GSoC 2012 LTE MAC Scheduler: TDMT, FD-MT, TTA, TD-BET and FD-BET

GSoC 2012 LTE MAC Scheduler: TDMT, FD-MT, TTA, TD-BET and FD-BET

[Tom Henderson, 2012-07-20 20:00:59]

Looks great!  Keep up the good work.

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rst
File src/lte/doc/source/lte-design.rst (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:695: The Throughput to Average (TTA) scheduler
[FCapo2012]_ can be considered as an intermidiate between MT and PF.
intermediate solution

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:696: The metric used in TTA is caculated as
follows:
calculated

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:705: caculation method already is shown in MT
and PF. Meanwhile, :math:`R_{i}(t)` in bit/s stands
calculation

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:707: achievable rates is how to get MCS. For
:math:`R_{i}(k,t)`, MCS is caculatd by subband cqi while
calculated

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:709: the achievable rate of paritular RBG is
only related to FD scheduling.
particular

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:715: used in TTA is caculated as follows:
calculated

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:722: where :math:`T_{j}(t)` is the past
througput performance perceived by the user :math:`j` and can be caculated by
the
throughput

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:722: where :math:`T_{j}(t)` is the past
througput performance perceived by the user :math:`j` and can be caculated by
the
calculated

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:726: prioriy metric). Then scheduler assigns
one RBG to this UE, it calculates expected throughput of this UE and uses it
priority

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.rst
File src/lte/doc/source/lte-testing.rst (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:356: scheduler and Time Domain Maximum
Throughput (TDMT) scheduler repectively,
respectively

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:367: throughput value for first UE by the
throughput achievable of a sinlge UE
single

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:379: The curves labeled "MT" in Figure XX
represent the test values
Figure XX?

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:409: to 1. Therefore, the throughput get by
TD-BET is equal to that of PF. On the other hand, FD-BET performs
s/get/obtained

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
File src/lte/model/fdbet-ff-mac-scheduler.cc (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
src/lte/model/fdbet-ff-mac-scheduler.cc:246: "The number of TTIs a CQI is valid
(default 1000 - 1 sec.)",
would be clearer to state "(default 1000 ms)" so that the units are clearer.

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
src/lte/model/fdbet-ff-mac-scheduler.cc:298: {
please run "utils/check-style.py -l 3 -i <filename>" on all of your new files to
fix indentation coding style issues and other things.

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
src/lte/model/fdbet-ff-mac-scheduler.cc:347: // TODO: Implementation of the API
can a user actually mistakenly call this from the public API, and get no-op
behaviour?   It may be safer to make a "NS_FATAL_ERROR ("Unimplemented");" or
remove from the API here rather than just return.  Or is this soft error OK from
a modeling perspective.  Please discuss with your mentors.

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
src/lte/model/fdbet-ff-mac-scheduler.cc:455: std::map <uint16_t, double>
estAveThr;                                // store exptected average throughput
for UE
expected

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
src/lte/model/fdbet-ff-mac-scheduler.cc:601: //NS_LOG_DEBUG ( Simulator::Now()
<< this << "Allocate user " << newEl.m_rnti << " tbSize  "<< tbSize << " PRBs "
<< rbgPerRnti*rbgSize << " mcs " << (uint16_t) newDci.m_mcs.at (0) << " layers "
<< nLayer);
remove dead code

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
File src/lte/model/fdbet-ff-mac-scheduler.h (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
src/lte/model/fdbet-ff-mac-scheduler.h:46: unsigned int lastTtiBytesTrasmitted;
Transmitted

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
src/lte/model/fdbet-ff-mac-scheduler.h:53: * \defgroup FF-API
FdBetFfMacScheduler
Is doxygen grouping to this low level necessary?

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
src/lte/model/fdbet-ff-mac-scheduler.h:90: void
TransmissionModeConfigurationUpdate (uint16_t rnti, uint8_t txMode);
missing doxygen

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdmt-ff-mac-schedu...
File src/lte/model/fdmt-ff-mac-scheduler.cc (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdmt-ff-mac-schedu...
src/lte/model/fdmt-ff-mac-scheduler.cc:301: return;
another duplicate

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdmt-ff-mac-schedu...
src/lte/model/fdmt-ff-mac-scheduler.cc:479: if ((cqi1 > 0)||(cqi2 > 0)) // CQI
== 0 means "out of range" (see table 7.2.3-1 of 36.213)
should this be && instead of || ?  (not sure what the model is here, but it
seems from the comment that you may want to check both values)

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdmt-ff-mac-schedu...
src/lte/model/fdmt-ff-mac-scheduler.cc:583: worstCqi.at (j) = 1;
should this code break out of the for loop, or can multiple values of j be
labeled "worst one"?

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdmt-ff-mac-schedu...
File src/lte/model/fdmt-ff-mac-scheduler.h (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdmt-ff-mac-schedu...
src/lte/model/fdmt-ff-mac-scheduler.h:82: void
TransmissionModeConfigurationUpdate (uint16_t rnti, uint8_t txMode);
doxygen

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tdbet-ff-mac-sched...
File src/lte/model/tdbet-ff-mac-scheduler.cc (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tdbet-ff-mac-sched...
src/lte/model/tdbet-ff-mac-scheduler.cc:306: return;
delete line 306

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tdbet-ff-mac-sched...
File src/lte/model/tdbet-ff-mac-scheduler.h (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tdbet-ff-mac-sched...
src/lte/model/tdbet-ff-mac-scheduler.h:89: void
TransmissionModeConfigurationUpdate (uint16_t rnti, uint8_t txMode);
doxygen

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tta-ff-mac-schedul...
File src/lte/model/tta-ff-mac-scheduler.cc (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tta-ff-mac-schedul...
src/lte/model/tta-ff-mac-scheduler.cc:311: for (uint16_t i = 0; i <
params.m_logicalChannelConfigList.size (); i++)
? I don't see where your loop variable is being used.

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tta-ff-mac-schedul...
src/lte/model/tta-ff-mac-scheduler.cc:525: if (metric - metricMax > 1e-3)
perhaps make 1e-3 a defined constant or attribute rather than magic number

http://codereview.appspot.com/6353082/diff/1/src/lte/test/lte-test-fdbet-ff-m...
File src/lte/test/lte-test-fdbet-ff-mac-scheduler.cc (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/test/lte-test-fdbet-ff-m...
src/lte/test/lte-test-fdbet-ff-mac-scheduler.cc:59: : TestSuite
("lte-fdbet-ff-mac-scheduler", SYSTEM)
UNIT?

http://codereview.appspot.com/6353082/diff/1/src/lte/test/lte-test-fdbet-ff-m...
src/lte/test/lte-test-fdbet-ff-mac-scheduler.cc:204: //   LogComponentEnable
("LteEnbPhy", LOG_LEVEL_ALL);
delete dead code (here and other places)

[Biljana Bojović, 2012-07-24 12:26:04]

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rst
File src/lte/doc/source/lte-design.rst (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:655: Maximum Throughphut (MT) Scheduler
Throughput

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:664: The caculation of achievable rate in FDMT
and TDMT is as same as the one in PF.
calculation

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:689: faireness to UEs in poor chennel
condition.
fairness

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:689: faireness to UEs in poor chennel
condition.
channel

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:689: faireness to UEs in poor chennel
condition.
channel

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:708: :math:`R_{i}(t)` is caculated by wideband
cqi. TTA scheduler can only be implemented in frequency domain (FD) because
calculated

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:724: with largest priority metirc and
allocates all RBGs to this UE. On the other hand, in the frequency domain blind
metric

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-design.rs...
src/lte/doc/source/lte-design.rst:729: math:`T_{j}(t)` of all UE. Then the
scheduler will use the same way to allocate RBG for a new UE which has the
the next

[Biljana Bojović, 2012-07-24 13:43:23]

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.rst
File src/lte/doc/source/lte-testing.rst (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:356: scheduler and Time Domain Maximum
Throughput (TDMT) scheduler repectively,
.

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:360: SINR value and a different number of
UEs. The test consists on
values

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:364: scheduler allocates all RBGs to the
first UE defined in script. This is because
the

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:368: at the given SNR, while reference
throughput value for other UEs by zero.
maybe to reformulate phrase "calculate by sth."

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:389: TTA test case are as same as the test
for MT scheduler. More complex test case need to be
are the same as

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:389: TTA test case are as same as the test
for MT scheduler. More complex test case need to be
needs

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:411: is an integer multiple of the number of
RBG(UE). In this case, FD-BET always allocate the same number of RBGs
Could you reformulate the sentence? It is not very clear, because in the same
sentence is explained in which case those schedulers behave similarly, and the
definition of UE(RBG) is provided

http://codereview.appspot.com/6353082/diff/1/src/lte/doc/source/lte-testing.r...
src/lte/doc/source/lte-testing.rst:419: both scheduler should give the same
amount of throughput to each user.
per TTI or averaged?

[Biljana Bojović, 2012-07-25 16:42:44]

Very nice work, there are just a few things to be changed.

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
File src/lte/model/fdbet-ff-mac-scheduler.cc (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
src/lte/model/fdbet-ff-mac-scheduler.cc:306: return;
one return is redundant

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdbet-ff-mac-sched...
src/lte/model/fdbet-ff-mac-scheduler.cc:530: 
remove dead code

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdmt-ff-mac-schedu...
File src/lte/model/fdmt-ff-mac-scheduler.cc (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdmt-ff-mac-schedu...
src/lte/model/fdmt-ff-mac-scheduler.cc:570: //                 NS_LOG_DEBUG
(this << " RBG " << (*itMap).second.at (k) << " CQI " <<
(uint16_t)((*itCqi).second.m_higherLayerSelected.at ((*itMap).second.at
(k)).m_sbCqi.at (0)) );
to remove unused code

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdmt-ff-mac-schedu...
src/lte/model/fdmt-ff-mac-scheduler.cc:576: {
why those 2 "if" conditions are not only one condition?

http://codereview.appspot.com/6353082/diff/1/src/lte/model/fdmt-ff-mac-schedu...
src/lte/model/fdmt-ff-mac-scheduler.cc:601: }
those 2 else conditions have the same body, could you make only one of them?

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tdbet-ff-mac-sched...
File src/lte/model/tdbet-ff-mac-scheduler.cc (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tdbet-ff-mac-sched...
src/lte/model/tdbet-ff-mac-scheduler.cc:455: //NS_LOG_DEBUG (this << "
ALLOCATION for RBG " << i << " of " << rbgNum);
remove if unused

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tdbet-ff-mac-sched...
src/lte/model/tdbet-ff-mac-scheduler.cc:476: else
if this algorithm should work like described here:
https://www.nsnam.org/wiki/index.php/GSOC2012LTEScheduling, it should select the
user with minimum average past throughput and assign resources to him, or to
devide among RBs among the users depending on this last average throughput.

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tdmt-ff-mac-schedu...
File src/lte/model/tdmt-ff-mac-scheduler.cc (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tdmt-ff-mac-schedu...
src/lte/model/tdmt-ff-mac-scheduler.cc:496: return;
this could be checked before "for loop"?

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tta-ff-mac-schedul...
File src/lte/model/tta-ff-mac-scheduler.cc (right):

http://codereview.appspot.com/6353082/diff/1/src/lte/model/tta-ff-mac-schedul...
src/lte/model/tta-ff-mac-scheduler.cc:548: allocationMap.insert (std::pair
<uint16_t, std::vector <uint16_t> > ((*itMax), tempMap));
if i understand good RB are always assigned to the user with the highest metric,
or the highest achievable rate, but in
https://www.nsnam.org/wiki/index.php/GSOC2012LTEScheduling is explained that it
will " averaging the resources evenly between the users"