Patent Application
20060053424
The present invention relates to an automated load balancing and substantially streamlined resource management in a communication system. The present invention is particularly applicable when high-speed packet based links are interfaced with signal processing resources.
Typically, in present communication systems a resource management functionality plays a significant role in any network element providing heterogeneous signal processing services. This means that several processing units must be dedicated to provide the resource management functionality. Also, quite a high amount of the internal communication bandwidth must be reserved for exchanging the resource management related control messages. However, increasing the number of the media processing units may cause the resource manager to become a bottleneck that may mean a reduced overall cost-efficiency, and, in practice, longer latency times when responding to new service requests.
However, the prior art described above suffers from the following drawbacks. This conventional resource management system is not scaleable which means that there is a large number of dedicated units as well as a large number of control messages. In addition, delays in responding to service requests occur. Moreover, the utilization of statistical multiplexing inflicts difficulties.
Particularly, the processing times in the packet switched connection according to the prior art are not deterministic. Specifically, in the conventional arrangement, one digital signal processor (DSP) is receiving and processing several channels (e.g. 8 . . . 16) simultaneously. This means that a packet must wait a non-deterministic time before it is processed, that is, depending on how many channels are processed before that specific channel.
Accordingly, it is an object of the present invention to overcome these shortcomings of the prior art and to provide a streamlined and cost-effective way to manage the available resource pool.
According to the present invention, the object is solved by providing a method of balancing the load of resources in a packet switched connection within a communication system, said system comprising processing units for performing communication, at least one load balancing unit for distributing the load to said processing units, and a data storage, said method comprising the steps of: obtaining a current connection state as well as a current load state of said processing units from said data storage; selecting by said load balancing unit a processing unit on a per-packet basis; and maintaining information about the load state of each processing unit so that said selecting step is performed by selecting a processing unit to serve and process a respective packet based on the load state.
Here, the data storage can be accessed to by said load balancing unit or said processing units. Further, the information about the load state may be maintained as a Boolean state, i.e. to indicate free or not free.
The selection of a processing unit can be done in a round-robin fashion.
Further, a supported service profile for each processing unit can be maintained in addition. In this case, the supported service profile can be used as additional selection criteria.
In the method according to the present invention, the load balancing unit can obtain a load state from each processing unit upon a hardware based mechanism or a packet based mechanism. In the latter case, a load state of a processing unit may be inserted into a packet processed by said unit or a packet returned by a processing unit may be interpreted as a flag for a free resource.
Besides, should excess traffic occur it can be redirected to another load balancing unit, wherein said excess traffic would be defined upon the number of active processing units.
The method according to the present invention provides a more effective utilization of the media processing resources, since the resources are managed on the basis of an effective resource allocation of the whole network element instead of managing resource allocations of single processing units. Thus, the benefits of a statistical multiplexing can be exploited easily.
An additional benefit is the deterministic, i.e. optimal, processing time that a single packet always encounters, since a processing unit serves only one packet at a time.
Moreover, with the method according to the present invention, the processing delay of a received packet is always optimal and very constant. Consequently, a constant processing time minimizes unwanted jitter and other possible fluctuations of the traffic flow.
According to the present invention, the object is further solved by providing a device unit for serving and processing packets of a communication connection, comprising means adapted to inform a load state of said device to a balancing unit; and means adapted to obtain a state of said communication connection.
In this processing device unit, said obtaining means can be adapted to retrieve said communication connection state from a data storage or from a packet being under processing.
According to the present invention, the object is still further solved by providing a device unit for balancing a load of each of multiple processing units-performing a packet switched communication connection, comprising: means for maintaining a load state of each of said processing units; and means adapted to select a processing unit on the basis of a respective load state.
In this balancing device unit, a load state of a processing unit may be contained in a table. The state can be expressed as a Boolean state or as value which corresponds to the percentage of load.
Further, said selecting means can be adapted such that a processing unit is selected also on the basis of a parameter indicating the service profile supported by a respective processing unit. In this case, said parameter could be contained in a table.
As a modification, the load balancing device unit may further comprise means adapted to insert a communication connection state into a packet to be routed.
In a preferred embodiment, the processing units are comprised of multicore digital signal processing means having a shared data storage for all cores, whereby said device comprises a first level of load balancing for selecting a digital signal processing means and a second level of load balancing for selecting a single core. As another modification, the load balancing device unit may further comprise means for redirecting excess traffic to another load balancing device unit according to the present invention, wherein said excess traffic is defined upon the number of active processing units.
Furthermore, a system adapted to perform the method according to the present invention and/or comprising one or more devices according to the present invention does also solve the object.
Further details and advantages of the present invention as well as further modifications thereof are apparent from the detailed description of the preferred embodiments which are to be taken in conjunction with the appended drawings, in which:
The present invention introduces a load balancing unit (or DSP selector) in front of the DSP resource pool. The main idea behind the load balancing unit is to remove (or at least substantially reduce) the need for separate and poorly scaleable resource management layers.
The
Specifically, in
According to the arrangement depicted in
Next, in
According to the arrangement depicted in
What is common to both implementations is that the load balancing unit keeps track of the total utilization of the processing units and this overall load information can be provided for other network management processes. Specifically, the arrangement that is depicted in
Furthermore, the conveyance of the load state from each processing unit to the load balancing unit may happen either by a hardware based mechanism (such as dedicated pin, shared memory etc.) or a packet based mechanism (such as inserting the load state to returning (processed) packets or just interpreting a returning packet as a flag for a free resource). The load balancing unit and the processing units may be interconnected for example with Ethernet/IP, thus they do not require a physical co-location.
Some functionalities that the processing units, the load balancing unit and the data storage provide in preferred embodiments of the invention are outlined in the following.
The processing unit implements a mechanism to inform the load status to the load balancing unit which can be a hardware based mechanism (dedicated pin, shared memory etc.) or a packet based, e.g. inserting the status in the processed packets. Further, the processing unit comprises means for obtaining the connection state from the data storage or from the received packet.
The load balancing unit which can also be one of the processing units implements a table that contains the load status of each DSP unit in a Boolean format (free or not free) or as a percentage of load (0 . . . 100% load). Further, it comprises means for selecting a resource based on the load status, wherein a parameter that indicates the supported service profile for each processing unit (e.g. only EFR codec) may also be used as an additional selection variable. Optionally, the load balancing unit may also comprise means for inserting the connection-state to a routed packet.
The data storage has to maintain the connection states by mapping them to suitable connection identifiers such as UDP ports and may lock the states in order to handle bursts of packets.
While the above may be considered as a basic arrangement according to the present invention, further developments of the same invention may be as follows.
Since the state-of-the art signal processors consist of multiple cores (e.g. 4-8) per one physical chip, one possibility could be to implement the load balancing unit functionality inside each multi-core DSP device which usually have a shared memory (data storage) for all cores. This way, there would be two levels of load balancing: one for selecting the DSP and a second level for selecting a single core.
In addition, it would also be preferred to have a redirecting functionality. That is, the excess traffic of a first load balancing unit could be redirected to another load balancing unit if a certain limit (or load) is exceeded, i.e. when all or most of the processing units are active when a new packet arrives.
According to the above, the benefits of a streamlined resource management, an automated load balancing between a high number of processing units, a more efficient utilization of the DSP resources resulting in a statistical multiplexing with a high number of processing units managed as a whole, the guaranteeing of a deterministic processing delay for each packet leading to a minimum delay and a smooth traffic pattern, and maintaining the possibility to still be able to dedicate processing units to a specific service as an optimal utilization of memory thus gaining a highest number of channels can be achieved.
According to the above, it is provided a method of balancing the load of resources in a packet switched connection within a communication system, said system comprising processing units 11; 21 for performing communication, at least one load balancing unit 12; 22 for distributing the load to said processing units 11; 21, and a data storage 14; 24, said method comprising the steps of: obtaining a current connection state as well as a current load state of said processing units from said data storage 14; 24; selecting by said load balancing unit 12; 22 a processing unit on a per-packet basis; and maintaining information about the load state of each processing unit 11; 21 so that said selecting step is performed by selecting a processing unit to serve and process a respective packet based on the load state.
While it is described above what is presently considered to be the preferred embodiments of the present invention, it is apparent to those skilled in the art that various modifications are possible to the present invention without departing from the spirit and scope thereof which is defined in the appended claims.
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/IB02/02492 | 6/28/2002 | WO | 1/13/2005 |
