Considerations for 3G and 4G over VSAT - Global · PDF fileConsiderations for 3G and 4G over VSAT ... • Most significant KPI moving forward to 3G is RRC ... – Optimization solution

Considerations for 3G

and 4G over VSAT

Michele Di Paolo

VP, Mobile Network Operator (MNO) Solution and Business Development

ComtechEFData

Considerations for 3G

over VSAT

( and why 3G failed to deliver over VSAT)

2

UMTS (3G) Topology

• The Universal Mobile Telecommunications

System (UMTS) is a third generation mobile cellular

system

3

Considerations for 3G over VSAT

• MNO expectations:

– Deliver services with terrestrial level KPIs.

– Respect existing network transport (L2 transport)

– Respect MNO defined QOS

– High availability

4

• Keys to achieving MNO expectations:

– Low jitter and delay variations

– Support L2 protocols (VLAN/MPLS) and L3

– 3GPP Compliant QOS performance

– Error free delivery (AUPC/ACM)

Maintaining highest KPI

• Most significant KPI moving forward to 3G is RRC (Radio

Resource Control) Success Rate which manages the control

plane between RNC and the Ue and is responsible for Radio

Resource management.

• Jitter in excess of 10msecs resulting in packets not being

delivered on time, retransmissions, dropped Ues, etc.

5

It is not just about delivering bandwidth

RRC KPI

Success rate

of 99.95%

recorded.

Maintaining highest throughput

• 3G NodeB manages UL traffic from the Ues.

• Additional delay and/or jitter variation in UL traffic is

detected at RNC which sends congestion notification

to NodeB to throttle down UL.

• This results in FLOW control which negatively impact

UL and DL traffic which we call the Bart effect.

6

3G traffic expectations

• TCP session throughput expectation across GEO satellite

<1Mbps/session.

• The disruptive effect of packet loss, congestion control, etc

can reduce effective real-life speeds to 160kbps or less.

• Packet loss is a real issue for 3G Iub with reported 5% RAN

packet being considered lost/delayed over TDMA… source

Ericsson.

7

Packet per second considerations

• Requires highly capable processing engines to deal

with high packet per second rates found in 3G

environment.

• Expect 2000 packets per 1 Mbps of 3G traffic!

8

Overhead: Iub Voice

Comtech EF Data Confidential 9

• IP NodeB voice is supported via R99 techniques

which call for voice to be encapsulated in

IP/UDP/tunnels and so also carry significant

overhead.

– L2/L3 overhead greater than 50%

benefits from header compression are well known with savings on the order of 55% (AMR 12.2) to 75% (AMR 4.7).

Total bytes 4

# of Bytes 14 8 20 8 7 1 1 12-32 2 4

L1 HDR L2 HDR IP HDR UDP HDR FP HDR MAC HDR RLC HDR AMR CODEC FP CRC FCS

COMP HDR

# of Bytes 1-3

50 23-43

COMP PAYLOAD

variable

Overhead: Iub Data

Comtech EF Data Confidential 10

• Data (3G) consists of multiplexed traffic from multiple

users segmented into 40/80 bytes slices every

2msec.

• Iub data average overhead is in the range of 27%.

Typical benefit of Header and payload compression can deliver 30%

savings.

Total bytes 4

# of Bytes 14 8 20 8 7 1 1 12-1600 2 4

L1 HDR L2 HDR IP HDR UDP HDR FP HDR MAC HDR RLC HDR DATA (3G) FP CRC FCS

COMP HDR

# of Bytes 1-3

50 23-1600

COMP PAYLOAD

variable

Why can’t we optimize 3G

performance?• PDU are segments/slices of traffic from End User

equipment which are multiplexed together at the RNC

to fill a 2msec burst time at the NodeB.

11tr

FP

MU

X

User traffic is segmented into 40 and 80 bytes

PDU at the RNC.

The RNC multiplexes up to 21 or 42 PDU into a

single FP mux equivalent to 2msec radio burst.

Do we need Jumbo frames?

• 3 types of HSDPA codes per cell are defined which also defines

the type and size of PDU and FP frames.

• HSDPA can support 5, 10, and 15 codes per cell across 3 cells.

** requires Jumbo frame support

* average Iub data overhead 1.27

12

Codes/

cell

PDU

(bytes)

PDU/

FP

FP MTU

(bytes)

PPS/

cell/

BW/ cell

(Mbps)

Iub BW/ cell

(Mbps)

5 40 21 840* 500 3.4Mbps 4.3Mbps

10 80 21 1680** 500 6.8Mbps 8.5Mbps

15 80 42 3360** 500 13.6Mbps 17.1Mbps

5

5

5

10

10

10

15

15

15

High availability and QOS

• QOS and two way AUPC/ACM required to maintain

highest site availability.

13

> 17 dB Fade

(VSAT BW

reduced

~80%)

Conclusion on 3G

• Need to maintain highest level of KPI by minimizing

delay and jitter

• Support for different MNO backhaul preferences - L2

(VLAN/MPLS) and/or L3 IP processing

• Need to support high pps rate, provide effective

3GPP compliant QOS and high availability

mechanisms to ensure RIGHT traffic is prioritized and

delivered.

14

Introduction to LTE

performance

(and what it means for end users throughput.)

15

LTE rollout

16

• Most commercial rollout started with Cat 3 or 4

networks supporting 100/150Mbps over 20Mhz

spectrum.

• Carrier Aggregation evolved to raise capacities.

• In most Urban areas, 2/3/4CA are prevalent while the

race is on for 1G using either 100Mhz spectrum or

4CA with 4x4 MIMO (and newer 256QAM MCS)

Definitions and speed?

• What does a 20Mhz 2X2 MIMO (referred to as 150Mbps link)

actually mean?

– 20Mhz – 10% guard band = 18Mhz

– LTE subcarrier are 15khz (18Mhz/15khz=1200 subcarriers)

– 12 subcarriers are combined into a Resource block (1200/12 =

100RB which is also equivalent to 12X15khz=180khz).

– Time domain slots is 0.5msecs

– There are 7 OFDM symbols in normal case.

– Finally, Resource Element (RE) is smallest element and defined the how many modulated symbols can be processed in 1msec:

▪ 1200 subcarrier X 7 OFDM symbols X 2 Time slots = 16800 REs

17

Actual speed…

– With 16800 RE and LTE 64QAM modulation we get :

▪ 16800 x 6 bits/hz = 100.8Mbps

– 2 X 2 MIMO implies 2 outbounds so:

▪ 2 x 100.8Mbps = 201.6Mbps

– Overhead and additional elements reduce BW by 25%.

▪ 201.6Mbps X (1-25%) = 150Mbps

18

• BUT, those are ideal conditions…actual throughput is

determined by RF performance of receiving handset.

How much backhaul is enough?

• Reality check.. 90% of subscribers in

rural are medium to far distance from

tower.

• eNb selects from 28 MCS (Modulation

and Coding Scheme) to find best match

for transmission.

• Assumption for average BW: – 10% close, 60% medium, 30% far -> 50/60Mbps

19

How does CA play into this?

• For rural applications, lower frequency range is best suited for

wide coverage (700/800/900khz).

• Overlays of higher frequency can be added to provide “boost”

capacity.

20

• Devices which do not support

CA can be pushed onto best

access frequency and operate

in single frequency mode.

• CA supporting devices can

receive load balanced BW from

both carriers.

• Carriers to be aggregated need

not be same size (10+10,

10+20, 20+20+10 etc)

CA allows for exciting deployment options!

• CA two way communications

• CA with supplemental downlink

21

So where are we now?

22

5G?

LTE capacity conclusions

• LTE backhaul capacity needs are regularly over-

stated

• Reported capacity is fixed per site and only

determined by base station technology… need to

work through the information to determine the

backhaul requirement.

• When it comes to rural, extended range (low

frequency) is best way to monetize the investment.

23

Optimizing LTE for

Quality of Experience

24

LTE Network Topology

25

Key backhaul interface is S1

interface between the Serving

GW and the eNb.

Main advantage of LTE eNb vs 3G RNC/NodeB is the unification of the radio

resource and packet management into the eNb.

26

Ue HTTP

Ue TCP

Ue IP

GTP

UDP

IP

L2

L1

HTTP

TCP

IP

Ue HTTP

Ue TCP

Ue IP

GTP

UDP

IP

L2

L1

S1

PDCP

RLC

MAC

L1

PDCP

RLC

MAC

L1

HTTP

TCP

IP

SGW/PGW (4G)) InterneteNb (4G)Ue)

TCP Session#1

LTE Traffic encapsulation

• LTE traffic between SGW/PGW and the eNb is

encapsulated across GPRS Tunneling Protocol.

• Opening up the GTP tunnel allows for vendors to

provide optimization solutions across actual Ue traffic

Why optimize LTE protocol?

• High bandwidth = better Quality of Experience (QoE)

• Contributors to poor QOE:

– Web sites complexity has grown: 4 yrs ago average website

size was 750kB and consisted of 20-30 objects. Today, average website in 2MB+ and consist of 200+ objects across dozens of

domain/hosts.

– Problem 1: DNS lookups must be performed to identify hosts server addresses on the internet before object DL can commence.

Over terrestrial links DNS lookups account for roughly 1/3 the time

to DL the site.

– Problem 2: Typical browsers can open 6 simultaneous TCP sessions per domain. If there are dozens of objects to DL from a

particular domain, those are managed sequentially.

27

Contributors to poor QOE

– Problem 3: TCP “Slow Start mechanism” highly sensitive to network delay

(greater the delay, the longer it takes TCP session to ramp up). More than

80% of web downloads are small files (<100kbytes) which means TCP

seldom reach full speed and most transfers are in “Slow Start”.

– Problem 4: Autoplay media downloads and buffers traffic slowing down

other object downloads.

– Problem 5: Encryption. Encryption. Encryption.

28

0

10000

20000

30000

40000

50000

600 1200 1800 2400 3000 3600 4200 4800

Byte

s

msec

100kB transfer over VSAT

Understanding Internet Traffic

• Sandvine predicts that “by the end of 2017… Internet traffic that

is more 75 to 80% encrypted in most markets. “

• It is expected that Internet encryption levels will stabilize at

roughly 90% within the next 2 yrs.

29

Internet Traffic (2015)

Encrypted Not Encrypted

Internet Traffic (2016)

Encrypted Not Encrypted

Internet Traffic (2017)

Encrypted Not Encrypted

LTE Optimization solutions

LTE optimization objectives:– Provide mobile users a better internet/broadband experience

though faster reaction time and DL speeds.

– More efficient backhaul bandwidth utilization (backhaul link is fuller

.. Longer)

– More efficient remote site radio management since DL are faster

thereby allowing radio resources to be re-allocated more often.

– Local acknowledgements at the remote so re-transmissions are managed locally (no end to end retransmissions across satellite).

– Optimization solution to scale to meet widest range of support LTE

capability deployments (integrated solution vs external appliance).

– Critical -> transparency to ensure accurate Bytes IN vs Bytes OUT

for billing purposes

30

Optimization: as fully perfect, functional, or effective as possible

Improve Mobile User Experience

• Over traditional IP networks (such as ISP or enterprise

networks), TCP acceleration has been used to mitigate and

resolve those performance limitations to improve Internet user

experience.

– TCP acceleration provides breaking of long end-to-end 3WHS

control loops to several smaller control loops by intercepting and

relaying TCP connections within the network

31

@

No TCP Acceleration –long end-to-end 3WHS (SYN,

SYN-ACK, ACK) and “slow start” for TCP ramp up.

VSAT

With TCP Acceleration – local 3WHS and local ACKs

mitigates the “slow start” for TCP.

Local TCP session Local TCP sessionPacket secure “relay”

TCP Acceleration benefit

• Faster throughput ramp up

• Higher throughput

• Faster packet loss recovery .. important in the case of

supporting LTE over Ka band.

32

DNS Caching

• DNS Caching is often performed in ISP environment to

“speed up” host address resolution over VSAT.

– Example: Modern websites such as CNN require 70+ DNS

inquiries, DNS inquires take 1/3rd of the time to DL a site and so

DNS caching significantly improves website QOE (see cnn-with-

dns-caching video)

• LTE optimization should provide DNS Caching to

eliminate long VSAT delays in Host name resolution

WHILE preserving and respecting BYTE counts for

billing.

33

What about overhead?

• LTE is very efficient at the radio level BUT it comes with significant

overhead. Considering L2 transport, each GTP packet has 94bytes

of overhead.

• Internet packet distribution: 55% < 100, 15% < 1200, 35% >1200

bytes, LTE overhead would average 12%.

– Example:

▪ TCP ACK is 4bytes but requires 98byte across LTE (20x overhead to payload ratio)

▪ TCP average packet size of 700bytes + 94bytes overhead = 12% overhead!!

34

EPC/LTE overhead (94 bytes per user TCP payload)MAC = 14 bytes, VLAN = 4 bytes, IP = 20 bytes, UDP = 8 bytes, GTP = 8 bytes, Ue IP = 20 bytes, Ue TCP = 20 bytes

Effectiveness of LTE Protocol

Optimization

Technique Technology Benefit

BW saving

optimizations

Image resizing, object caching, byte

caching

Data traffic savings: ???:

too much encryption).

Voice traffic: 50% or more

TCP Acceleration: Optimizes the TCP protocol across

VSAT eliminating the speed barrier of

modern OS.

Good (standard SCPC is

good for large transfers but

not as effective for small

transfers).

DNS Caching provides local DNS cache lookup to

eliminate long VSAT delays in Host

name resolution.

HIGH (speeds up time to

first byte and total DL time)

Web evolution (http2

and/or CEFD

Turbostreaming)

efficient multiplexing of single domain

objects into persistent high bandwidth

TCP connection

HIGH (multiplexing allows

small and large object

multiplexing into same TCP

flows allowing higher BW

and efficiency)

35

What about the modem side?

• Pt2MPt is more suited to LTE than Pt2Pt.

• High throughput LTE needs to be supported by high PPS

modem with high peak DL (100s of Mbps) and UL

(dozens of Mbps) traffic rates support.

• As higher BW tend toward Ka/Ku, dynamic two way BW

management and 3GPP compliant QOS is required to

maintain high availability and KPIs.

36

LTE Conclusion

• Performance optimized LTE over VSAT – Available,

proven, deployed with major operators worldwide.

• Traditional deployment options:

– Rural Macro cells: combine LTE with existing 2G/3G to provide

highest consumer coverage options.

– Rural Macro cells and small cells: exciting deployment options

using CA concepts.

• New markets:

– Small cells: fill “blank” spots in network coverage.

– Metro-edge market: provide highspeed LTE overlay capacity to

existing terrestrial backhaul.

37

Comtech EF Data

2114 West 7th Street

Tempe, AZ 85281

USA

Tel +1.480.333.2200

FAX +1.480.333.2540

[email protected]

www.comtechefdata.com