Commit 642d84d5 authored by Misbah Mubarak's avatar Misbah Mubarak
Browse files

Merging slimfly with codes master

parents db56bd38 a2f3680b
......@@ -22,6 +22,7 @@ extern "C" {
#include "model-net.h"
#include "model-net-sched.h"
#include "net/dragonfly.h"
#include "net/slimfly.h"
#include "net/loggp.h"
#include "net/simplenet-upd.h"
#include "net/simplep2p.h"
......@@ -120,12 +121,13 @@ typedef struct model_net_base_msg {
typedef struct model_net_wrap_msg {
msg_header h;
union {
model_net_base_msg m_base; // base lp
terminal_message m_dfly; // dragonfly
loggp_message m_loggp; // loggp
sn_message m_snet; // simplenet
sp_message m_sp2p; // simplep2p
nodes_message m_torus; // torus
model_net_base_msg m_base; // base lp
terminal_message m_dfly; // dragonfly
slim_terminal_message m_slim; // slimfly
loggp_message m_loggp; // loggp
sn_message m_snet; // simplenet
sp_message m_sp2p; // simplep2p
nodes_message m_torus; // torus
// add new ones here
} msg;
} model_net_wrap_msg;
......
......@@ -60,6 +60,7 @@ typedef struct mn_stats mn_stats;
X(SIMPLENET, "modelnet_simplenet", "simplenet", &simplenet_method)\
X(SIMPLEP2P, "modelnet_simplep2p", "simplep2p", &simplep2p_method)\
X(TORUS, "modelnet_torus", "torus", &torus_method)\
X(SLIMFLY, "modelnet_slimfly", "slimfly", &slimfly_method)\
X(DRAGONFLY, "modelnet_dragonfly", "dragonfly", &dragonfly_method)\
X(DRAGONFLY_ROUTER, "modelnet_dragonfly_router", "dragonfly_router", &dragonfly_router_method)\
X(LOGGP, "modelnet_loggp", "loggp", &loggp_method)\
......
/*
* Copyright (C) 2014 University of Chicago.
* See COPYRIGHT notice in top-level directory.
*
*/
#ifndef SLIMFLY_H
#define SLIMFLY_H
#include <ross.h>
typedef struct slim_terminal_message slim_terminal_message;
/* this message is used for both dragonfly compute nodes and routers */
struct slim_terminal_message
{
/* magic number */
int magic;
/* flit travel start time*/
tw_stime travel_start_time;
/* packet ID of the flit */
unsigned long long packet_ID;
/* event type of the flit */
short type;
/* category: comes from codes */
char category[CATEGORY_NAME_MAX];
/* final destination LP ID, this comes from codes can be a server or any other LP type*/
tw_lpid final_dest_gid;
/*sending LP ID from CODES, can be a server or any other LP type */
tw_lpid sender_lp;
tw_lpid sender_mn_lp; // source modelnet id
/* destination terminal ID of the dragonfly */
tw_lpid dest_terminal_id;
/* source terminal ID of the dragonfly */
unsigned int src_terminal_id;
/* local LP ID to calculate the radix of the sender node/router */
unsigned int local_id;
/* message originating router id */
unsigned int origin_router_id;
/* number of hops traversed by the packet */
short my_N_hop;
short my_l_hop, my_g_hop;
short saved_channel;
/* Intermediate LP ID from which this message is coming */
unsigned int intm_lp_id;
short new_vc;
short saved_vc;
/* last hop of the message, can be a terminal, local router or global router */
short last_hop;
/* For routing */
int intm_group_id;
int intm_router_id;
int chunk_id;
uint64_t packet_size;
uint64_t message_id;
uint64_t total_size;
int saved_remote_esize;
int remote_event_size_bytes;
int local_event_size_bytes;
// For buffer message
short vc_index;
int sender_radix;
int output_chan;
model_net_event_return event_rc;
int is_pull;
uint64_t pull_size;
/* for reverse computation */
short path_type;
tw_stime saved_available_time;
tw_stime saved_avg_time;
tw_stime saved_credit_time;
tw_stime saved_collective_init_time;
tw_stime saved_hist_start_time;
tw_stime msg_start_time;
short saved_completed_chunks;
int saved_hist_num;
int saved_occupancy;
/* for reverse computation of a node's fan in*/
int saved_fan_nodes;
tw_lpid sender_svr;
/* LP ID of the sending node, has to be a network node in the dragonfly */
tw_lpid sender_node;
tw_lpid next_stop;
struct sfly_qhash_entry * saved_hash;
};
#endif /* end of include guard: DRAGONFLY_H */
/*
* Local variables:
* c-indent-level: 4
* c-basic-offset: 4
* End:
*
* vim: ft=c ts=8 sts=4 sw=4 expandtab
*/
......@@ -80,6 +80,7 @@ nobase_include_HEADERS = \
codes/model-net-sched.h \
codes/model-net-inspect.h \
codes/net/dragonfly.h \
codes/net/slimfly.h \
codes/net/loggp.h \
codes/net/simplenet-upd.h \
codes/net/simplep2p.h \
......@@ -139,6 +140,7 @@ src_libcodes_a_SOURCES = \
src/networks/model-net/simplenet-upd.c \
src/networks/model-net/torus.c \
src/networks/model-net/dragonfly.c \
src/networks/model-net/slimfly.c \
src/networks/model-net/loggp.c \
src/networks/model-net/simplep2p.c \
src/networks/model-net/model-net-lp.c \
......@@ -148,6 +150,7 @@ src_libcodes_a_SOURCES = \
src/network-workloads/model-net-mpi-wrklds.c \
src/network-workloads/model-net-mpi-replay.c \
src/network-workloads/model-net-synthetic.c \
src/network-workloads/model-net-synthetic-slimfly.c \
src/network-workloads/model-net-dumpi-traces-dump.c
......@@ -169,6 +172,7 @@ bin_PROGRAMS += src/network-workloads/model-net-mpi-replay
bin_PROGRAMS += src/network-workloads/model-net-mpi-wrklds
bin_PROGRAMS += src/network-workloads/model-net-dumpi-traces-dump
bin_PROGRAMS += src/network-workloads/model-net-synthetic
bin_PROGRAMS += src/network-workloads/model-net-synthetic-slimfly
src_workload_codes_workload_dump_SOURCES = \
......@@ -178,6 +182,7 @@ src_network_workloads_model_net_mpi_replay_SOURCES = src/network-workloads/model
src_network_workloads_model_net_mpi_wrklds_SOURCES = src/network-workloads/model-net-mpi-wrklds.c
src_network_workloads_model_net_synthetic_SOURCES = src/network-workloads/model-net-synthetic.c
src_network_workloads_model_net_synthetic_slimfly_SOURCES = src/network-workloads/model-net-synthetic-slimfly.c
src_network_workloads_model_net_dumpi_traces_dump_SOURCES = src/network-workloads/model-net-dumpi-traces-dump.c
#bin_PROGRAMS += src/network-workload/codes-nw-test
......
LPGROUPS
{
MODELNET_GRP
{
repetitions="50";
server="3";
modelnet_slimfly="3";
slimfly_router="1";
}
}
PARAMS
{
packet_size="256";
modelnet_order=( "slimfly" );
# scheduler options
modelnet_scheduler="fcfs";
chunk_size="256";
# modelnet_scheduler="round-robin";
num_vcs="4";
num_routers="5";
num_terminals="3";
global_channels="5";
local_channels="2";
generator_set_X=("1","4"); # : Subgraph 0 generator set
generator_set_X_prime=("2","3"); # : Subgraph 1 generator set
local_vc_size="25600";
global_vc_size="25600";
cn_vc_size="25600";
local_bandwidth="9.0";
global_bandwidth="9.0";
cn_bandwidth="9.0";
router_delay="0";
link_delay="0";
message_size="512";
routing="minimal";
}
/*
* Copyright (C) 2015 University of Chicago.
* See COPYRIGHT notice in top-level directory.
*
*/
/*
* The test program generates some synthetic traffic patterns for the model-net network models.
* currently it only support the dragonfly network model uniform random and nearest neighbor traffic patterns.
*/
#include "codes/model-net.h"
#include "codes/lp-io.h"
#include "codes/codes.h"
#include "codes/codes_mapping.h"
#include "codes/configuration.h"
#include "codes/lp-type-lookup.h"
#define PAYLOAD_SZ 256
#define LP_CONFIG_NM (model_net_lp_config_names[SLIMFLY])
#define PRINT_WORST_CASE_MATCH 0
#define PARAMS_LOG 1
FILE * slimfly_results_log_2=NULL;
FILE * slimfly_ross_csv_log=NULL;
static int net_id = 0;
static int num_routers = 0;
static int num_servers = 0;
static int offset = 2;
static int traffic = 1;
static double arrival_time = 1000.0;
static double load = 0.0; //percent utilization of each terminal's uplink when sending messages
static double MEAN_INTERVAL = 0.0;
int this_packet_size = 0;
double this_global_bandwidth = 0.0;
double this_link_delay = 0;
int *worst_dest; //Array mapping worst case destination for each router
int num_terminals;
int total_routers;
/* whether to pull instead of push */
static int do_pull = 0;
static int num_servers_per_rep = 0;
static int num_routers_per_grp = 0;
static int num_nodes_per_grp = 0;
static int num_reps = 0;
static int num_groups = 0;
static int num_nodes = 0;
typedef struct svr_msg svr_msg;
typedef struct svr_state svr_state;
/* global variables for codes mapping */
static char group_name[MAX_NAME_LENGTH];
static char lp_type_name[MAX_NAME_LENGTH];
static int group_index, lp_type_index, rep_id, offset;
/* type of events */
enum svr_event
{
KICKOFF, /* kickoff event */
REMOTE, /* remote event */
LOCAL /* local event */
};
/* type of synthetic traffic */
enum TRAFFIC
{
UNIFORM = 1, /* sends message to a randomly selected node */
WORST_CASE = 2,
NEAREST_GROUP = 3, /* sends message to the node connected to the neighboring router */
NEAREST_NEIGHBOR = 4 /* sends message to the next node (potentially connected to the same router) */
};
struct svr_state
{
int msg_sent_count; /* requests sent */
int msg_recvd_count; /* requests recvd */
int local_recvd_count; /* number of local messages received */
tw_stime start_ts; /* time that we started sending requests */
tw_stime end_ts; /* time that we ended sending requests */
};
struct svr_msg
{
enum svr_event svr_event_type;
tw_lpid src; /* source of this request or ack */
int incremented_flag; /* helper for reverse computation */
};
static void svr_init(
svr_state * ns,
tw_lp * lp);
static void svr_event(
svr_state * ns,
tw_bf * b,
svr_msg * m,
tw_lp * lp);
static void svr_rev_event(
svr_state * ns,
tw_bf * b,
svr_msg * m,
tw_lp * lp);
static void svr_finalize(
svr_state * ns,
tw_lp * lp);
tw_lptype svr_lp = {
(init_f) svr_init,
(pre_run_f) NULL,
(event_f) svr_event,
(revent_f) svr_rev_event,
(final_f) svr_finalize,
(map_f) codes_mapping,
sizeof(svr_state),
};
const tw_optdef app_opt [] =
{
TWOPT_GROUP("Model net synthetic traffic " ),
TWOPT_UINT("traffic", traffic, "UNIFORM RANDOM=1, NEAREST NEIGHBOR=2 "),
TWOPT_STIME("arrival_time", arrival_time, "INTER-ARRIVAL TIME"),
TWOPT_STIME("load", load, "percentage of packet inter-arrival rate to simulate"),
TWOPT_END()
};
const tw_lptype* svr_get_lp_type()
{
return(&svr_lp);
}
static void svr_add_lp_type()
{
lp_type_register("server", svr_get_lp_type());
}
/* convert GiB/s and bytes to ns */
static tw_stime bytes_to_ns(uint64_t bytes, double GB_p_s)
{
tw_stime time;
/* bytes to GB */
time = ((double)bytes)/(1024.0*1024.0*1024.0);
/* MB to s */
time = time / GB_p_s;
/* s to ns */
time = time * 1000.0 * 1000.0 * 1000.0;
return(time);
}
/** Latest implementation of function to return an array mapping each router with it's corresponding worst-case router pair
* @param[in/out] *worst_dest Router ID to send all messages to
*/
void init_worst_case_mapping(int *worst_dest)
{
int i,j,k;
int r1,r2; //Routers to be paired
for(k=0;k<2;k++)
{
for(j=0;j<num_routers_per_grp-1;j+=2)
{
for(i=0;i<num_routers_per_grp;i++)
{
r1 = i + j*num_routers_per_grp + k*num_routers_per_grp*num_routers_per_grp;
r2 = i + (j+1)*num_routers_per_grp + k*num_routers_per_grp*num_routers_per_grp;
worst_dest[r1] = r2;
worst_dest[r2] = r1;
// printf("%d->%d, %d->%d\n",r1,worst_dest[r1],r2,worst_dest[r2]);
}
}
}
j = num_routers_per_grp-1;
for(i=0;i<num_routers_per_grp;i++)
{
r1 = i + j*num_routers_per_grp + 0*num_routers_per_grp*num_routers_per_grp;
r2 = i + j*num_routers_per_grp + 1*num_routers_per_grp*num_routers_per_grp;
worst_dest[r1] = r2;
worst_dest[r2] = r1;
// printf("%d->%d, %d->%d\n",r1,worst_dest[r1],r2,worst_dest[r2]);
}
}
static void issue_event(
svr_state * ns,
tw_lp * lp)
{
tw_event *e;
svr_msg *m;
tw_stime kickoff_time;
/* each server sends a dummy event to itself that will kick off the real
* simulation
*/
const char * anno;
configuration_get_value_int(&config, "PARAMS", "packet_size", NULL, &this_packet_size);
if(!this_packet_size) {
this_packet_size = 0;
fprintf(stderr, "Packet size not specified, setting to %d\n", this_packet_size);
exit(0);
}
configuration_get_value_double(&config, "PARAMS", "global_bandwidth", NULL, &this_global_bandwidth);
if(!this_global_bandwidth) {
this_global_bandwidth = 4.7;
fprintf(stderr, "Bandwidth of global channels not specified, setting to %lf\n", this_global_bandwidth);
}
/* configuration_get_value_double(&config, "PARAMS", "link_delay", anno, &this_link_delay);
if(!this_link_delay) {
this_link_delay = 0;
fprintf(stderr, "Link Delay not specified, setting to %lf\n", this_link_delay);
}
*/
if(arrival_time!=0)
{
MEAN_INTERVAL = arrival_time;
}
if(load != 0)
{
MEAN_INTERVAL = bytes_to_ns(this_packet_size, load*this_global_bandwidth) + this_link_delay;
}
/* skew each kickoff event slightly to help avoid event ties later on */
kickoff_time = g_tw_lookahead + MEAN_INTERVAL + tw_rand_exponential(lp->rng, (double)MEAN_INTERVAL/100);
e = tw_event_new(lp->gid, kickoff_time, lp);
m = tw_event_data(e);
m->svr_event_type = KICKOFF;
tw_event_send(e);
}
static void svr_init(
svr_state * ns,
tw_lp * lp)
{
issue_event(ns, lp);
return;
}
static void handle_kickoff_rev_event(
svr_state * ns,
tw_bf * b,
svr_msg * m,
tw_lp * lp)
{
ns->msg_sent_count--;
model_net_event_rc(net_id, lp, PAYLOAD_SZ);
}
static void handle_kickoff_event(
svr_state * ns,
tw_bf * b,
svr_msg * m,
tw_lp * lp)
{
char anno[MAX_NAME_LENGTH];
tw_lpid local_dest = -1, global_dest = -1;
svr_msg * m_local = malloc(sizeof(svr_msg));
svr_msg * m_remote = malloc(sizeof(svr_msg));
m_local->svr_event_type = LOCAL;
m_local->src = lp->gid;
memcpy(m_remote, m_local, sizeof(svr_msg));
m_remote->svr_event_type = REMOTE;
assert(net_id == SLIMFLY); /* only supported for dragonfly model right now. */
ns->start_ts = tw_now(lp);
// codes_mapping_get_lp_info(lp->gid, group_name, &group_index, lp_type_name, &lp_type_index, anno, &rep_id, &offset);
codes_mapping_get_lp_info(lp->gid, group_name, &group_index, lp_type_name, &lp_type_index, anno, &rep_id, &offset);
/* in case of uniform random traffic, send to a random destination. */
if(traffic == UNIFORM)
{
local_dest = tw_rand_integer(lp->rng, 0, num_nodes - 1);
// printf("\n LP %ld sending to %d ", lp->gid, local_dest);
}
else if(traffic == WORST_CASE)
{
// Assign the global router ID
// TODO: be annotation-aware
int num_lps = codes_mapping_get_lp_count(group_name, 1, LP_CONFIG_NM, anno, 0);
int src_terminal_id = (rep_id * num_lps) + offset;
// s->router_id=(int)s->terminal_id / (s->params->num_routers/2);
int src_router_id = src_terminal_id / (num_lps);
int dst_router_id = worst_dest[src_router_id];
// dst_lp = tw_rand_integer(lp->rng, total_routers + num_terminals*dst_router_id, total_routers + num_terminals*(dst_router_id+1)-1);
// local_dest = total_routers + num_terminals*dst_router_id + (src_terminal_id % num_terminals);
local_dest = num_lps*dst_router_id + (src_terminal_id % num_terminals);
// printf("packet path initialized. src_term:%d src_router:%d dest_router:%d dest_term:%d\n",src_terminal_id, src_router_id, dst_router_id, (int)local_dest);
}
else if(traffic == NEAREST_GROUP)
{
local_dest = (rep_id * 2 + offset + num_nodes_per_grp) % num_nodes;
// printf("\n LP %ld sending to %ld num nodes %d ", rep_id * 2 + offset, local_dest, num_nodes);
}
else if(traffic == NEAREST_NEIGHBOR)
{
local_dest = (rep_id * 2 + offset + 2) % num_nodes;
// printf("\n LP %ld sending to %ld num nodes %d ", rep_id * 2 + offset, local_dest, num_nodes);
}
//printf("1local_dest:%d global_dest:%d num_nodes:%d\n", local_dest, global_dest, num_nodes);
assert(local_dest < num_nodes);
codes_mapping_get_lp_id(group_name, lp_type_name, anno, 1, local_dest / num_servers_per_rep, local_dest % num_servers_per_rep, &global_dest);
//printf("2local_dest:%d global_dest:%d num_nodes:%d\n", local_dest, global_dest, num_nodes);
ns->msg_sent_count++;
model_net_event(net_id, "test", global_dest, PAYLOAD_SZ, 0.0, sizeof(svr_msg), (const void*)m_remote, sizeof(svr_msg), (const void*)m_local, lp);
issue_event(ns, lp);
return;
}
static void handle_remote_rev_event(
svr_state * ns,
tw_bf * b,
svr_msg * m,
tw_lp * lp)
{
ns->msg_recvd_count--;
}
static void handle_remote_event(
svr_state * ns,
tw_bf * b,
svr_msg * m,
tw_lp * lp)
{
ns->msg_recvd_count++;
}
static void handle_local_rev_event(
svr_state * ns,
tw_bf * b,
svr_msg * m,
tw_lp * lp)
{
ns->local_recvd_count--;
}
static void handle_local_event(
svr_state * ns,
tw_bf * b,
svr_msg * m,
tw_lp * lp)
{
ns->local_recvd_count++;
}
/* convert ns to seconds */
static tw_stime ns_to_s(tw_stime ns)
{
return(ns / (1000.0 * 1000.0 * 1000.0));
}
/* convert seconds to ns */
static tw_stime s_to_ns(tw_stime ns)
{
return(ns * (1000.0 * 1000.0 * 1000.0));
}
int index_mine = 0;
static void svr_finalize(
svr_state * ns,
tw_lp * lp)
{
ns->end_ts = tw_now(lp);
// printf("server %d server %llu recvd %d bytes in %f seconds, %f MiB/s sent_count %d recvd_count %d local_count %d \n", index_mine++,(unsigned long long)lp->gid, PAYLOAD_SZ*ns->msg_recvd_count, ns_to_s(ns->end_ts-ns->start_ts),
// ((double)(PAYLOAD_SZ*ns->msg_sent_count)/(double)(1024*1024)/ns_to_s(ns->end_ts-ns->start_ts)), ns->msg_sent_count, ns->msg_recvd_count, ns->local_recvd_count);
return;
}
static void svr_rev_event(
svr_state * ns,
tw_bf * b,
svr_msg * m,
tw_lp * lp)
{
switch (m->svr_event_type)
{
case REMOTE:
handle_remote_rev_event(ns, b, m, lp);
break;
case LOCAL:
handle_local_rev_event(ns, b, m, lp);
break;