What is RNC Load Balancing and How Implement It
Content
Enabling LDR Algorithms to reduce the CE congestion case
Author
Diego
Rojas
Representative
Date
Camilo
Internal
ID
703932
UMTS RAN
Product
Wireless
2010-05-25
Version
(VxxxRxxxCxxxBxxx)
RNC V200R010
Checked by
Name：
FAQ – What is RNC Load balancing and how implement it?
Phenonem
a：
How can RNC Load Balancing help to reduce congestion in SPU and DPU boards?
Alarm：
ALM-1302 DSP CPU Overload
ALM-1301 CPU Overload
Reason
Analysis
This guide aims to explain how the load balancing is implemented on the control
plane and the user plane.
：
Feature Description
MPU subsystem performs resources management on the user plane and control
plane within its own sub rack and within all RNC’s sub racks. The process for each
plane is described as follows:
LOAD BALANCING ON CONTROL PLANE
When a service arrives to the host SPU board [SPU where the NodeB is configured]
the SPU decides whether to process or forward the request service to the MPU in the
host sub rack [Sub rack where the NodeB is configured] according with its current
load.
The SPU can be in one of the next three stages:
1.
2015-8-26
Light Load: The SPU will process all services request that arrive to it and
No. 1, Total 8
Enabling LDR Algorithms to reduce the CE congestion case
2.
3.
Internal
will accept any other services the MPU transfer to it.
Sharing Out its Load: The SPU will forward all the services that arrive at it
to the MPU and will accept any service MPU send at it.
Overload: The SPU will forward all the services that arrive at it to the MPU
and won’t accept any service MPU send at it.
The thresholds that define the load sharing on control plane are configured by SET
CTRLPLANESHAREPARA:
LOAD BALANCING ON USER PLANE
The MPU manage and allocates all the user plane resources on the sub rack. When
the load in the sub rack is heavy the MPU forwards the resources request to other sub
racks.
When the service request arrives, it requests for user plane resources. In this case,
the SPU requests the MPU in the sub rack for resources. The MPU process as
follows:
1. If the user plane load of the host sub rack is lower than the value of
UserPlnSharingOutThd in the command SET USERPLNSHAREPARA, the MPU
will allocate the user plane resources on the DSP with the lightest load within the sub
rack.
2. If the user plane load of the host sub rack is higher than the value of
UserPlnSharingOutThd in the command SET USERPLNSHAREPARA, the MPU
will forward the request to the MPU in sub rack with the lightest load within the
RNC. Then, this MPU will allocate the user plane resources on the DSP with the
lightest load within its sub rack.
2015-8-26
No. 2, Total 8
Internal
Enabling LDR Algorithms to reduce the CE congestion case
Process
This Test procedure specify how to test the feature in one commercial RNC eith two
Subracks: [modify it according with the RNC configuration]
Step
Description
Time
1
Perform control plane test and check
results
5 minutes
Execute the MML: SET CTRLPLNSHAREPARA command to
set control plane resource pool parameter :SET
CTRLPLNSHAREPARA: CtrlPlnSharingOutThd=0;
2
Rollback control plane configuration
1 minute
Execute the MML: SET CTRLPLNSHAREPARA command to
set control plane resource pool parameter to the original
configuration :SET CTRLPLNSHAREPARA:
CtrlPlnSharingOutThd=50;
3
Perform user plane test and check results
5 minutes
Inhibit DPU boards on slot 11 and slot 12 by INH BRD:
INHT=PHYSICAL, SRN=0, SN=11; INH BRD:
INHT=PHYSICAL, SRN=0, SN=12;
Execute the MML: SET USERPLNSHAREPARA command to
set control plane resource pool parameter : SET
USERPLNSHAREPARA: UserPlnSharingOutThd=50,
UserPlnSharingOutOffset=5;
Remark: This threshold and the number of inhibit boards could
change according with the current load of the control plane
4
Rollback control plane configuration
5 minutes
Execute the MML: SET USERPLNSHAREPARA command to
set control plane resource pool parameter to the original
configuration: SET USERPLNSHAREPARA:
UserPlnSharingOutThd=90, UserPlnSharingOutOffset=5;
And execute UIN BRD: to uninhibit the DPU boards
UIN BRD: UINT=PHYSICAL, SRN=0, SN=11;
UIN BRD: UINT=PHYSICAL, SRN=0, SN=12;
Suggest
and
Summary
：
2015-8-26
Summary for User Plane test:
Bellow pictures show the DSP load on Subrack number 1 before, and during the test:
1.1
Before the test: DSP load on subrack 1 keeps at 7.1% for the whole
subrack in normal conditions as bellow.
No. 3, Total 8
Enabling LDR Algorithms to reduce the CE congestion case
1.2
2015-8-26
Internal
During the test: New services request start to be handle on subrack 1
after the user plane sharing out threshold change to 50% at 22:44 [red
line] and traffic load on subrack 0 keeps higher than this threshold.
No. 4, Total 8
Enabling LDR Algorithms to reduce the CE congestion case
Internal
Fig 1: Real time for DSP usage in Subrack 1 before and after
USRPLNSHAREOUTTHD change to 50%
Bellow picture shows how the user plane load for new connections is share from a
congested subrack [blue line] to another with light load.
Summary for Control Plane test:
2015-8-26
No. 5, Total 8
Internal
Enabling LDR Algorithms to reduce the CE congestion case
Attached picture shows how load is been forwarding to the lightest sub rack after the
SPU is in sharing out its load state:
* Picture provided by local TSD
Attached
：
Others:
Workaround for SPC090
The connection between SPU boards and all the other boards installed on the RNC is
managed by IPC links. These links also are used to communicate all the MPU
subsystems allocated on the RNC. Due to a software bug on SPC050 this
communication is broken and just can be recovered resetting the RNC or patching it
to SPC090.
IPC status
---------Low-end
state
=
=
2015-8-26
=
IPC
administrative
Available
Low-end
time
operation
Link broken
Low-end
state
IPC
operate
status
changed
<NULL>
No. 6, Total 8
Internal
Enabling LDR Algorithms to reduce the CE congestion case
Not sent codes in low-end send
queue
=
0
Connect
sent
=
request
received
=
request
=
Packets
low-end
0
Connect
response
Packets
low-end
0
Connect
received
=
response
=
Packets
low-end
0
Codes
sent
low-end
10049230
Connect
sent
Packets
low-end
successfully
0
Codes
switched
=
=
receive
and
discarded
due
to
low-end
send
queue
but
switch
0
Codes
failed
successfully
0
Codes
overflow
low-end
=
low-end
successfully receive
0
The total bytes of message with low priority due to low-end
sent
=
0
The total bytes of message with low priority due to low-end
received
=
0
High-end
state
=
=
=
IPC
administrative
<NULL>
High-end
time
operation
<NULL>
High-end
state
IPC
operate
status
changed
<NULL>
Not send codes in high-end send
queue
=
0
Number of connect request Packets high-end
sent
=
0
Number
received
=
of
connect
request
Packets
high-end
0
Number of connect response Packets high-end
sent
=
0
Number
received
=
of
connect
response
Packets
high-end
0
Number of codes high-end successfully
sent
=
0
Number
switched
2015-8-26
=
of
codes
high-end
successfully
receive
and
0
No. 7, Total 8
Enabling LDR Algorithms to reduce the CE congestion case
Internal
Number of codes discarded due to high-end send queue
overflow
=
0
Number of codes high-end successfully receive but switch
failed
=
0
The total bytes of message with low priority due to high-end
sent
=
0
The total bytes of message with low priority due to high-end
received
=
0
As a result, load sharing between sub racks was affected due to communication
broken between the MPUs. SPC090 solves this problem.
2015-8-26
No. 8, Total 8
Author
Diego
Rojas
Representative
Date
Camilo
Internal
ID
703932
UMTS RAN
Product
Wireless
2010-05-25
Version
(VxxxRxxxCxxxBxxx)
RNC V200R010
Checked by
Name：
FAQ – What is RNC Load balancing and how implement it?
Phenonem
a：
How can RNC Load Balancing help to reduce congestion in SPU and DPU boards?
Alarm：
ALM-1302 DSP CPU Overload
ALM-1301 CPU Overload
Reason
Analysis
This guide aims to explain how the load balancing is implemented on the control
plane and the user plane.
：
Feature Description
MPU subsystem performs resources management on the user plane and control
plane within its own sub rack and within all RNC’s sub racks. The process for each
plane is described as follows:
LOAD BALANCING ON CONTROL PLANE
When a service arrives to the host SPU board [SPU where the NodeB is configured]
the SPU decides whether to process or forward the request service to the MPU in the
host sub rack [Sub rack where the NodeB is configured] according with its current
load.
The SPU can be in one of the next three stages:
1.
2015-8-26
Light Load: The SPU will process all services request that arrive to it and
No. 1, Total 8
Enabling LDR Algorithms to reduce the CE congestion case
2.
3.
Internal
will accept any other services the MPU transfer to it.
Sharing Out its Load: The SPU will forward all the services that arrive at it
to the MPU and will accept any service MPU send at it.
Overload: The SPU will forward all the services that arrive at it to the MPU
and won’t accept any service MPU send at it.
The thresholds that define the load sharing on control plane are configured by SET
CTRLPLANESHAREPARA:
LOAD BALANCING ON USER PLANE
The MPU manage and allocates all the user plane resources on the sub rack. When
the load in the sub rack is heavy the MPU forwards the resources request to other sub
racks.
When the service request arrives, it requests for user plane resources. In this case,
the SPU requests the MPU in the sub rack for resources. The MPU process as
follows:
1. If the user plane load of the host sub rack is lower than the value of
UserPlnSharingOutThd in the command SET USERPLNSHAREPARA, the MPU
will allocate the user plane resources on the DSP with the lightest load within the sub
rack.
2. If the user plane load of the host sub rack is higher than the value of
UserPlnSharingOutThd in the command SET USERPLNSHAREPARA, the MPU
will forward the request to the MPU in sub rack with the lightest load within the
RNC. Then, this MPU will allocate the user plane resources on the DSP with the
lightest load within its sub rack.
2015-8-26
No. 2, Total 8
Internal
Enabling LDR Algorithms to reduce the CE congestion case
Process
This Test procedure specify how to test the feature in one commercial RNC eith two
Subracks: [modify it according with the RNC configuration]
Step
Description
Time
1
Perform control plane test and check
results
5 minutes
Execute the MML: SET CTRLPLNSHAREPARA command to
set control plane resource pool parameter :SET
CTRLPLNSHAREPARA: CtrlPlnSharingOutThd=0;
2
Rollback control plane configuration
1 minute
Execute the MML: SET CTRLPLNSHAREPARA command to
set control plane resource pool parameter to the original
configuration :SET CTRLPLNSHAREPARA:
CtrlPlnSharingOutThd=50;
3
Perform user plane test and check results
5 minutes
Inhibit DPU boards on slot 11 and slot 12 by INH BRD:
INHT=PHYSICAL, SRN=0, SN=11; INH BRD:
INHT=PHYSICAL, SRN=0, SN=12;
Execute the MML: SET USERPLNSHAREPARA command to
set control plane resource pool parameter : SET
USERPLNSHAREPARA: UserPlnSharingOutThd=50,
UserPlnSharingOutOffset=5;
Remark: This threshold and the number of inhibit boards could
change according with the current load of the control plane
4
Rollback control plane configuration
5 minutes
Execute the MML: SET USERPLNSHAREPARA command to
set control plane resource pool parameter to the original
configuration: SET USERPLNSHAREPARA:
UserPlnSharingOutThd=90, UserPlnSharingOutOffset=5;
And execute UIN BRD: to uninhibit the DPU boards
UIN BRD: UINT=PHYSICAL, SRN=0, SN=11;
UIN BRD: UINT=PHYSICAL, SRN=0, SN=12;
Suggest
and
Summary
：
2015-8-26
Summary for User Plane test:
Bellow pictures show the DSP load on Subrack number 1 before, and during the test:
1.1
Before the test: DSP load on subrack 1 keeps at 7.1% for the whole
subrack in normal conditions as bellow.
No. 3, Total 8
Enabling LDR Algorithms to reduce the CE congestion case
1.2
2015-8-26
Internal
During the test: New services request start to be handle on subrack 1
after the user plane sharing out threshold change to 50% at 22:44 [red
line] and traffic load on subrack 0 keeps higher than this threshold.
No. 4, Total 8
Enabling LDR Algorithms to reduce the CE congestion case
Internal
Fig 1: Real time for DSP usage in Subrack 1 before and after
USRPLNSHAREOUTTHD change to 50%
Bellow picture shows how the user plane load for new connections is share from a
congested subrack [blue line] to another with light load.
Summary for Control Plane test:
2015-8-26
No. 5, Total 8
Internal
Enabling LDR Algorithms to reduce the CE congestion case
Attached picture shows how load is been forwarding to the lightest sub rack after the
SPU is in sharing out its load state:
* Picture provided by local TSD
Attached
：
Others:
Workaround for SPC090
The connection between SPU boards and all the other boards installed on the RNC is
managed by IPC links. These links also are used to communicate all the MPU
subsystems allocated on the RNC. Due to a software bug on SPC050 this
communication is broken and just can be recovered resetting the RNC or patching it
to SPC090.
IPC status
---------Low-end
state
=
=
2015-8-26
=
IPC
administrative
Available
Low-end
time
operation
Link broken
Low-end
state
IPC
operate
status
changed
<NULL>
No. 6, Total 8
Internal
Enabling LDR Algorithms to reduce the CE congestion case
Not sent codes in low-end send
queue
=
0
Connect
sent
=
request
received
=
request
=
Packets
low-end
0
Connect
response
Packets
low-end
0
Connect
received
=
response
=
Packets
low-end
0
Codes
sent
low-end
10049230
Connect
sent
Packets
low-end
successfully
0
Codes
switched
=
=
receive
and
discarded
due
to
low-end
send
queue
but
switch
0
Codes
failed
successfully
0
Codes
overflow
low-end
=
low-end
successfully receive
0
The total bytes of message with low priority due to low-end
sent
=
0
The total bytes of message with low priority due to low-end
received
=
0
High-end
state
=
=
=
IPC
administrative
<NULL>
High-end
time
operation
<NULL>
High-end
state
IPC
operate
status
changed
<NULL>
Not send codes in high-end send
queue
=
0
Number of connect request Packets high-end
sent
=
0
Number
received
=
of
connect
request
Packets
high-end
0
Number of connect response Packets high-end
sent
=
0
Number
received
=
of
connect
response
Packets
high-end
0
Number of codes high-end successfully
sent
=
0
Number
switched
2015-8-26
=
of
codes
high-end
successfully
receive
and
0
No. 7, Total 8
Enabling LDR Algorithms to reduce the CE congestion case
Internal
Number of codes discarded due to high-end send queue
overflow
=
0
Number of codes high-end successfully receive but switch
failed
=
0
The total bytes of message with low priority due to high-end
sent
=
0
The total bytes of message with low priority due to high-end
received
=
0
As a result, load sharing between sub racks was affected due to communication
broken between the MPUs. SPC090 solves this problem.
2015-8-26
No. 8, Total 8
