GPRS Network Infrastructure Dimensioning and Performance

Transcription

1 GPRS Network Infrastructure Dimensioning and Performance Jim Donahue, BTCellnet, GPRS Network Programme Manager Peter Lisle, GPRS Programme Manager, BTCellnet, BT Cellnet is one of the first operators in the world to deploy General Packet Radio Services (GPRS) in its live GSM network. This paper will explain the key GPRS network infrastructure elements and review the key drivers for network dimensioning and network performance based upon experience to date. The following topics will be covered: 1. GPRS Network Infrastructure Overview 2. Dimensioning GPRS Elements Radio network dimensioning Core domain dimensioning 3. GPRS Network Performance Peak throughput drivers Latency drivers 1. GPRS Network Infrastructure Overview GPRS has been designed as an extension to the existing GSM network infrastructure to provide a connectionless packet data service. GPRS introduces a number of new functional elements that support the end to end transport of IP based packet data. GPRS has been developed by the GSM standards bodies, resulting in a system with defined functionality, interfaces and inter-network operation for roaming support. Two major new core network elements are introduced: the Serving GPRS Support Node (SGSN) and the Gateway GPRS Support node (GGSN). The SGSN monitors the state of the mobile station and tracks its movements within a given geographical area. It is also responsible for establishing and managing the data connections between the mobile user and the destination network. The GGSN provides the point of attachment between the GPRS domain and external data networks such as the Internet and Corporate Intranets. Each external network is given a unique Access Point Name (APN) which is used by the mobile user to establish the connection to the required destination network. The GSM Base Station Subsystem (BSS) has been adapted to support the GPRS connectionless packet mode of operation. A new functional node called the Packet Control Unit has been introduced (as part of the BSS) to control and manage the allocation of GPRS radio resources to the mobile users. The modifications to the radio infrastructure and additional functionality introduced by GPRS means that new mobile stations (MS) are required.

2 The GPRS network infrastructure is summarised Figure 1. VLR MSC PSTN/ ISDN MSC - Mobile Switching Centre VLR - Visitor Location Register HLR - Home Location Register BSS - Base Station System PSPDN - Packet Switched Public Data Network SGSN - Serving GPRS Support Node GGSN - Gateway GPRS Support Node A Gs HLR Gr Gc BSS Gb SGSN Gn GGSN Gi PSPDN N.B. Gc & Gs interfaces are optional Figure 1: Overview of GPRS Network Infrastructure As can be seen, there are a number of new standardised network interfaces introduced: Gb - Frame relay connection between the SGSN and the PCU within the BSS. This transports both user data and signalling messages to/from the SGSN. Gn - The GPRS backbone network, implemented using IP LAN/WAN technology. Used to provide virtual connections between the SGSN and GGSN. Gi - The point of connection between GPRS and the external networks, each referenced by the Access Point Name. This will normally be implemented using IP WAN technology. Gr - Interface between the HLR and SGSN that allows access to customer subscription information. This has been implemented using enhancements to the existing GSM C7 MAP interface. Gs - Optional interface that allows closer co-ordination between the GSM and GPRS networks Gc - Optional interface that allows the GGSN access to customer location information Page 2

3 2. Dimensioning GPRS Elements The network dimensioning process defines the quantity and configuration of the key network components required to support demand forecast. Typical outputs include: Number of extra radio carriers required Number SGSNs/GGSNs Number of switch sites and their location Sizing of Gn/Gi/Gb links This output is achieved through analysis of demand forecast, system planning limits, and operator planning rules. An overview of these key drivers is shown in Table 1. Dimensioning Driver Key Elements Demand Forecast Input Number GPRS Subscribers Busy Hour Throughput per Subscriber % Attached Subscribers % Active PDP Contexts GPRS Hotspots/Busy Hour System Planning Limits (supplier specific) Max cells/sgsn Peak Throughput per SGSN/GGSN Max attached Subs/SGSN Max Active PDP Contexts/ GGSN Operator Planning Rules Peak Planning Threshold per Component Capacity Lead Time; Burstiness; Headroom Service Restoration Policy Table 1 Key GPRS Network Dimensioning Drivers The demand forecast typically includes the projected number of subscribers over the planning period along with the average throughout per subscriber. The call mix associated with the demand forecast will drive the average throughput rates. Typical applications and relative throughput level are shown in Table 2 for different call mix types. Typical Applications Consumers Subshcribers WAP Information services Banking Business Subscribers Mobile office / schedule File transfer WWW browsing Machine-to-Machine Subscribers Telemetry Metering Vehicle location Page 3

4 E-commerce Average Throughputs Medium High Low Table 2 GPRS Call Mix Types The following sections show how these drivers impact the dimensioning of the GPRS radio and core network domains. Radio domain dimensioning Since the radio interface is shared with voice and data, both GSM and GPRS demand forecasts must be considered together as part of the GPRS radio dimensioning process. The output of this process will be the number of additional radio carriers required to support the demand forecast, some of which were drive by voice requirements, the others required to support GPRS capacity, even though they will be shared by both in the operational network. The way in which GPRS shares radio capacity with circuit switched traffic in a GSM cell is depicted in Figure 2. In this figure, the individual timeslots are allocated to either circuit switched traffic or IP traffic according to demand. This brings with it the first design decision for mobile networks: do you allocate capacity on a first come first served basis, or provide priority to either the circuit switched or IP traffic? Furthermore, if priority schemes are employed, are these used uniformly across all the timeslots available in the cell, or are some timeslots prioritised to voice and others to IP? Packed switched Figure 2. Shared radio capacity between circuit switched and GPRS packet switched traffic Before making these decisions on priority, it is worth noting the outcome that lack of capacity has on the two different services: - On circuit switched services (typically voice), new calls either originating or terminating in the cell must be granted a free timeslot. If a timeslot is not available, Page 4 Packed switched calls variable rate, variable delay Circuit switched calls (e.g. voice) guaranteed rate, minimum delay Time Circuit switched calls (e g

5 the set up will fail. Outgoing calls will result in the annoying three beeps failure tone whilst incoming calls will either route to an announcement or to a voice mailbox. - On IP services, data encountering a fully utilised cell can still be transported. If there is already IP data using one or more timeslots on the cell, the new data can share the timeslot or timeslots. In this situation the new data flow and the existing data flow receive reduced throughput. This shared use of timeslots can handle up to 7 active data sessions concurrently per timeslot. If, despite this shared timeslot capability, there is no capacity available for the new data session (e.g. all timeslots are in use on circuit switched calls), the data goes into retry mode, where the IP data is proffered to the cell every 5 seconds. With these two options for shared or delayed carriage of data, there is a very high probability of being able to transport IP data, even in a heavily loaded cell. The outcome of these different characteristics for circuit switched and IP traffic means that lack of capacity for voice calls is much more catastrophic than lack of capacity for IP traffic. Thus in the majority of circumstances, it is appropriate to allocate priority use of cell capacity to circuit switched calls, with IP traffic employing whatever residue is available. The notion of running a new data service on residue capacity in the GSM network may seem at first to relegate GPRS to the status of a second class service. However, a few simple calculations will show us that the residue capacity in a typical cell is more than enough to provide a high level of service to IP traffic. Table 3 shows the capacity of a shared 4 carrier ( 30 channel) cell operating at a circuit switched blocking level of 1% - a typical design level. Cell capacity 30 channels Circuit switched 1% blocking 20 Erlangs,ie 20 channels average Resultant capacity for IP data traffic 10 channels Resultant end user IP throughput available 1 100kbit/s Note 1: assumes Coding Scheme 2 ( CS2) ie approximately 10kbit/s per channel. Table 3. Typical loading capability of a GSM cell ( 4 carriers) What this tells us is that in a cell where we can support an average of 20 voice users we can also support a data throughput of 100kbit/s. If each of the data users requires an average throughput of 5kbit/s ( not untypical in a bursty data environment) the cell can also support 20 data users. In practice since only 10-20% of data users will want to transfer data simultaneously, the peak data rate available per user will be in the region of 25 50kbit/s. This simplistic calculation needs to be refined to take account of the probability of multiple users all requiring instantaneous transmission of large files of data, but in practice when such occasions arise the end result will simply be that all users will experience slow data transfer: the files will still transfer successfully. What the foregoing example tells us is that for a large number of cells in a GSM network, the existing capacity of the network will suffice to provide a good quality data service to a large community of data customers. In practice, the take-up of GPRS will not be instantaneous across Page 5 Ti

6 the GSM customer base from day one, so it will be possible to monitor usage and performance as GPRS usage grows, to validate performance expectations. There are a number of cases where the existing capacity of a GSM network will not be sufficient to provide a satisfactory level of service to GPRS users: - In existing network hotspots, where the circuit switched network is congested. - In locations where high usage of GPRS data service is encountered ( e.g. in-building cells) - In multilayer networks where one layer of the network is used in high utilisation mode ie where the blocking level on circuit switched traffic is knowingly driven up in order to achieve high levels of channel utilisation. In all these cases, additional carrier capacity must be provided to offer GPRS traffic suitable throughput. The real answer to the radio network dimensioning challenge will come from experience. Experience will tell us whether the busy hour for voice traffic ( circuit switched ) coincides with that for data traffic. Experience will also tell us whether the geographic spread of data usage matches that of voice. Finally, experience will tell us what sort of use customers make of GPRS, what sort of file sizes are transported, and what sort of speeds they require. Careful monitoring of loading and service levels experienced on GPRS in the growth phase of the service will enable dimensioning decisions to be made ahead of growth. All the foregoing analysis and discussion has assumed provision of equal performance across GPRS users on a GSM network. However, the GPRS standards proved for users to be given differential service levels. In particular, users may be offered a precedence class that promotes their data to first in the queue when encountering shared radio (or Core network) resources. Once this feature is developed by equipment vendors (probably in 2001) it will be possible to offer a subset of GPRS users premium service, guaranteeing high levels of throughput even if the cell they are in is heavily loaded. Finally, dimensioning of the GPRS radio capability requires the dimensioning of the specific GSM network base station equipment and base station controllers (BSCs), and the GPRS Packet Control Units (PCUs) associated with each BSC. The primary system constraints and operator planning rules that drive BTS and BSC dimensioning are specified in Table 4. GPRS Radio Element Base Station Transceiver (BTS) Base Station Controller/ Packet Controller Unit (BSC/PCU) Key Dimensioning Factors Busy Hour (BH) kbps per cell Spare capacity in GSM network Location of GPRS hotspots Coincidence of GPRS BH with GSM Number of cells monitored Number of active TS per cell Number of E1 ports for Gb interface Table 4 Key radio dimensioning factors Page 6

7 Core domain dimensioning The system constraints and operator planning assumptions impact the dimensioning of each GPRS Core network element differently. An overview of how these drivers apply to each core GPRS Network element is shown in Table 5. GPRS Core Element Key Dimensioning Factors Gb Network Average throughput per user Number of cells per BSC/PCU SGSN BH number of Attached Subscribers BH peak throughput Number of cells per SGSN Number of Routing Areas per SGSN Number of Gb E1 ports Gn Network BH number of Active PDP contexts BH peak throughput Total number of Routing Areas Number of switch sites GGSN BH number of Active PDP contexts BH peak throughput Number of Gi ports required Gi Network APN throughput requirements Customer security requirements Customer cost considerations Table 5: Dimensioning Factors for GPRS Network Elements Dimensioning example The following SGSN/GGSN dimensioning example shows how the call mix can determine which factors drive the final system configuration. In this example, two different call mixes are used, one driven by a high percentage of consumer subscribers, the other driven by corporate customers. The systems constraints and operator planning rules are assumed to be constant for both scenarios. The assumptions used for this example are shown in Table 6, and explained below. System Constraints SGSN Capacity: 25,000 Attached Subscribers 4 Mbps Peak Throughput GGSN Capacity: 50,000 Active PDP Contexts Page 7 Operator Planning Rules 60% Burst Margin 70% Planning Headroom 6 Month Look Ahead Assume 60% Attached Subs Assume 95% Active PDP contexts

8 10 Mbps Peak Throughput Table 6. Example system constraints and operator planning assumptions Attached Subscribers (AS) Any subscribers that have the GPRS functionality on their phone enabled is considered attached and is monitored by the SGSN. The SGSN monitors subscriber activity to maintain always connected capability whether or not any data is being transferred. A 60% attached subscribers planing rule assumes that 60% of all GPRS subscribers have their GPRS functionality enabled at the GPRS busy hour (BH). Active PDP Contexts Subscribers that have set up a virtual link with an Access Point Name (APN) require connectivity to an internet service or other GPRS service require an active context on the GGSN. The GGSN maintains knowledge of this context as long as the subscriber is attached, whether or not data is being transferred. A 95% active PDP contexts planning rule assumes that almost all attached subscribers have requested access to at least one APN. An average number of active PDP contexts per subscriber must also be assumed, probably between one and two per subscriber. Burst Margin The average throughput provided in a call mix as a percentage of the peak throughput that an SGSN or GGSN can support in the GPRS busy hour (BH). If the call mix provides an average throughput for dimensioning purposes, then a 60% burst margin indicates that only 60% of an SGSN or GGSN s peak throughput should be used when calculating equipment quantities. Planning Headroom To allow for unforeseen surges in AS and PDP contexts for specific SGSNs or GGSNs, a planning limit should be assumed below the SGSN or GGSN s peak capacity. This planning head room can also allow spare capacity to be used in one or more switch sites as part of a service restoration policy to support the capability that would be lost in the event of the loss of a switching complex. Look ahead This is the demand forecast lead time allowed for in network dimensioning. This is typically calculated in terms of total subscribers. If a planning rule for a six month lookahead is used to dimension a network to be used in Jan 2001, for example, the network would be sized to meet the number of subscribers forecasted for June Using these system constraints and planning rules, the required number of SGSNs and GGSNs can be calculated from a given Demand Forecast and Call Mix. In the example shown in Table 7, consumer and corporate call mixes are used for the same number of subscribers, resulting in different equipment quantities. Consumer User Call Mix Corporate User Call Mix Avg BH throughput per user: 14 bps Avg BH throughput per user:100 bps Subscribers Year 1: 1 m Subscribers - Year 1: 1 m Subscribers Year 1 + 6months: 1.5 m Subscribers - Year 1 + 6months: 1.5 m Resulting Configuration Resulting Configuration Required SGSNs (Attached Subs): 52 Required SGSNs (Attached Subs): 52 Page 8

9 Required SGSNs (Mbps): 9 Required GGSNs (PDP Contexts): 24 Required GGSNs (Mbps): 4 Required SGSNs (Mbps): 63 Required GGSNs (PDP Contexts): 24 Required GGSNs (Mbps): 25 Table 7. Dimensioning Example for Consumer vs. Corporate Call Mixes In each case, two different methods of calculating the number of SGSNs and GGSNs are used. In the consumer call mix, due to the high number of subscribers and low average throughput, the number of SGSNs is driven by the number of Attached Subscribers, and the GGSN by the number of Active PDP Contexts. In the second example, the call mix is predominantly comprised of corporate customers, and the average throughput is significantly higher. In this case, both the SGSNs and GGSNs are driven by throughput system constraints. Overall, the required equipment levels in the network are slightly higher for the corporate customers than the consumer call mix due to the greater throughput, given the same number of subscribers. In a real world example, the number of subscribers would also vary for the two different call mixes. In summary, for a given set of system constraints and operator planning rules, the call mix will drive the SGSN and GGSN dimensioning: The number of SGSNs required is driven by either throughput or Attached Subscribers - A high % of consumer applications will make the number of Attached Subscribers the driver - A large number of corporate customers may make throughput the driver The number of GGSNs required is driven by throughput or attached users with active PDP contexts - A high % of consumer applications may make the number of Active PDP Contexts the driver - A large number of corporate customers may make throughput the driver The number of GPRS switching sites follows from the number of SGSNs and GGSNs combined with supplier and operator constraints Supplier constraints can drive the minimum or maximum configuration of SGSNs and GGSNs in a particular location Operator planning rules can drive the number of switch sites required based on service restoration requirements, customer demographics, available space in existing switch sites, etc. 3. Network Performance The two major measures of GPRS performance are: Peak Throughput: the rate at which data is transferred Latency: the time taken for data packets to pass through the GPRS bearer Peak Throughput Page 9

10 An overview of GPRS peak throughputs based on the number of timeslosts available in GPRS handsets, and the Coding Schemes supported by the network is shown in Table 8. Timeslots CS-1 (kbps) Raw throughput/ Useable data Page 10 CS-2 (kbps) Raw throughput/ Useable data / / / / / / / /40 Table 8 Typical GPRS Peak Throughputs The key drivers for peak throughput are: Mobile terminal timeslots / available radio capacity Radio coding scheme Protocol overhead Radio blocking level Timeslots - As shown in the table, the number of timeslots (TS) that a mobile terminal has will drive the peak throughput. Initial GPRS terminals are expected to be on the order of 1 TS uplink and 2 TS downlink (1U/2D). Future handsets are likely to have at least 4 TS downlink, and perhaps multiple uplink TSs. It is also important to remember that the throughputs in Table 8 are peak throughputs and are only achievable if there is sufficient capacity available in the radio network support them. In busy times when multiple GSM and GPRS users are vying for the same timeslots, the actual throughput will vary and will often be well below the peak level. Coding Schemes - The second key driver of throughput is the radio interface coding scheme. As shown in Table 9, higher coding schemes offer greater throughputs. GPRS offers four coding schemes, but initial supplier GPRS radio infrastructure offerings are expected to be limited to CS-1 and CS-2. Higher CS coding levels also result in greater C/I levels which results in reduced coverage areas. For CS-2 the coverage area is not significantly lower that CS-1, but coverage is progressively reduced for CS-3 and 4. All four coding schemes are based on a standard GPRS coded block of 425 bits, which consists of the Uplink State Flag (USF), the user data block (which is of varying size depending on the coding scheme being used) and a Block Check Sequence (BCS for error detection). For CS1, CS2 and CS3, this radio block is then further coded with a ½ rate convolutional code. For CS2 & CS3 this is then punctured (some of the resulting bits of the code are removed) in order to return the total coded length back to 456 bits for transmission. For CS4, no forward error correction code is used and the only error checking is the BCS. The full parameters of the coding schemes are shown in Table 8 below, together with the achieved raw user data rates. Coding Code USF Pre- Radio BCS Tail Coded Punctured Raw

11 Scheme rate bits coded USF bits Block bits excl. USF and BCS bits bits bits bits User Data rate kb/s CS-1 ½ CS-2 2/ CS-3 3/ CS Table 9 - Coding Parameters for the Coding Schemes. Protocol Overhead This causes the true user throughput to be significantly less than the peak raw throughput. The raw user data rates assume an error free channel, and exclude any higher layer protocol overheads, such as TCP/IP, and the link establishment and control overheads. Therefore, the true peak user throughput rates for any of these coding schemes will be lower, as shown earlier in Table 6. Page 11

12 An overview of the GPRS protocols that impact the useable peak data rate is shown below in Figure 3. CS2, 2 TS CS2, 4 TS kb/s kb/s 22.2 kb/s kb/s Application Data TCP/IP kb/s kb/s 23.2 kb/s 26.8 kb/s kb/s kb/s 46.4 kb/s 53.6 kb/s SNDCP Logical link Radio link Radio layer Figure 3 Impact of GPRS protocol overheads on peak throughput By adding headers and error detection trailers, each protocol reduces the effective amount of useable data that is transmitted with a given packet. The method used for the numbers quoted in Table 6 for useable data is the throughput that includes the TCP/IP overhead. This is consistent with data rates quoted for other Internet communications; however, TCP/IP itself adds a 40 bytes header per packet, leaving the final peak throughout of actual application data at 6.81 kbps for CS-1 or kbps for CS-2, assuming no header compression. Radio Blocking Finally, the actual useable peak throughput will be influenced by the quality of the radio environment. The numbers for useable throughput described in this section are all based on an ideal radio environment. The useable throughput achieved in a real world radio environment is likely to be less than this, and can vary widely at different times and locations in the network based on radio blocking levels and number of required re-transmissions. Page 12

13 Latency The major elements of latency and representative latency figures are provided below in Table 9. Latency Element Uplink TBF Establishment 1TS Ongoing Uplink Latency 1 TS Downlink TBF Establishment 2TS Ongoing Downlink Latency 2TS MS Delay 250 ms 100 ms 150 ms 150 ms TBF Establishment 400 ms ms 0 Over the Air Delay 400 ms 400 ms 200 ms 200 ms SGSN/GGSN 50 ms 50 ms 50 ms 50 ms Latency Total 1.1 seconds.55 seconds 1.4 second.4 seconds Table 10 GPRS Latency Examples for 1 TS Uplink, and 2TS Downlink These latency calculations are from the Mobile Station (MS) to the Gi GGSN interface to external networks. Any delays in external to the GPRS network from interconnections via the Internet or in application processing are not included. They are based on BT Labs calculations of the latency that would be expected from the GPRS standard and BT Cellnet experience measuring latency in the lab and operational environment. In this example, a round trip ping which measures the time to send a packet to a server outside the network and receive a response, the total time would be at approximately 2.5 seconds (1.1 uplink plus 1.4 downlink). Based on a 500 ms variance, a round trip ping should generally take 2-3 seconds since radio resources must be allocated for a one time ping. Subsequent transfers would only require about one second round trip as long as the radio resources are allocated to GPRS, since Temporary Block Flow (TBF) establishment would not be necessary. The actual latency experienced by the user could also vary based on the specific way the infrastructure is implemented by suppliers and the applications accessed. More operational experience is required to understand which types of applications will require frequent TBF setups and hence have greater latency. The key elements of GPRS latency are defined below: RLC Block Error Rate - the time taken to retransmit erroneous information due to errors caused by the hostile radio environment. This rate is highly variable depending on radio conditions. For the purposes of the examples in Table 8, ideal radio conditions are assumed and no delay is accounted for. Mobile Station (MS) delay - the time taken by the Mobile Station (MS) to process an IP datagram and request radio resource. This includes the delay from the PC to MS, and the MS processing Page 13

14 time. This delay is typically less than 100ms, with the exception of the processing to establish a request for an uplink TBF channel, which could be in the order of ms. Temporary Block Flow (TBF) Establishment/Cleardown Time - the time it takes the BSS/PCU to provide and release the radio resources required by the user to enable data transfer to take place in either the uplink or downlink. This only occurs on the first transmission, and is not required for subsequent transmissions as long as the resources are allocated to GPRS. The time for TBF establishment can be on the order of 500 ms to 1s and is independent upon the amount of data to be transferred. Throughput over the air delay - the rate at which user data is physically transmitted from the MS to the SGSN once a TBF is established. This delay is directly related to the size of the IP datagram being sent. The smaller the packet size the shorter the delay. For the examples used we are assuming an MTU (Message Transmission Unit) of kbyes for a 400ms delay for 1 TS. This delay is proportionally reduced for multiple timeslot MSs. SGSN/GGSN delay - the delay for the packet to transit through the SGSN and GGSN. This should be almost negligible, and is assumed to be less than 50ms. Acknowledgements The authors acknowledge the excellent work done by the BT Cellnet and BT Labs GPRS teams in developing the technical basis for this report. In particular, Ian Paton s work in developing the latency and peak throughput analysis; Ian Miller, John Williamson, for the GPRS radio feasibility study; Richard Russell and Mark Anderson, for the GPRS Core Network Dimensioning Process; and John Button and Paul Snow from BT Labs for their GPRS performance studies and radio feasibility analysis. Biographies Jim Donahue serves as GPRS Network Programme Manager for BT Cellnet. Prior to his work at BT Cellnet, he was a consultant for Booz, Allen & Hamilton where he served in a variety of IT and network technology consulting and programme management roles. Former clients included major telecom operators in Europe and North America and various U.S. government agencies. He had previously worked for Rockwell International and Grumman Aerospace on the Space Station Programme where he was responsible for designing on-board and satellite-based communications systems and integrating NASA s international partner contributions. Peter Lisle was with BT for 20 year in a variety of network strategy and planning roles, before joining BT Cellnet 4 years ago to lead its network design and planning department. He has now taken on the role of programme manager responsible for all technical and commercial aspects of the development and launch of GPRS-based services. Page 14