You are on page 1of 54

INTER FREQ LOAD BALANCING IN LTE CONTEXT

Prepared by Thierry Billon


december 2013,

Agenda

1. Scope/rational
2. MAIN ASSUMPTIONS
3. REFERENCE CASE :ALL SITES EQUIPED WITH ONLY ONE CARRIER
4. F2 CARRIER ONLY ON CENTER SITE
5. ALL SITES EQUIPPED WITH F1 and F2
1. Comparison equal loading from idle mode and unequal
loading (70/30) without Load balancing
2. Comparison unequal loading from idle (70/30) with without
load balance
Conclusion and further study

Copyright 2009 Alcatel-Lucent, d.r.

Scope/rational
This document deals with load balancing between carrier in a LTE context
Previous documents have analyzed idle mode selection as root for
establishing an initial balance between frequency layers, with different
carrier priority ( priority to F2 layer, equal priority). Some solution with
adaptive tuning of the idle mode threshold targeting equal %PRB load were
tested.( Interfreq_loadbalance_FDD_SON_5 dec 2012)
Connected mode load balancing was studied , with as conclusion that its
efficiency could be linked to the existence of relatively stable connection in
demand and in time (e.g. GBR) , and traffic profile dependant. Short data
session, dormancy timer, measurement gap were seen to cause difficulties
(IFLB_IDLE_CONNECTED_september 2013)
The previous traffic model was very short data session. Here we use an
updated model , more heavy tail, with larger mean data volume
On previous version, most cases were done with all sites equipped with dual
band. Here we also look at the case of dual band only in center site

3 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

MAIN ASSUMPTIONS
o. NETWORK TOPOLOGY
o. LOAD BALANCING
o. RADIO PARAMETERS

4 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

NETWORK TOPOLOGY

We use a wrap round 7* 3*2 sector sites topology.


The two carriers are 780 Mhz and 2100 Mhz, with 5 Mhz bandwidth and look at
two cases
The first case is with dual band only on the center site
The second case with where each site is equipped with 3 sectors on F1
and 3 sectors on F2 ( Collocated )

5 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

LOAD BALANCING (1/2)


The load balancing algorithm is the load equalization one
If ( %PRB load cell > Threshold1 and delta%PRB load with other cell >
Threshold2)
Offload up to N users to other cell
The user to be offloaded are selected by taking among all connected users
those that have the highest average PRB consumption
In the simulation the filter time constant for the cell %PRB load is taken to
20s. The filter time constant for the per user PRB consumption is 2s, with one
sample every 1 ms
N is set to 10 users maximum to be offloaded per second. Users are selected
until their %PRB contribution is lower than half the delta%PRB or N=10
Threshold1 is set to 50%, threshold2 to 20%
Compared to november edition, we add in this december edition some cases,
in particular a comparison between using actual prb load or using semi static
prb
Copyright 2009 Alcatel-Lucent, d.r.

LOAD BALANCING (2/2)

Once a user has been selected, it is configured in measurement Gap with 80


ms cycle, and offload if it fullfills an A4 event
The A4 event is set in the simulation to -104 dbm. In the simulation all users
fulfills this criteria. Hand over failure are not simulated
All users are dual band
The simulation does not model the data transfert from one cell to the other
( assume immediat transfert)

Copyright 2009 Alcatel-Lucent, d.r.

RADIO PARAMETERS

8 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

TRAFIC MODEL

9 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

TRAFIC MODEL (1/3)


Along this document we use best effort traffic only, with a traffic profile that
match about the statistic as provided by Omar Salvador the 24 of october
about Mobile originated and mobile terminating user traffic from NY market
lower Manhattan (see two next chart)
We use 70% connection Mobile originated
The dormancy timer is set to 5s
The traffic is generated as a Poisson process, and the average arrival rate is
varied from 1 to 20 call attempt per cell per second.
For each arrival rate, each simulation corresponds to 500 second of
observation
The maximum number of connected users par cell is set to 200. Arrival when
the number of connected user is 200 are rejected

10 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

TRAFIC MODEL :Best effort short data packet session (2/3)

11 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

TRAFIC MODEL :Best effort short data packet session (3/3)


UL traffic (Kbytes) per connection for Mobile Terminated (MT)
Min. 1st Qu. Median
0.00

0.09

0.25

Mean 3rd Qu.


6.97

Max.

2.00

74560.00

DL traffic (Kbytes) per connection for Mobile Terminated (MT)


Min. 1st Qu.

Median

0.00

0.44

0.22

Mean 3rd Qu.


40.70

2.00

Max.
210500.00

DL traffic (Kbytes) per connection for Mobile Originated (MO)


Min. 1st Qu.
0.0

0.1

Median
1.0

Mean 3rd Qu.


104.6

9.0

Max.
1375000.0

UL traffic (Kbytes) per connection for Mobile Originated (MO)


Min. 1st Qu.

Median

0.00

1.00

0.21

12 | Titre de la prsentation | Mois 2009

Mean 3rd Qu.


16.38
Copyright 2009 Alcatel-Lucent, d.r.

5.00

Max.
260100.00

RESULTS
For each cases, for each arrival rate, we provide as results
The % PRB consumption, the served traffic per cell, the mean and 5% worst case and
the number of rejected or offloaded users.
We provide the results,
With the case were idle mode provide a good balance
Compare the case where idle mode provide unbalance w/o load balancing

13 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

REFERENCE CASE : ALL SITES EQUIPED WITH ONLY ONE CARRIER F1

14 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

REFERENCE CASE( 1 carrier 5Mhz)

15 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

REFERENCE CASE ( 1 carrier 5Mhz)

With one carrier 5 Mhz, the maximum number of users (200) is achieved with
an average arrival rate of 10 call attemps per second. At this load, the served
downlink traffic is about 4.3 Mbit/s , all resources consumed . The uplink
traffic is much less

16 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

F2 carrier only on center site, F1 on all sites

1) For the simulation, we create an unbalance between F1 and


F2 by having 70% of access on F1 and 30% on F2 .We compare
results with without load balancing
2)Comparison on using actual PRB load and static PRB load, and
Idle mode setting benchmark

17 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

F2 carrier only on center site, F1 on all sites

1) Unbalance 70/30 and comparison results with without load


balancing

18 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

PRB consumption w/o Load balancing: carrier 5Mhz unequal user repartition
(70 /30) from idle mode, F2 only on center site, F1 on all sites.

19 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Served traffic on F1 and F2 w/o load balancing : carrier 5Mhz unequal user
repartition (70 30) from idle mode, F2 only on center site, F1 on all sites.

20 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Rejected and offloaded users w/o Load balancing: carrier 5Mhz unequal user
repartition (70 30) from idle mode, F2 only on center site, F1 on all sites:

21 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Traffic repartition on F1 and F2 w/o load balancing : carrier 5Mhz unequal user
repartition (70 30) from idle mode, F2 only on center site:

22 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Connection time on F1 and F2 w/o load balancing 2 carrier 5Mhz unequal user
repartition (70 30) from idle mode, F2 only on center site

23 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Load balancing conclusion with dual band F1 F2 only on center site, all other
sites on F1, with initial 70/30 users unbalance
With an initial unbalance of 70/30 for call attempts in F1 and F2, load
balance allows to go up to 20 call attempts per second while without load
balance, user start to be rejected ( more than 200 connected user in F1) at 13
call attempt per second.
The time of connection with load balance is minimized and the served traffic
better ( 10 Mbit/s at 20 call attempts per second, rather than 7 Mbits without
load balancing)
This asks, at this load, for 6 hand over per second.
When comparing the case of initial unbalance 70/30 plus load balance and the
case of a good tuning of idle mode threshold ( either manually or by self
adaptive mechanism, see next chart) we obtain almost the same results with
the difference that idle mode tuning dont need hand over ,is not trafic
profile dependant, and has some trend to better equalize.

24 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

PRB and Served traffic on F1 and F2 : 2 carrier 5Mhz, F2 only on center site,
with idle mode access tuning for equal %PRB

25 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

F2 carrier only on center site, F1 on all sites


2)Comparison on using actual PRB load and static PRB load, and
Idle mode setting benchmark

26 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Comparison when using actual or static PRB as trigger, and idle mode setting
benchmark
Here we compare Load balancing using either actual PRB consumption or
static PRB consumption .For these simulation we assume that priority has
been set badly set to F2 resulting to most of the user entering on the F2
carrier .For static PRB , we use a minBeBitRate of 60 kbit/s
For idle mode setting benchmark, we use priority on F2, with self tuned
Threshold X high using PRB load from time 0. The SIB information is updated
every 15 s
The simulation is conducted from time 0 ( no load on the network) to 1000 s
The load ( call attempts per second per dual cell) is increasing linearly with
time, as follow

27 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

2 carrier 5Mhz dual band on center site Idle mode Priority F2 badly set, LB with
actual PRB

28 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

2 carrier 5Mhz dual band on center site Idle mode Priority F2 badly set, LB with
static PRB

29 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

2 carrier 5Mhz dual band on center site Idle mode Priority F2, self learnt
ThreshXhigh, no LB

30 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Conclusion for single site dual band


Offloading with PRB load as trigger fails to recover large user unbalance that
could exist due to idle mode setting. The trend is to offload the more PRB
demanding users, that will fullfill rapidly the PRB of the other layer, which
will preclude further offloading.
For this case of dual band in the center site only , loading first F1 ( case with
70% in F1 and 30% in F2) leads to different Load balancing behavior than
loading first F2. F2 having no interference , the PRB consumption to trig load
balancing from F2 to F1 is obtain latter than in F1
Using static PRB as trigger rather than actual PRB load gives better
performance ( In term of user balance). The reason is that the static PRB
mechanism include a weighting with the number of connected users and a
minimum bit rate .
If large user unbalance between frequencies exist due too idle mode setting,
load balancing using PRB either actual or static fails to recover the unbalance,
and the hard limit on maximum number of users is hit on one frequency, well
before this limit is hit with a good idle tuning
31 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

F1 and F2 carrier on all sites


1) Comparison equal loading from idle mode and unequal
loading (70/30) without Load balancing
2) Comparison unequal loading from idle (70/30) with without
load balance
3) Comparison on using actual PRB load and static PRB load and
idle mode setting benchmark

32 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

F1 and F2 carrier on all sites


1) Comparison equal loading from idle mode and unequal
loading (70/30) without Load balancing

33 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

2 carrier 5Mhz all sites equal and unequal user repartition (70 30) from idle
mode no load balance

34 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Served traffic 2 carriers 5Mhz all sites equal and unequal user repartition from
idle mode no load balance

35 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Connection time 2 carrier 5Mhz all sites equal and unequal user repartition
from idle mode no load balance

36 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Traffic volume ratio 2 carriers 5Mhz all sites equal and unequal user repartition
from idle mode no load balance

37 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Rejected users (limit=200 connected per cell) 2 carrier 5Mhz equal and unequal
users repartition (70 30) from idle mode no load balance

38 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Conclusion for 2 carriers 5Mhz all sites ,equal and unequal user repartition
from idle mode no load balance
For all sites equipped with 2 carriers, when each carriers have sufficient
coverage ( small ISD makes that 2100 can cover the cell edge of 780 Mhz)
equal user repartition from idle mode allows to go up to 20 call attempts per
second with equal load on each layer
When there is unbalance from idle mode( 70/30), the maximum number of
connected user is hit at an arrival rate of 13 call attempts per second

39 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

F1 and F2 carrier on all sites


2) Comparison unequal loading from idle (70/30) with without
load balance using actual %PRB load as trigger

40 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

2 carrier 5Mhz all sites unequal user repartition (70 30) from idle mode W/o
load balance, selecting more PRB consuming users for offload

41 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Served traffic 2 carriers 5Mhz all sites unequal user repartition from idle mode
W/0 load balance

42 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Connection time 2 carrier 5Mhz all sites unequal user repartition from idle
mode w/O load balance

43 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Traffic volume ratio 2 carriers 5Mhz all sites unequal user repartition from idle
mode w/o load balance

44 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Rejected users (limit=200 connected per cell) 2 carrier 5Mhz equal and unequal
users repartition (70 30) from idle mode no load balance

45 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Load balancing conclusion with dual band F1 F2 on all sites


When all sites are equipped with dual band and with an initial unbalance of
70/30 for call attempts in F1 and F2, load balance with actual % PRB load is
not efficient. We do not obtain at all equivalent performance or trends as
when idle mode ensure equal user repartition.
While the % PRB load tends to be equalized between each band, the offload
traffic (number of offloaded users ) is low, and the limit of 200 maximum
connected users in F1 is not pushed far away than without load balancing
This is due to the algorithm that offload the more consuming PRB users. With
the trafic profile we use (chart 10, 11 and 12), a large proportion of users
have small payload, while 15% have large demand. Since we offload the users
with largest PRB consumption, the trend is to select the largest payload users,
and probably also the worst in SNR, and after offloading , they will occupied
all the bandwidth they get, making the PRB load of F2 going high ( see chart
34), precluding further offloading from F1 .

46 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

F1 and F2 carrier on all sites


3) Comparison on using actual PRB load and static PRB load, and
Idle mode setting benchmark

47 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Comparison when using actual or static PRB as trigger, and idle mode setting
benchmark
Here we compare Load balancing using either actual PRB consumption or
static PRB consumption .For these simulation we assume that priority has
been set to F1 resulting to all the users entering on the same carrier .For
static PRB , we use a minBeBitRate of 60 kbit/s
For idle mode setting benchmark, we use priority on F2, with self tuned
Threshold X high using PRB load from time 0. The SIB information is updated
every 15 s
The simulation is conducted from time 0 ( no load on the network) to 1000 s
The load ( call attempts per second per dual cell) is increasing linearly with
time, as follow

48 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

2 carrier 5Mhz all sites Idle mode Priority F1, LB with actual PRB load

49 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

2 carrier 5Mhz all sites, Idle mode Priority F1, LB with static PRB load

50 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

2 carrier 5Mhz all sites Idle mode Priority on F2, self learnt ThreshXhigh, no LB

51 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

Conclusion
Offloading with PRB load as trigger fails to recover large user unbalance that
could exist due to idle mode setting. The trend is to offload the more PRB
demanding users, that will fullfill rapidly the PRB of the other layer, which
will preclude further offloading.
Using static PRB as trigger rather than actual PRB load gives better
performance ( In term of user balance). The reason is that the static PRB
mechanism include a weighting with the number of connected users and a
minimum bit rate .
If large user unbalance between frequency exist due too idle mode setting,
load balancing using PRB either actual or static fails to correct the unbalance,
and the hard limit on maximum number of users is hit on one frequency, well
before this limit is hit with a good idle tuning.
In the simulation, when the hard limit of maximum connected users is hit,
new arrival are rejected. We do not redirect them . Would we redirect them
on the other frequency, this will probably reestablish some balance
52 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

www.alcatel-lucent.com
www.alcatel-lucent.com

53 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

www.alcatel-lucent.com
www.alcatel-lucent.com

54 | Titre de la prsentation | Mois 2009

Copyright 2009 Alcatel-Lucent, d.r.

You might also like