1
| |
2/* 3 * Copyright (c) 1999-2008 Mark D. Hill and David A. Wood 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions are 8 * met: redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer; 10 * redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution; 13 * neither the name of the copyright holders nor the names of its 14 * contributors may be used to endorse or promote products derived from 15 * this software without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 18 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 19 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 20 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 21 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 22 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 23 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 27 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29
| 1/* 2 * Copyright (c) 1999-2008 Mark D. Hill and David A. Wood 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions are 7 * met: redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer; 9 * redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution; 12 * neither the name of the copyright holders nor the names of its 13 * contributors may be used to endorse or promote products derived from 14 * this software without specific prior written permission. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28
|
30/* 31 * $Id$ 32 * 33 * Description: see Throttle.hh 34 * 35 */ 36 37#include "mem/ruby/network/simple/Throttle.hh"
| 29#include "base/cprintf.hh" 30#include "mem/protocol/Protocol.hh"
|
38#include "mem/ruby/buffers/MessageBuffer.hh" 39#include "mem/ruby/network/Network.hh"
| 31#include "mem/ruby/buffers/MessageBuffer.hh" 32#include "mem/ruby/network/Network.hh"
|
40#include "mem/ruby/system/System.hh"
| 33#include "mem/ruby/network/simple/Throttle.hh"
|
41#include "mem/ruby/slicc_interface/NetworkMessage.hh"
| 34#include "mem/ruby/slicc_interface/NetworkMessage.hh"
|
42#include "mem/protocol/Protocol.hh"
| 35#include "mem/ruby/system/System.hh"
|
43 44const int HIGH_RANGE = 256; 45const int ADJUST_INTERVAL = 50000; 46const int MESSAGE_SIZE_MULTIPLIER = 1000; 47//const int BROADCAST_SCALING = 4; // Have a 16p system act like a 64p systems 48const int BROADCAST_SCALING = 1; 49const int PRIORITY_SWITCH_LIMIT = 128; 50 51static int network_message_to_size(NetworkMessage* net_msg_ptr); 52
| 36 37const int HIGH_RANGE = 256; 38const int ADJUST_INTERVAL = 50000; 39const int MESSAGE_SIZE_MULTIPLIER = 1000; 40//const int BROADCAST_SCALING = 4; // Have a 16p system act like a 64p systems 41const int BROADCAST_SCALING = 1; 42const int PRIORITY_SWITCH_LIMIT = 128; 43 44static int network_message_to_size(NetworkMessage* net_msg_ptr); 45
|
53extern std::ostream * debug_cout_ptr;
| 46extern std::ostream *debug_cout_ptr;
|
54
| 47
|
55Throttle::Throttle(int sID, NodeID node, int link_latency, int link_bandwidth_multiplier)
| 48Throttle::Throttle(int sID, NodeID node, int link_latency, 49 int link_bandwidth_multiplier)
|
56{
| 50{
|
57 init(node, link_latency, link_bandwidth_multiplier); 58 m_sID = sID;
| 51 init(node, link_latency, link_bandwidth_multiplier); 52 m_sID = sID;
|
59} 60
| 53} 54
|
61Throttle::Throttle(NodeID node, int link_latency, int link_bandwidth_multiplier)
| 55Throttle::Throttle(NodeID node, int link_latency, 56 int link_bandwidth_multiplier)
|
62{
| 57{
|
63 init(node, link_latency, link_bandwidth_multiplier); 64 m_sID = 0;
| 58 init(node, link_latency, link_bandwidth_multiplier); 59 m_sID = 0;
|
65} 66
| 60} 61
|
67void Throttle::init(NodeID node, int link_latency, int link_bandwidth_multiplier)
| 62void 63Throttle::init(NodeID node, int link_latency, int link_bandwidth_multiplier)
|
68{
| 64{
|
69 m_node = node; 70 m_vnets = 0;
| 65 m_node = node; 66 m_vnets = 0;
|
71
| 67
|
72 ASSERT(link_bandwidth_multiplier > 0); 73 m_link_bandwidth_multiplier = link_bandwidth_multiplier; 74 m_link_latency = link_latency;
| 68 ASSERT(link_bandwidth_multiplier > 0); 69 m_link_bandwidth_multiplier = link_bandwidth_multiplier; 70 m_link_latency = link_latency;
|
75
| 71
|
76 m_wakeups_wo_switch = 0; 77 clearStats();
| 72 m_wakeups_wo_switch = 0; 73 clearStats();
|
78} 79
| 74} 75
|
80void Throttle::clear()
| 76void 77Throttle::clear()
|
81{
| 78{
|
82 for (int counter = 0; counter < m_vnets; counter++) { 83 m_in[counter]->clear(); 84 m_out[counter]->clear(); 85 }
| 79 for (int counter = 0; counter < m_vnets; counter++) { 80 m_in[counter]->clear(); 81 m_out[counter]->clear(); 82 }
|
86} 87
| 83} 84
|
88void Throttle::addLinks(const Vector<MessageBuffer*>& in_vec, const Vector<MessageBuffer*>& out_vec)
| 85void 86Throttle::addLinks(const Vector<MessageBuffer*>& in_vec, 87 const Vector<MessageBuffer*>& out_vec)
|
89{
| 88{
|
90 assert(in_vec.size() == out_vec.size()); 91 for (int i=0; i92 addVirtualNetwork(in_vec[i], out_vec[i]); 93 }
| 89 assert(in_vec.size() == out_vec.size()); 90 for (int i=0; i<in_vec.size(); i++) { 91 addVirtualNetwork(in_vec[i], out_vec[i]); 92 }
|
94
| 93
|
95 m_message_counters.setSize(MessageSizeType_NUM); 96 for (int i=0; i<MessageSizeType_NUM; i++) { 97 m_message_counters[i].setSize(in_vec.size()); 98 for (int j=0; j<m_message_counters[i].size(); j++) { 99 m_message_counters[i][j] = 0;
| 94 m_message_counters.setSize(MessageSizeType_NUM); 95 for (int i = 0; i < MessageSizeType_NUM; i++) { 96 m_message_counters[i].setSize(in_vec.size()); 97 for (int j = 0; j<m_message_counters[i].size(); j++) { 98 m_message_counters[i][j] = 0; 99 }
|
100 }
| 100 }
|
101 }
| |
102} 103
| 101} 102
|
104void Throttle::addVirtualNetwork(MessageBuffer* in_ptr, MessageBuffer* out_ptr)
| 103void 104Throttle::addVirtualNetwork(MessageBuffer* in_ptr, MessageBuffer* out_ptr)
|
105{
| 105{
|
106 m_units_remaining.insertAtBottom(0); 107 m_in.insertAtBottom(in_ptr); 108 m_out.insertAtBottom(out_ptr);
| 106 m_units_remaining.insertAtBottom(0); 107 m_in.insertAtBottom(in_ptr); 108 m_out.insertAtBottom(out_ptr);
|
109
| 109
|
110 // Set consumer and description 111 m_in[m_vnets]->setConsumer(this); 112 string desc = "[Queue to Throttle " + NodeIDToString(m_sID) + " " + NodeIDToString(m_node) + "]"; 113 m_in[m_vnets]->setDescription(desc); 114 m_vnets++;
| 110 // Set consumer and description 111 m_in[m_vnets]->setConsumer(this); 112 string desc = "[Queue to Throttle " + NodeIDToString(m_sID) + " " + 113 NodeIDToString(m_node) + "]"; 114 m_in[m_vnets]->setDescription(desc); 115 m_vnets++;
|
115} 116
| 116} 117
|
117void Throttle::wakeup()
| 118void 119Throttle::wakeup()
|
118{
| 120{
|
119 // Limits the number of message sent to a limited number of bytes/cycle. 120 assert(getLinkBandwidth() > 0); 121 int bw_remaining = getLinkBandwidth();
| 121 // Limits the number of message sent to a limited number of bytes/cycle. 122 assert(getLinkBandwidth() > 0); 123 int bw_remaining = getLinkBandwidth();
|
122
| 124
|
123 // Give the highest numbered link priority most of the time 124 m_wakeups_wo_switch++; 125 int highest_prio_vnet = m_vnets-1; 126 int lowest_prio_vnet = 0; 127 int counter = 1; 128 bool schedule_wakeup = false;
| 125 // Give the highest numbered link priority most of the time 126 m_wakeups_wo_switch++; 127 int highest_prio_vnet = m_vnets-1; 128 int lowest_prio_vnet = 0; 129 int counter = 1; 130 bool schedule_wakeup = false;
|
129
| 131
|
130 // invert priorities to avoid starvation seen in the component network 131 if (m_wakeups_wo_switch > PRIORITY_SWITCH_LIMIT) { 132 m_wakeups_wo_switch = 0; 133 highest_prio_vnet = 0; 134 lowest_prio_vnet = m_vnets-1; 135 counter = -1; 136 }
| 132 // invert priorities to avoid starvation seen in the component network 133 if (m_wakeups_wo_switch > PRIORITY_SWITCH_LIMIT) { 134 m_wakeups_wo_switch = 0; 135 highest_prio_vnet = 0; 136 lowest_prio_vnet = m_vnets-1; 137 counter = -1; 138 }
|
137
| 139
|
138 for (int vnet = highest_prio_vnet; (vnet*counter) >= (counter*lowest_prio_vnet); vnet -= counter) {
| 140 for (int vnet = highest_prio_vnet; 141 (vnet * counter) >= (counter * lowest_prio_vnet); 142 vnet -= counter) {
|
139
| 143
|
140 assert(m_out[vnet] != NULL); 141 assert(m_in[vnet] != NULL); 142 assert(m_units_remaining[vnet] >= 0);
| 144 assert(m_out[vnet] != NULL); 145 assert(m_in[vnet] != NULL); 146 assert(m_units_remaining[vnet] >= 0);
|
143
| 147
|
144 while ((bw_remaining > 0) && ((m_in[vnet]->isReady()) || (m_units_remaining[vnet] > 0)) && m_out[vnet]->areNSlotsAvailable(1)) {
| 148 while (bw_remaining > 0 && 149 (m_in[vnet]->isReady() || m_units_remaining[vnet] > 0) && 150 m_out[vnet]->areNSlotsAvailable(1)) {
|
145
| 151
|
146 // See if we are done transferring the previous message on this virtual network 147 if (m_units_remaining[vnet] == 0 && m_in[vnet]->isReady()) {
| 152 // See if we are done transferring the previous message on 153 // this virtual network 154 if (m_units_remaining[vnet] == 0 && m_in[vnet]->isReady()) { 155 // Find the size of the message we are moving 156 MsgPtr msg_ptr = m_in[vnet]->peekMsgPtr(); 157 NetworkMessage* net_msg_ptr = 158 safe_cast<NetworkMessage*>(msg_ptr.ref()); 159 m_units_remaining[vnet] += 160 network_message_to_size(net_msg_ptr);
|
148
| 161
|
149 // Find the size of the message we are moving 150 MsgPtr msg_ptr = m_in[vnet]->peekMsgPtr(); 151 NetworkMessage* net_msg_ptr = dynamic_cast<NetworkMessage*>(msg_ptr.ref()); 152 m_units_remaining[vnet] += network_message_to_size(net_msg_ptr);
| 162 DEBUG_NEWLINE(NETWORK_COMP,HighPrio); 163 DEBUG_MSG(NETWORK_COMP, HighPrio, 164 csprintf("throttle: %d my bw %d bw spent enqueueing " 165 "net msg %d time: %d.", 166 m_node, getLinkBandwidth(), m_units_remaining[vnet], 167 g_eventQueue_ptr->getTime()));
|
153
| 168
|
154 DEBUG_NEWLINE(NETWORK_COMP,HighPrio); 155 DEBUG_MSG(NETWORK_COMP,HighPrio,"throttle: " + int_to_string(m_node) 156 + " my bw " + int_to_string(getLinkBandwidth()) 157 + " bw spent enqueueing net msg " + int_to_string(m_units_remaining[vnet]) 158 + " time: " + int_to_string(g_eventQueue_ptr->getTime()) + ".");
| 169 // Move the message 170 m_out[vnet]->enqueue(m_in[vnet]->peekMsgPtr(), m_link_latency); 171 m_in[vnet]->pop();
|
159
| 172
|
160 // Move the message 161 m_out[vnet]->enqueue(m_in[vnet]->peekMsgPtr(), m_link_latency); 162 m_in[vnet]->pop();
| 173 // Count the message 174 m_message_counters[net_msg_ptr->getMessageSize()][vnet]++;
|
163
| 175
|
164 // Count the message 165 m_message_counters[net_msg_ptr->getMessageSize()][vnet]++;
| 176 DEBUG_MSG(NETWORK_COMP,LowPrio,*m_out[vnet]); 177 DEBUG_NEWLINE(NETWORK_COMP,HighPrio); 178 }
|
166
| 179
|
167 DEBUG_MSG(NETWORK_COMP,LowPrio,*m_out[vnet]); 168 DEBUG_NEWLINE(NETWORK_COMP,HighPrio); 169 }
| 180 // Calculate the amount of bandwidth we spent on this message 181 int diff = m_units_remaining[vnet] - bw_remaining; 182 m_units_remaining[vnet] = max(0, diff); 183 bw_remaining = max(0, -diff); 184 }
|
170
| 185
|
171 // Calculate the amount of bandwidth we spent on this message 172 int diff = m_units_remaining[vnet] - bw_remaining; 173 m_units_remaining[vnet] = max(0, diff); 174 bw_remaining = max(0, -diff);
| 186 if (bw_remaining > 0 && 187 (m_in[vnet]->isReady() || m_units_remaining[vnet] > 0) && 188 !m_out[vnet]->areNSlotsAvailable(1)) { 189 DEBUG_MSG(NETWORK_COMP,LowPrio,vnet); 190 // schedule me to wakeup again because I'm waiting for my 191 // output queue to become available 192 schedule_wakeup = true; 193 }
|
175 } 176
| 194 } 195
|
177 if ((bw_remaining > 0) && ((m_in[vnet]->isReady()) || (m_units_remaining[vnet] > 0)) && !m_out[vnet]->areNSlotsAvailable(1)) { 178 DEBUG_MSG(NETWORK_COMP,LowPrio,vnet); 179 schedule_wakeup = true; // schedule me to wakeup again because I'm waiting for my output queue to become available 180 } 181 }
| 196 // We should only wake up when we use the bandwidth 197 // This is only mostly true 198 // assert(bw_remaining != getLinkBandwidth());
|
182
| 199
|
183 // We should only wake up when we use the bandwidth 184 // assert(bw_remaining != getLinkBandwidth()); // This is only mostly true
| 200 // Record that we used some or all of the link bandwidth this cycle 201 double ratio = 1.0 - (double(bw_remaining) / double(getLinkBandwidth()));
|
185
| 202
|
186 // Record that we used some or all of the link bandwidth this cycle 187 double ratio = 1.0-(double(bw_remaining)/double(getLinkBandwidth())); 188 // If ratio = 0, we used no bandwidth, if ratio = 1, we used all 189 linkUtilized(ratio);
| 203 // If ratio = 0, we used no bandwidth, if ratio = 1, we used all 204 linkUtilized(ratio);
|
190
| 205
|
191 if ((bw_remaining > 0) && !schedule_wakeup) { 192 // We have extra bandwidth and our output buffer was available, so we must not have anything else to do until another message arrives. 193 DEBUG_MSG(NETWORK_COMP,LowPrio,*this); 194 DEBUG_MSG(NETWORK_COMP,LowPrio,"not scheduled again"); 195 } else { 196 DEBUG_MSG(NETWORK_COMP,LowPrio,*this); 197 DEBUG_MSG(NETWORK_COMP,LowPrio,"scheduled again"); 198 // We are out of bandwidth for this cycle, so wakeup next cycle and continue 199 g_eventQueue_ptr->scheduleEvent(this, 1); 200 }
| 206 if (bw_remaining > 0 && !schedule_wakeup) { 207 // We have extra bandwidth and our output buffer was 208 // available, so we must not have anything else to do until 209 // another message arrives. 210 DEBUG_MSG(NETWORK_COMP, LowPrio, *this); 211 DEBUG_MSG(NETWORK_COMP, LowPrio, "not scheduled again"); 212 } else { 213 DEBUG_MSG(NETWORK_COMP, LowPrio, *this); 214 DEBUG_MSG(NETWORK_COMP, LowPrio, "scheduled again"); 215 216 // We are out of bandwidth for this cycle, so wakeup next 217 // cycle and continue 218 g_eventQueue_ptr->scheduleEvent(this, 1); 219 }
|
201} 202
| 220} 221
|
203void Throttle::printStats(ostream& out) const
| 222void 223Throttle::printStats(ostream& out) const
|
204{
| 224{
|
205 out << "utilized_percent: " << getUtilization() << endl;
| 225 out << "utilized_percent: " << getUtilization() << endl;
|
206} 207
| 226} 227
|
208void Throttle::clearStats()
| 228void 229Throttle::clearStats()
|
209{
| 230{
|
210 m_ruby_start = g_eventQueue_ptr->getTime(); 211 m_links_utilized = 0.0;
| 231 m_ruby_start = g_eventQueue_ptr->getTime(); 232 m_links_utilized = 0.0;
|
212
| 233
|
213 for (int i=0; i<m_message_counters.size(); i++) { 214 for (int j=0; j<m_message_counters[i].size(); j++) { 215 m_message_counters[i][j] = 0;
| 234 for (int i = 0; i < m_message_counters.size(); i++) { 235 for (int j = 0; j < m_message_counters[i].size(); j++) { 236 m_message_counters[i][j] = 0; 237 }
|
216 }
| 238 }
|
217 }
| |
218} 219
| 239} 240
|
220void Throttle::printConfig(ostream& out) const
| 241void 242Throttle::printConfig(ostream& out) const
|
221{
| 243{
|
222
| |
223} 224
| 244} 245
|
225double Throttle::getUtilization() const
| 246double 247Throttle::getUtilization() const
|
226{
| 248{
|
227 return (100.0 * double(m_links_utilized)) / (double(g_eventQueue_ptr->getTime()-m_ruby_start));
| 249 return 100.0 * double(m_links_utilized) / 250 double(g_eventQueue_ptr->getTime()-m_ruby_start);
|
228} 229
| 251} 252
|
230void Throttle::print(ostream& out) const
| 253void 254Throttle::print(ostream& out) const
|
231{
| 255{
|
232 out << "[Throttle: " << m_sID << " " << m_node << " bw: " << getLinkBandwidth() << "]";
| 256 out << "[Throttle: " << m_sID << " " << m_node 257 << " bw: " << getLinkBandwidth() << "]";
|
233} 234
| 258} 259
|
235// Helper function 236 237static 238int network_message_to_size(NetworkMessage* net_msg_ptr)
| 260int 261network_message_to_size(NetworkMessage* net_msg_ptr)
|
239{
| 262{
|
240 assert(net_msg_ptr != NULL);
| 263 assert(net_msg_ptr != NULL);
|
241
| 264
|
242 // Artificially increase the size of broadcast messages 243 if (BROADCAST_SCALING > 1) { 244 if (net_msg_ptr->getDestination().isBroadcast()) { 245 return (RubySystem::getNetwork()->MessageSizeType_to_int(net_msg_ptr->getMessageSize()) * MESSAGE_SIZE_MULTIPLIER * BROADCAST_SCALING); 246 } 247 } 248 return (RubySystem::getNetwork()->MessageSizeType_to_int(net_msg_ptr->getMessageSize()) * MESSAGE_SIZE_MULTIPLIER);
| 265 int size = RubySystem::getNetwork()-> 266 MessageSizeType_to_int(net_msg_ptr->getMessageSize()); 267 size *= MESSAGE_SIZE_MULTIPLIER; 268 269 // Artificially increase the size of broadcast messages 270 if (BROADCAST_SCALING > 1 && net_msg_ptr->getDestination().isBroadcast()) 271 size *= BROADCAST_SCALING; 272 273 return size;
|
249}
| 274}
|