart of a preferred embodiment of the de-multiplexing algorithm is shown in
Having still in mind the idle time experienced by the ONUs while waiting for the TS, and aiming for the maximum data optimisation possible, a further embodiment of the invention includes application of a compression algorithm at the ONU and of a decompression algorithm in the OLT. Such mechanism should be applied to a composite frame after the aggregation was done taking advantage of the increased size of the obtained frame to achieve a higher compression ratio. As is known to a skilled person in the art, there are several high performance compression/decompression algorithms available which operate in an iterative manner, ideal for a fast network implementation. Since there is a certain probability that a sequence of frames directed to the same DA and originating from the same SA belong to a single communication, there is an increased likelihood that the data comprised in the individual frames share certain characteristics which can be exploited by compression algorithms. Thus, application of a compression algorithm to the composite frame is likely to yield a higher compression ratio than application of the same compression algorithm to each frame individually or to all data arriving at an ONU between two TS as a whole if the order of arrival is maintained unchanged. In a preferred embodiment of the invention the “size of block”- and “Length Type”-indicators are stored in sequence within the composite frame such that the remaining part of the composite frame will only comprise contiguous blocks of payload data extracted from a sequence of packets having the same source/destination addresses. In this way the compression ratio can be even more increased because packet sizes and length type-indicators have a high probability of being the same for a single communication stream. Furthermore, a large portion of the composite frame payload comprises only payloads collected from the stream of packets which usually leads to better compression ratios because there is only data of the same type in the large portion of the composite frame payload.
Simulations of a realistic network scenario indicate that the communication overhead in an EPON can be reduced from an average of 9% to only 1%. In other words, the effective maximum transmission bandwidth was raised from 91% to 99% of the channel capacity. Thus, the invention enables an EPON previously providing data services to 50 ONUs to now accommodate 54 ONUs. If compression is applied, the total improvement can reach up to 25%.
The invention provides the following advantages:
TS Time-Slot
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/EP08/60020 | 7/30/2008 | WO | 00 | 4/25/2011 |
