Sunteți pe pagina 1din 7

Latency in Broad-band Mobile Networks

Clara Serrano, Beatriz Garriga, Julia Velasco, Julio Urbano, Santiago Tenorio and Manuel Sierra 3G Radio Product Vodafone Group Networks Madrid, Spain clara.serrano@vodafone.com, beatriz.garriga@vodafone.com, julia.velasco@vodafone.com, julio.urbano@vodafone.com, santiago.tenorio@vodafone.com, manuel.sierra@vodafone.com Abstract Whilst currently available speeds on Mobile 3G
Networks are comparable with many fixed broadband carriers, RTT is still away from the values achieved in fixed NWs. There is still significant room for improvement of latency in 3 G NWs to provide the best quality of experience for broadband services, therefore, latency is a key parameter that should be taken into account in future 3G NW configurations. The goal of this paper is to demonstrate the high impact latency has on current Mobile NWs and also to present ways that should be followed to reduce RTT, based on real NW measurements. In realizing this challenge, the current state of HSPA NWs in terms of latency have been analyzed, the sensitivity to latency of that traffic carried by the 3G NW quantified and measuring the proportional contribution of RTT from all the elements of the NW. Initial analysis shows that the UTRAN is, currently, the predominant factor in E2E latency for 3G NWs and therefore this study focuses more on this aspect. The next step has been to study the reduction of latency on 3G real NWs by means of current technology optimization(s) and planned technology deployment. Finally, it was investigated how QoS differentiation can be optimised for sustaining RTT over a potentially congested NW in several of its interfaces and/or nodes. Keywords-component: Latency; RTT; UTRAN; Transport Network; Abbreviations UMTS: Universal Mobile Telecommunications System UTRAN: UMTS Terrestrial Radio Access Network 3G NW: 3rd Generation Network R99: Release 99 of UMTS CN: Core Network RNC: Radio Network Controller Node B: UMTS Base Station UE: User Equipment SGSN: Service GPRS Support Node GGSN: Gateway GPRS Support Node NGME: GGSN Uu interface: UE to Node B interface Iub interface: Node B to RNC interface Iu interface: RNC to SGSN (or MSC) interface Gn interface: SGSN to GGSN interface Gi interface: GGSN to GGSN interface RTT: Round Trip Time TTI: Transmission Time Interval RLC: Radio Link Control E2E: End to end QoS: Quality of Service HSPA: High-Speed Packet Access. HSDPA: High-Speed Downlink Packet Access HSUPA: High-Speed Uplink Packet Access HSPA ph1: HSPA phase 1 implementation LTE: Long Term Evolution ST: Service Time MTU: Maximum Transmission Unit IP: Internet Protocol ATM: Asynchronous Transfer Mode HTTP: HyperText Transfer Protocol Mbps: Mega Bits Per Second ms: Miliseconds DSL: Digital Subscriber Line ACK: Acknowledge (return message)

I.

INTRODUCTION

Today, mobile communication can be ubiquitous (anytime, anywhere, anyplace), personal (instant messaging, picture cards, video messaging) or interactive (push-to-talk [PTT], video telephony, video sharing). Using mobile communication services has never been easier or more entertaining. It is now hard to imagine a world without wireless applications and services. Worldwide, mobile services are playing increasingly important roles in many facets of our society. Just a decade ago, mobile services consisted primarily of basic voice communication whilst today we depend on mobile services not only for voice communication, but also for education, entertainment, healthcare, location and m-commerce. Mobile services have also made significant inroads in developing nations, by improving the quality of life for many of their citizens. As mobile services and their consumer adoption increases around the world, one may say that mobile services are indeed becoming an integral and indispensable part of life. [1] With the introduction of broad-band mobile services, operators now have the capability of offering latency-sensitive and simultaneous voice and data services such as VoIP, PTT, video telephony, video sharing and live multiplayer gaming. The evolution of mobile services has a symbiotic relationship with the evolution of wireless technologies. RAN evolution from R99 to HSPA bearers has significantly increased Mobile 3G NWs peak rates and reduced its Latency. However, while speeds currently available are comparable with many Fixed Broadband Carriers, the RTT is still not comparable to the low values achievable by fixed NWs where Mobile NW can be

978-1-4244-2517-4/09/$20.00 2009 IEEE

seen as a long pipe in comparison [1][3][[4]]. Latency values in the range of only 50 milliseconds for example, may still engender customer dissatisfaction with mobile broadband compared to fixed connections. Where an instantaneous service is used, even small latency contribution could bring a disproportionate effect on the overall user experience. This could well be the killer differentiator for broadband-mobile services. Never mind price, or fancy mashups - mobile companies should think about a really obvious and tangible benefit to end users: Instantaneous has a lot of value.[[5]] If mobile services want really to become truly indispensable by providing the best quality of experience for ubiquitous, personal and interactive services, latency should be one of the main parameters to take into account when considering current and future mobile networks. This leaves us with a wide range of specific techniques for optimizing e2e performance [6] by lowering latency and presents an additional key objective to bear in mind in improving the user experience as opposed to only going for higher peak rates. II. A. CURRENT STATE OF 3G NW IN TERMS OF LATENCY HSPA Network Sensitivity to Latency

Then, sensL is the Latency contribution to the total Service Time of a burst. L will be a constant value between two specific points of the NW, while ST depends on the transmitted burst size. Therefore, sensibility to latency depends strongly on the burst size if the packets to be sent. Considering typical values of 1Mbps for throughput and 100msec for latency, sensitivity to latency is analyzed for a range of typical Burst Sizes present on HSPA NWs (see Figure 1. ). Its observed that sensitivity decreases as burst size increases, because the bigger the transmission time is, the less latency contributes to ST. A service based on traffic bursts up to 50Kbytes size say is generally significantly sensitive to latency.
100

Sensitivity to Latency

80 60 40 20 0
100B 1KB 10KB 100KB 1MB 10MB 100MB

In this section it is shown the high impact latency has on broad-band mobile services, quantifying the sensitivity to latency of mobile NWs and a comprehensive analysis of the current 3G real NW traffic. . How can sensitivity to latency be quantified? For example: if a large amount of data is sent over a period of 2 seconds, with a typical latency of 100ms, then quite probably the latency here isn't very noticeable, but if instead it were a smaller amount of data sent and it were to takes 0,01ms, then the latency is more than the transmission time. The main factor that affects sensitivity with respect to latency is the type of traffic the network transmits, i.e. the size of the traffic bursts. Therefore, a detailed analysis of the burst size distribution in 3G NWs is required to characterize Sensitivity to latency in 3G NW. Sensitivity to Latency can be defined as
L sensL(%) = ST

Burst Size

and more

Figure 1.

Sensitivity to Latency of Burst Sizes

Volume of HSPA NWs Bursts


166.374

Carried Traffic Frequency of Burst

122.869

51.834 30.316 12.068 1.809 262 25

100B

1KB

10KB

100KB

1MB

10MB

100MB

where L is latency between two elements of the NW and ST is the Average Service Time, defined as
A Thr

Burst Size

and more

Figure 2. Volume of Application Data Traffic collected from IU interface in Vodafone Spain NW on June 2008 for HSPA connections

ST = L + Ttx = L +

where Ttx is the time of transmission of the traffic burst, depending on the Throughput (Thr) and the Size of the burst to transmit (A).

In Figure 2. , an example of Burst size distribution for HSPA connections typically observed on a 3G NW is presented. It can be seen that in a typical 3G NW small bursts of traffic (0 to 1KB) are very frequent and makes up the majority of burst sizes, although it should be noted that most of the traffic is carried in much bigger transmissions, ranging from 1MB to 100MB and more.

As a result, combining both analysis (Figure 1. and Figure 2. ), we can deduce that 80% of the data bursts transmitted over the 3G NW have a sensitivity to latency higher than 90% (see Figure 2. ). This means that for nearly 80% of the bursts transmitted over the NW any improvement in throughput is irrelevant, while reductions in latency will result in almost 1:1 improvement of the service time.
Bursts 100
90 80 70 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100

in HSPA NWs CDF

If we consider each session as an experience of an individual user, it can be said that the service given to the 77% of the users in 3G NWs is sensitive to latency a substantial proporption.. This finding corroborates the thinking that latency should be one of the main factors to be taken into account when considering the design of current and future mobile networks. With this in mind, in the following sections an analysis of Latency in a 3 G real NW is shown.
Browsing P2P Streaming Gaming Otros

1% 2% 11% 10 % 77 %

Sensitivity to Latency (%)

Figure 3.

Sensitivity to Latency Distribution of Bursts in the HSPA NW

However, it has to be taken into account the percentage of traffic these bursts represent out of the total traffic a 3G network carries. Conducting the same analysis to the traffic itself but in the quantity of bytes transmitted, not the quantity of bursts, brings the reverse conclusion: 90% of the traffic on HSPA NWs is carried on big bursts (see. Figure 2. ), and big bursts are not sensitive to latency (see Figure 1. ), but are throughput sensitive. In practice, service time for a given application will always be a combination of both types of traffic: throughput sensitive and latency sensitive. For example, in browsing applications, the different composition of objects to download or a number of server-client protocol implementations will introduce a latency component into the throughput-driven massive volume part. An analysis of typical application burst profiles and latency dependence has been performed. Typical applications that can be found in 3G NWs are browsing, P2P, Streaming and Gaming. Taking into account the burst profile found in the different applications from an analysis in existing 3G NWs it was found that browsing applications have a sensitivity to latency between 20-30 %, considered medium sensitivity. An extreme case of latency dependent application has been found in On-line Gaming. Users with ping times to the game server lower than 80..100msec are referred as Low Ping Rates, while users with higher ping times are called High Ping Whiners for using this as an excuse for their relative lack of skills [7]. Whilst on the other hand an extreme case of throughput dependent application are P2P applications. Finally, with an analysis of the proportion of sessions each application represents in the 3G NW (see Figure 5. ), a 3GNW can be characterized in terms of sensitivity to latency : 77% of the sessions performed in 3GNW have medium sensitivity to latency, 10 % have zero sensitivity to latency and 1% of the network is extremely sensitive.

Figure 4.

Service sessions distribution in an existing 3G NW.

B. Contribution to RTT from all HSPA NW elements In order to obtain the contribution of each network element to the RTT, some measurements were performed in a real 3G NW, and the result is shown below in Figure 5. (the breakdown is valid for IP packet size lower than 300bytes). The measurements reveal an intrinsic latency in the 3G NW (HSDPA and HSUPA phase 1) in the range of 65 ms to 180 ms, depending on the transport network topology. The contribution of the radio network goes from 65 to 115 ms, which makes up between 60% and 90% of the time needed to receive the ACK from the server.

65 115*ms

* Over DSL Tx

2-20ms

1-25ms

1-18ms

70-180ms
Low-end latency target <60ms is feasible thanks to new HSPA enhancements from this year

Measured in VF ES Network Huawei June-Aug 2008

High-end latency can be reduced significantly by resolutions in CN & Transport topology/architecture

Figure 5.

Contribution to RTT from all network elements.

The high variance found is explained by the different configurations of the network, and not only in the Uu interface. In Transport and Core networks there are different aspects that impact greatly the RTT, such as the type of connection between the node B and RNC (ATM vs IP protocol, tx topology,

number of intermediate nodes, etc) and the paths defined in the Gn and Gi interfaces. Usually the Core network is designed to be secure, much more focussed on redundancy and robustness than on the time delay, so by optimising the packet routing the RTT could be strongly reduced (see next section). Figure 5. shows the latency for different transport network configurations at the Iub interface (measured in the real NW, from UE to RNC). The difference between a direct connection from node B to RNC with a fixed line (fiber or coaxial feeder) and a microwave link implies on average approximately 20% more delay. Furthermore, if more than one hop is needed when using a microwave link, the latency becomes 50 ms higher than the direct connection from the node B to the RNC. Provision of the Iub interface over DSL link is becoming very common nowadays, but it is not the optimum way to achieve low latency, and in fact it increases itI by on average 20 ms (30% higher RTT) with respect to the optimum link. In summary, RTT is strongly dependent on the access transmission technology and on the number of intermediary hops nodes there are in the connectivity. The lowest values of latency correspond to those nodes connected directly with the RNC, while highest values come from those nodes with several hops (either in the last mile, or through the backhaul).
Latency by Access Tx technology
160 140 120 100 80 60 40 20 0 Direct connectionn to RNC Microwave link 1 hop Microwave link >=2hops Unbundling

Bearing in mind that a radio network contributes to at least 60% of the overall latency one immediate piece of advice to all MNOs wanting to reduce RTT values in their 3G NWs is to implement HSUPA Ph1 (10msec TTI). This technology, compared to R99 radio bearers, improves significantly the Uu interface behaviour in the uplink, reducing the overall RTT in 20% (compared to HSDPA plus R99 in uplink). In addition to the latency gain, HSUPA increases the UL throughput by being able to provide up to 2 Mbps instead of the 384 kbps in R99 (384 kbps is the maximum radio bearer in release 99). This improvement in data rate has also a significant impact in latency when transmitting heavy packets. Transport network topology optimization. If an IP network is used as the core transport (CPN: Core Packet Network), an optimization on the tx routes (these networks are usually built under robustness criteria, not focusing on optimized RTT) could bring a significant improvement (up to 8ms based on the tests performed in Vodafone Spain). In order to optimize RTT values over both the Iu and Gn interfaces, the network topology i.e. distribution of network elements has to be considered: for Iu, RTT values are optimized when RNC and SGSN are co-located, and similarly so for the SGSN and GGSN and the Gn interface. In such topologies, up to 20ms improvements can be achieved (based on the tests performed in Vodafone Spain). Parameter optimization There are three aspects that need to be optimized due to their high influence in latency: HSUPA Grant, IP MTU and RLC protocol. HSUPA Grant[8]: Error! Reference source not found. shows the latency comparison in setting different packet sizes. The measurements were performed in a real network, in a low loaded scenario, with HSDPA and HSUPA ph1 available, and the server directly connected to the Core Network. As packet sizes increases so too does latency, but the magnitude of the linear increase reveals a poor exploit of throughput capabilities. 40msec/KB equates to a 200Kbps net
200 180 160 Average RTT Max RTT Min RTT

Figure 6.

Contribution of different transmission technologies in Iub

III.

LATENCY REDUCTION

As described in the previous section, 3G NWs have a current intrinsic latency between 65 and 180ms - a quite high figure when comparing with the service time of the typical burst of traffic. It has been also shown that for these values of latency, it has a high impact on the services given by 3G NWs. Therefore measures should be investigated as to lowering latency values in order to improve the user experience these are now described. a. Optimization of Current Technology Detailed analysis of the results of the previous section leads us to focus the optimization of the current 3G NWs technology in three aspects: UTRAN software upgrades, transport network topology optimization and parameter optimization. UTRAN software upgrades

Currently too high dispersion. Packet loss suspected, resulting in retransmissions and increased RTT.

Time (ms)

140 120 100 80 60 40 20 0 0 200 400 600 800 1000 1200

RLC protocol affects since the RNC sends the ACKs in DL every 10 ms. Ping Size (Bytes)

The UPA Grant assigns initially the Minimum Bit Rate to all users. For larger packets the impact is higher.

Figure 7.

Packet size (Ping size) impact on RRT

throughput. A conservative setting of the UPA Grant algorithm could be considered thereby permitting differentiated improvement according to QoS class. The linear increase comes from the fact that the high load data packets are transmitted in different TTIs (time intervals in RLC layer). The maximum data to be transmitted in one TTI depends on the HSUPA Grant algorithm, and it is usually conservative initially, starting from a very low data rate, and increasing as the link is proven to be good enough. A less conservative algorithm, at least for high priority connections, would allow transmitting much faster in the radio interfaces. IP MTU: The maximum size of an IP packet is 65,535 bytes while the typical MTU for Ethernet is by default 1,500 bytes. Interfaces using the IP protocol are impacted by MTU since it implies dividing the IP packet in different smaller packets, which in turn requires higher processing capabilities in the intervening or endpoint elements . On the other hand, the smaller the packet is, the faster it reaches the target. Therefore, if the path is reliable enough (i.e. where a non-significant percentage of retransmissions take place), it is more efficient to reduce the MTU - if the processing capability allows it. RLC protocol in UTRAN (UE-RNC)[9] impacts latency as well due to the impact on the packet scheduler. A fixed interval of every 10ms means that even where relatively small packet delays from Core network side (e.g. 2ms) occur, these have to be added to (due to the scheduler) bringing about additional delay in the RNC (e.g. 2ms+8ms). Further analysis is underway in order to minimize this issue.

allows the NW to have different RLC packet sizes, depending on the data load to be sent. Enhanced Always On is the name given to a group of features related to improving the connection (setup) time and battery life: Enhanced F-DPCH, SRB on HSDPA, and Continuous Packet Connectivity. The deployment of LTE is expected to show big improvement in latency. This new technology almost fully specified is/was primarily developed with a focus on high speed data connections. The topology of a LTE network differs from the 3G networks due to the absence of an RNC element (where such RNC functions are moved to the NodeB), and a more distributed Core Network permits more direct and thereby faster communication between the node B and the Core. In addition to this, LTE will not be hindered by existing the RLC protocol limitations impacting 3G. Assuming an optimum transport and Core network configuration and a delay contribution of less than 5 ms, the latency could become less than 20ms.
180 160 140 Time (ms) 120 100 80 60 40 20 0 R99 DL, HSDPA HSDPA HSDPA HSDPA, HSPA+ R99 UL QPSK, 16QAM, 16QAM, HSUPA R99 UL R99 UL HSUPA 2 ms LTE Core NodeB&RNC Computer& UE

b. Planned Technology Deployment In the coming years additions and modifications to the specifications in question will bring about many enhancements in the radio network which will help to reduce the transmission delay such that the bottleneck could even be moved/found somewhere else in the network. Expected RRT improvements for these different technology enhancements are displayed in Error! Reference source not found. This figure contains data from real NW measurements for R99, HSDPA and HSUPA ph1 (10ms), and assumptions for HSUPA ph2 (where a TTI of 2ms is used), HSPA+ and LTE. In recent years, latency has been reduced due to the deployment of HSDPA and HSUPA technologies. Further improvements are expected via new extended features in the Radio network part such as HSUPA 2ms (UPA phase 2), such as Enhanced Layer 2 and Enhanced Always On. Based on estimations, HSUPA phase 2 could reduce RTT by 20% and bring overall RTT to less than 65ms. Enhanced Layer 2 is a feature which improves the efficiency of both the usage of the required RLC (HSDPA related part) and MAC-hs protocols enabling the support of high data rates as well as improved terminal battery life. It
Figure 8.

Expected average RTT for coming radio network technologies

c. Sustain RTT on Congested UTRAN :QoS differentiation In general, the delay in which packets are delivered increases under NW congestion. QoS Load differentiation is one method which can allocate high priority users with better relative performance in terms of delay in addition to throughput [10]. Investigating RTT improvements via QoS Differentiation via smart management of the queues in all the congested interfaces whereby packets from high priority users are placed in the higher priority queue resulted in better responsiveness of the system. The main UTRAN features impacting delay performance per priority can be divided in three blocks[11][12]: - HSPA Schedulers (UL and DL) can now schedule packets on a per user basis i.e. select high priority users over others in the filling up of the TTI.

-Traffic Mapping of QoS in ATM, and application of Diffserv Priorities of IP ensures a differentiated E2E delay. Application of these techniques will permit users a minimum bit rate according to their priority and in addition to higher minimum bit rates, lower delays are possible if burst size is high enough (i.e burst size maximum throughput is higher than minimum bit rate). RTT measurements per priority.: Measurements presented in this section were performed in a UTRAN network. HSDPA 7,2 Mbps and HSUPA 1,4 Mbps technologies were available during measurements performed with a correspondingly enabled mobile in both instances. Measurements were performed on a loaded node B during periods of full Iub occupancy. Investigating how QoS Load Differentiation can impact upon RTT, three groups of users were defined in analyzing the impact and resultant performance in the handling of each group of users over a range of QoS settings: Gold (high priority), Silver (medium priority) and Bronze (low priority). Figure 9. shows that the difference in terms of ping delay increases where Burst size as minimum bit rate (MinBR) settings for low priority users lose effect. Measured delay of Silver~Bronze users increases on average by 84%~245% with respect to Gold users as Burst size increases. Under the same conditions, it has also been measured the time to download a browsing application, as seen in previous sections, one of the most used application in 3G NW. The time to download a typical web page (size 221KB) has been measured in two situations, under congestion and w/o load (Figure 10. ). For Gold users, results for loaded NW kept close to results for unloaded NW. For Silver and especially for Bronze the congestion is more apparent, their RTT is significantly higher than Gold users. This behavior is key for delay sensitive applications. The objective when configuring QoS Load differentiation mechanisms is, for Gold users, to have the same RTT under loaded NW as when NW has no load.
Delay (ms)

Time for web page to download (sec)

- Introducing Flow Control at the Iub enable the buffer to provide more BW to high priority users.

10 9 8 7 6 5 4 3 2 1 0 Gold Silver w/o Load Bronze


2,4

9,3 5,7 3,4


2,4 2,4

Priority Class Figure 10. Ping Download time for a typical web page per priority class (ms)

IV.

CONCLUSION

With the introduction of broad-band mobile services, operators now have the capability of offering latency-sensitive voice and data services. However, at present RTT in Mobile 3G NWs still does not compare to those values achieved in fixed NWs. Optimizing e2e performance should also include aiming to reduce latency values when trying to improve user experience, in addition to higher peak rates or indeed as an objective in itself. The high impact the current values of latency have in 3G NW services was shown and with this in mind ways to quantify the sensitivity to latency in services i.e. understand the impact latency has on the service time of a given application. It is shown that the high impact burst size has on sensitivity: small data bursts usually implies more sensitive traffic and thus applications. Bearing this in mind, a study of the distribution of burst size on a real 3G network was conducted which revealed that total traffic consisted of a high percentage of small bursts. With respect to the services that produce the data packets and the size of these bursts of data, 80% of such data bursts transmitted over the 3G NW are latency sensitive in more than 90% of cases. In terms of Traffic (quantity of bytes transmitted), the result is the reverse - 90% of data on HSPA NWs arise via large bursts, and these are in most if not all cases not sensitive to e2e latency. In practice, applications will always be sensitive to a combination of both of these metrics of throughput and latency. Taking into account both sensitivity to latency and the size of the data burst,, the high proportion of bursts and high proportion of users with applications that are sensitive to latency, latency reduction should be amongst the main goals when considering current and future mobile networks. Measurements in a real NW show presently an inherent latency in the range of 70 ms to 180 ms via the HSPA NW (HSDPA and HSUPA phase 1) although this does , depend on the transport network topology. Where HSUPA is still not available, typical latency values are at least 20 ms longer. RTT is very dependent on the Transport Network architecture and in

Ping Size (Bytes)

Figure 9.

Ping Delay per priority class (ms)

particular the number of hops i.e. intermediary nodes there are. The Iu, Gn and Gi interfaces display a high variance in the measured latency due to - in most cases - the different network topology in the connectivity of an RNC and an SGSN, and an SGSN with an GGSN. In other words, interface delays usually arise from a Transport and Core Networks design that is not focussed upon i.e. not optimized for latency. These three interfaces contribute between 5 and 60 ms to the latency. Based on measurements taken and the results obtained both in existing NWs and via simulations, the key actions to be taken in order to improve the network performance and reduce overall RTT are summarized here together with the impact on latency they have:. - Current network parameter optimization, HSUPA Grant, IP MTU and RLC protocol . - Optimize the Transport network topology so that the route taken by the packets is the most efficient. - Via SW in most cases, maximizing deployment of HSUPAph1 (10msec TTI) across networks, deploy HSUPA phase 2 (2 ms TTI) when available, the key features of HSPA+ and when available in the future deploy LTE. - by applying QoS Load differentiation mechanisms, RTT can be sustained even in a congested UTRAN, providing high priority users with a better performance in terms of delay. It is important to note that the contribution to the sum Latency total of the Radio interface is progressively decreasing. Previously contributing some 75% of latency, Network traces from current HSPA network now show that the UTRAN (RNC, Iub, Node B and Uu interface) contribution is now in some cases below 60% due to Transport network optimizations already. It is expected that eventually, the Radio part will not be the only bottleneck if indeed the bottleneck does not move to another part of the chain - and therefore other NW aspects may need to be optimized too. The Latency analyses performed in a real 3G NWs prove that this is a key factor in the quality of experience of the broad-band mobile services. It has also been shown that even in presently deployed 3G NWs and indeed in future 3G NWs various techniques are available now and in the near future that permit the decreased latency values thereby improving the user experience. REFERENCES
[1] QUALCOM INCORPORATED (2007, December). Evolution of Wireless Applications and Services [Consulted: June 2008] Internet available: http://www.qualcomm.com/common/documents/white_papers/EvolutionWirelessAppsSer vices.pdf [2] CHESHIRE STUART (2001). Its the Latency , Stupid. [Consulted: June 2008] Internet available: http://www.stuartcheshire.org/rants/Latency.html [3] WILLIAM DOUGHERTY (2007, 31 May). Its still the Latency, Stupidpt1. [Consulted: June 2008] Internet available: http://www.edgeblog.net/2007/its-stillthe-latency-stupid/ [4] Flarion, "Low latency: The forgotten piece of the mobile broadband puzzle," Flarion Technologies., NJ 2003.

[5] Dean Bubley (2008, 29 July). Latency in mobile more needs to be done. [Consulted: August 2008] Internet available: http://disruptivewireless.blogspot.com/2008/07/latency-in-mobile-more-needs-to-bedone.html [6] DOUGHERTY WILLIAM (2007, 5 June). Its still the Latency, Stupidpt2. [Consulted: June 2008] Internet available: http://www.edgeblog.net/2007/its-stillthe-latency-stupid-pt2/ [7] The True Gamers Edge [Consulted: June 2008] The Saint [online] Internet available: http://homepages.newnet.co.uk/thesaint/ [8] 3GPP TS 25.321 v8.4.0: Meduyn Access Control (MAC) protocol specificatioin [9] 3GPP TS 25.322 V8.3.0: Radio Link Control (RLC) protocol specification

[10] B. Garriga, F. Dominguez, C.Serrano, S. Tenorio and E. Asensio, ,QoS Load Differentiation Application in a UTRAN, IEEE, to be published in April 2009 [11] 3GPP TS 23.107 v5.13.0,: " QoS Concept and Architecture ".

[12] 3GPP TS 25.433: " UTRAN Iub interface Node B Application Part (NBAP) signalling ".

S-ar putea să vă placă și