maki/openairinterface5g
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Harmonized FlexRAN scheduler with MAC changes for support of multiple UEs

Browse Source
This commit is contained in:
Xenofon Foukas
2017-01-25 20:02:38 +02:00
parent 422c06c932
commit 974e5beebc
4 changed files with 41 additions and 138 deletions
Download Patch File Download Diff File Expand all files Collapse all files
openair2/LAYER2/MAC/flexran_agent_mac_proto.h
-13
View File
@@ -131,19 +131,6 @@ void _assign_rbs_required (module_id_t Mod_id,
uint16_t nb_rbs_allowed_slice[MAX_NUM_CCs][MAX_NUM_SLICES],
int min_rb_unit[MAX_NUM_CCs]);
int _maxround(module_id_t Mod_id,
uint16_t rnti,
int frame,
sub_frame_t subframe,
uint8_t ul_flag );
int _maxcqi(module_id_t Mod_id,
int32_t UE_id);
void _sort_UEs (module_id_t Mod_idP,
int frameP,
sub_frame_t subframeP);
void _dlsch_scheduler_pre_processor (module_id_t Mod_id,
int slice_id,
frame_t frameP,
openair2/LAYER2/MAC/flexran_agent_scheduler_dlsch_ue.c
+39 -124
View File
@@ -137,8 +137,9 @@ void _store_dlsch_buffer (module_id_t Mod_id,
UE_list_t *UE_list = &eNB_mac_inst[Mod_id].UE_list;
UE_TEMPLATE *UE_template;
for (UE_id=UE_list->head; UE_id>=0; UE_id=UE_list->next[UE_id]) {
for (UE_id = 0; UE_id < NUMBER_OF_UE_MAX; UE_id++) {
if (UE_list->active[UE_id] != TRUE) continue;
if (flexran_slice_member(UE_id, slice_id) == 0)
continue;
@@ -221,7 +222,8 @@ void _assign_rbs_required (module_id_t Mod_id,
// UE_TEMPLATE *UE_template;
// clear rb allocations across all CC_ids
for (UE_id=UE_list->head; UE_id>=0; UE_id=UE_list->next[UE_id]) {
for (UE_id = 0; UE_id < NUMBER_OF_UE_MAX; UE_id++) {
if (UE_list->active[UE_id] != TRUE) continue;
if (flexran_slice_member(UE_id, slice_id) == 0)
continue;
@@ -229,6 +231,10 @@ void _assign_rbs_required (module_id_t Mod_id,
pCCid = UE_PCCID(Mod_id,UE_id);
rnti = UE_list->UE_template[pCCid][UE_id].rnti;
/* skip UE not present in PHY (for any of its active CCs) */
if (!phy_stats_exist(Mod_id, rnti))
continue;
//update CQI information across component carriers
for (n=0; n<UE_list->numactiveCCs[UE_id]; n++) {
CC_id = UE_list->ordered_CCids[n][UE_id];
@@ -293,115 +299,6 @@ void _assign_rbs_required (module_id_t Mod_id,
}
}
// This function scans all CC_ids for a particular UE to find the maximum round index of its HARQ processes
int _maxround(module_id_t Mod_id,uint16_t rnti,int frame,sub_frame_t subframe,uint8_t ul_flag )
{
uint8_t round,round_max=0,UE_id;
int CC_id;
UE_list_t *UE_list = &eNB_mac_inst[Mod_id].UE_list;
for (CC_id=0; CC_id<MAX_NUM_CCs; CC_id++) {
UE_id = find_UE_id(Mod_id,rnti);
round = UE_list->UE_sched_ctrl[UE_id].round[CC_id];
if (round > round_max) {
round_max = round;
}
}
return round_max;
}
// This function scans all CC_ids for a particular UE to find the maximum DL CQI
int _maxcqi(module_id_t Mod_id,int32_t UE_id)
{
LTE_eNB_UE_stats *eNB_UE_stats = NULL;
UE_list_t *UE_list = &eNB_mac_inst[Mod_id].UE_list;
int CC_id,n;
int CQI = 0;
for (n=0; n<UE_list->numactiveCCs[UE_id]; n++) {
CC_id = UE_list->ordered_CCids[n][UE_id];
eNB_UE_stats = mac_xface->get_eNB_UE_stats(Mod_id,CC_id,UE_RNTI(Mod_id,UE_id));
if (eNB_UE_stats==NULL) {
mac_xface->macphy_exit("maxcqi: could not get eNB_UE_stats\n");
return 0; // not reached
}
if (eNB_UE_stats->DL_cqi[0] > CQI) {
CQI = eNB_UE_stats->DL_cqi[0];
}
}
return(CQI);
}
// This fuction sorts the UE in order their dlsch buffer and CQI
void _sort_UEs (module_id_t Mod_idP,
int frameP,
sub_frame_t subframeP)
{
int UE_id1,UE_id2;
int pCC_id1,pCC_id2;
int cqi1,cqi2,round1,round2;
int i=0,ii=0;//,j=0;
rnti_t rnti1,rnti2;
UE_list_t *UE_list = &eNB_mac_inst[Mod_idP].UE_list;
for (i=UE_list->head; i>=0; i=UE_list->next[i]) {
for(ii=UE_list->next[i]; ii>=0; ii=UE_list->next[ii]) {
UE_id1 = i;
rnti1 = UE_RNTI(Mod_idP,UE_id1);
if(rnti1 == NOT_A_RNTI)
continue;
if (UE_list->UE_sched_ctrl[UE_id1].ul_out_of_sync == 1)
continue;
pCC_id1 = UE_PCCID(Mod_idP,UE_id1);
cqi1 = _maxcqi(Mod_idP,UE_id1); //
round1 = _maxround(Mod_idP,rnti1,frameP,subframeP,0);
UE_id2 = ii;
rnti2 = UE_RNTI(Mod_idP,UE_id2);
if(rnti2 == NOT_A_RNTI)
continue;
if (UE_list->UE_sched_ctrl[UE_id2].ul_out_of_sync == 1)
continue;
cqi2 = _maxcqi(Mod_idP,UE_id2);
round2 = _maxround(Mod_idP,rnti2,frameP,subframeP,0); //mac_xface->get_ue_active_harq_pid(Mod_id,rnti2,subframe,&harq_pid2,&round2,0);
pCC_id2 = UE_PCCID(Mod_idP,UE_id2);
if(round2 > round1) { // Check first if one of the UEs has an active HARQ process which needs service and swap order
swap_UEs(UE_list,UE_id1,UE_id2,0);
} else if (round2 == round1) {
// RK->NN : I guess this is for fairness in the scheduling. This doesn't make sense unless all UEs have the same configuration of logical channels. This should be done on the sum of all information that has to be sent. And still it wouldn't ensure fairness. It should be based on throughput seen by each UE or maybe using the head_sdu_creation_time, i.e. swap UEs if one is waiting longer for service.
// first check the buffer status for SRB1 and SRB2
if ( (UE_list->UE_template[pCC_id1][UE_id1].dl_buffer_info[1] + UE_list->UE_template[pCC_id1][UE_id1].dl_buffer_info[2]) <
(UE_list->UE_template[pCC_id2][UE_id2].dl_buffer_info[1] + UE_list->UE_template[pCC_id2][UE_id2].dl_buffer_info[2]) ) {
swap_UEs(UE_list,UE_id1,UE_id2,0);
} else if (UE_list->UE_template[pCC_id1][UE_id1].dl_buffer_head_sdu_creation_time_max <
UE_list->UE_template[pCC_id2][UE_id2].dl_buffer_head_sdu_creation_time_max ) {
swap_UEs(UE_list,UE_id1,UE_id2,0);
} else if (UE_list->UE_template[pCC_id1][UE_id1].dl_buffer_total <
UE_list->UE_template[pCC_id2][UE_id2].dl_buffer_total ) {
swap_UEs(UE_list,UE_id1,UE_id2,0);
} else if (cqi1 < cqi2) {
swap_UEs(UE_list,UE_id1,UE_id2,0);
}
}
}
}
}
void _dlsch_scheduler_pre_processor_allocate (module_id_t Mod_id,
int UE_id,
uint8_t CC_id,
@@ -575,7 +472,9 @@ void _dlsch_scheduler_pre_processor (module_id_t Mod_id,
min_rb_unit[CC_id]=get_min_rb_unit(Mod_id,CC_id);
for (i=UE_list->head; i>=0; i=UE_list->next[i]) {
for (i = 0; i < NUMBER_OF_UE_MAX; i++) {
if (UE_list->active[i] != TRUE) continue;
UE_id = i;
// Initialize scheduling information for all active UEs
@@ -603,24 +502,24 @@ void _dlsch_scheduler_pre_processor (module_id_t Mod_id,
_assign_rbs_required (Mod_id,slice_id, frameP,subframeP,nb_rbs_required,nb_rbs_allowed_slice,min_rb_unit);
// Sorts the user on the basis of dlsch logical channel buffer and CQI
_sort_UEs (Mod_id,frameP,subframeP);
sort_UEs (Mod_id,frameP,subframeP);
total_ue_count = 0;
// loop over all active UEs
for (i=UE_list->head; i>=0; i=UE_list->next[i]) {
for (i=UE_list->head; i>=0; i=UE_list->next[i]) {
rnti = flexran_get_ue_crnti(Mod_id, i);
if(rnti == NOT_A_RNTI)
continue;
if (UE_list->UE_sched_ctrl[i].ul_out_of_sync == 1)
continue;
UE_id = i;
if (flexran_slice_member(UE_id, slice_id) == 0)
continue;
// if there is no available harq_process, skip the UE
if (UE_list->UE_sched_ctrl[UE_id].harq_pid[CC_id]<0)
if (!phy_stats_exist(Mod_id, rnti))
continue;
for (ii=0; ii < UE_num_active_CC(UE_list,UE_id); ii++) {
@@ -629,6 +528,10 @@ void _dlsch_scheduler_pre_processor (module_id_t Mod_id,
ue_sched_ctl->max_allowed_rbs[CC_id]=nb_rbs_allowed_slice[CC_id][slice_id];
flexran_get_harq(Mod_id, CC_id, UE_id, frameP, subframeP, &harq_pid, &round);
// if there is no available harq_process, skip the UE
if (UE_list->UE_sched_ctrl[UE_id].harq_pid[CC_id]<0)
continue;
average_rbs_per_user[CC_id]=0;
frame_parms[CC_id] = mac_xface->get_lte_frame_parms(Mod_id,CC_id);
@@ -670,6 +573,15 @@ void _dlsch_scheduler_pre_processor (module_id_t Mod_id,
for(i=UE_list->head; i>=0; i=UE_list->next[i]) {
rnti = UE_RNTI(Mod_id,i);
if(rnti == NOT_A_RNTI)
continue;
if (UE_list->UE_sched_ctrl[i].ul_out_of_sync == 1)
continue;
if (!phy_stats_exist(Mod_id, rnti))
continue;
if (flexran_slice_member(i, slice_id) == 0)
continue;
@@ -734,9 +646,13 @@ void _dlsch_scheduler_pre_processor (module_id_t Mod_id,
// LOG_D(MAC,"UE %d rnti 0x\n", UE_id, rnti );
if(rnti == NOT_A_RNTI)
continue;
if (UE_list->UE_sched_ctrl[UE_id].ul_out_of_sync == 1)
continue;
if (!phy_stats_exist(Mod_id, rnti))
continue;
transmission_mode = mac_xface->get_transmission_mode(Mod_id,CC_id,rnti);
//rrc_status = mac_eNB_get_rrc_status(Mod_id,rnti);
/* 1st allocate for the retx */
@@ -1061,11 +977,15 @@ flexran_schedule_ue_spec_common(mid_t mod_id,
uint32_t ndi;
flexran_agent_mac_create_empty_dl_config(mod_id, dl_info);
#if 0
if (UE_list->head==-1) {
return;
}
#endif
start_meas(&eNB->schedule_dlsch);
VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_SCHEDULE_DLSCH,VCD_FUNCTION_IN);
@@ -1107,8 +1027,9 @@ flexran_schedule_ue_spec_common(mid_t mod_id,
if (mbsfn_flag[CC_id]>0)
continue;
for (UE_id=UE_list->head; UE_id>=0; UE_id=UE_list->next[UE_id]) {
rnti = flexran_get_ue_crnti(mod_id, UE_id);
ue_sched_ctl = &UE_list->UE_sched_ctrl[UE_id];
@@ -1121,12 +1042,6 @@ flexran_schedule_ue_spec_common(mid_t mod_id,
continue;
}
if (flexran_get_ue_crnti(mod_id, UE_id) == NOT_A_RNTI) {
LOG_D(MAC,"[eNB] Cannot find UE\n");
// mac_xface->macphy_exit("[MAC][eNB] Cannot find eNB_UE_stats\n");
continue;
}
if ((ue_sched_ctl->pre_nb_available_rbs[CC_id] == 0) ) {
continue;
}
openair2/LAYER2/MAC/pre_processor.c
+1 -1
View File
@@ -68,7 +68,7 @@
/* this function checks that get_eNB_UE_stats returns
* a non-NULL pointer for all CCs for a given UE
*/
static int phy_stats_exist(module_id_t Mod_id, int rnti)
int phy_stats_exist(module_id_t Mod_id, int rnti)
{
int CC_id;
for (CC_id = 0; CC_id < MAX_NUM_CCs; CC_id++)
openair2/LAYER2/MAC/proto.h
+1
View File
@@ -536,6 +536,7 @@ void store_ulsch_buffer(module_id_t module_idP, int frameP, sub_frame_t subframe
void sort_ue_ul (module_id_t module_idP,int frameP, sub_frame_t subframeP);
void assign_max_mcs_min_rb(module_id_t module_idP,int frameP, sub_frame_t subframeP,uint16_t *first_rb);
void adjust_bsr_info(int buffer_occupancy, uint16_t TBS, UE_TEMPLATE *UE_template);
int phy_stats_exist(module_id_t Mod_id, int rnti);
/*! \fn UE_L2_state_t ue_scheduler(const module_id_t module_idP,const frame_t frameP, const sub_frame_t subframe, const lte_subframe_t direction,const uint8_t eNB_index)
\brief UE scheduler where all the ue background tasks are done. This function performs the following: 1) Trigger PDCP every 5ms 2) Call RRC for link status return to PHY3) Perform SR/BSR procedures for scheduling feedback 4) Perform PHR procedures.