comm = mbox->put_init(msg, size)->set_rate(mig_speed)->wait_for(timeout);
else
comm = mbox->put_async(msg, size)->wait_for(timeout);
- } catch (xbt_ex& e) {
+ } catch (const xbt_ex&) {
if (comm) {
sg_size_t remaining = static_cast<sg_size_t>(comm->get_remaining());
XBT_VERB("timeout (%lf s) in sending_migration_data, remaining %llu bytes of %llu", timeout, remaining, size);
bool skip_stage2 = false;
size_t remaining_size = ramsize;
- size_t threshold = 0.0;
+
+ double clock_prev_send;
+ double clock_post_send;
+ double bandwidth;
+ size_t threshold;
/* check parameters */
if (ramsize == 0)
sg_vm_start_dirty_page_tracking(vm_);
double computed_during_stage1 = 0;
- double clock_prev_send = s4u::Engine::get_clock();
+ clock_prev_send = s4u::Engine::get_clock();
try {
/* At stage 1, we do not need timeout. We have to send all the memory pages even though the duration of this
} else if (sent > ramsize)
XBT_CRITICAL("bug");
- } catch (xbt_ex& e) {
+ } catch (const xbt_ex&) {
// hostfailure (if you want to know whether this is the SRC or the DST check directly in send_migration_data code)
// Stop the dirty page tracking an return (there is no memory space to release)
sg_vm_stop_dirty_page_tracking(vm_);
return;
}
- double clock_post_send = s4u::Engine::get_clock();
+ clock_post_send = s4u::Engine::get_clock();
mig_timeout -= (clock_post_send - clock_prev_send);
if (mig_timeout < 0) {
XBT_VERB("The duration of stage 1 exceeds the timeout value, skip stage 2");
}
/* estimate bandwidth */
- double bandwidth = ramsize / (clock_post_send - clock_prev_send);
- threshold = bandwidth * max_downtime;
+ bandwidth = ramsize / (clock_post_send - clock_prev_send);
+ threshold = bandwidth * max_downtime;
XBT_DEBUG("actual bandwidth %f (MB/s), threshold %zu", bandwidth / 1024 / 1024, threshold);
/* Stage2: send update pages iteratively until the size of remaining states becomes smaller than threshold value. */
if (not skip_stage2) {
int stage2_round = 0;
- for (;;) {
- sg_size_t updated_size = 0;
- if (stage2_round == 0) {
- /* just after stage1, nothing has been updated. But, we have to send the data updated during stage1 */
- updated_size = get_updated_size(computed_during_stage1, dp_rate, dp_cap);
- } else {
- double computed = sg_vm_lookup_computed_flops(vm_);
- updated_size = get_updated_size(computed, dp_rate, dp_cap);
- }
+ /* just after stage1, nothing has been updated. But, we have to send the data updated during stage1 */
+ sg_size_t updated_size = get_updated_size(computed_during_stage1, dp_rate, dp_cap);
+ remaining_size += updated_size;
+ XBT_DEBUG("mig-stage2.%d: remaining_size %zu (%s threshold %zu)", stage2_round, remaining_size,
+ (remaining_size < threshold) ? "<" : ">", threshold);
+
+ /* When the remaining size is below the threshold value, move to stage 3. */
+ while (threshold < remaining_size) {
XBT_DEBUG("mig-stage 2:%d updated_size %llu computed_during_stage1 %f dp_rate %f dp_cap %llu", stage2_round,
updated_size, computed_during_stage1, dp_rate, dp_cap);
- /* Check whether the remaining size is below the threshold value. If so, move to stage 3. */
- remaining_size += updated_size;
- XBT_DEBUG("mig-stage2.%d: remaining_size %zu (%s threshold %zu)", stage2_round, remaining_size,
- (remaining_size < threshold) ? "<" : ">", threshold);
- if (remaining_size < threshold)
- break;
-
- sg_size_t sent = 0;
- double clock_prev_send = s4u::Engine::get_clock();
+ sg_size_t sent = 0;
+ clock_prev_send = s4u::Engine::get_clock();
try {
XBT_DEBUG("Stage 2, gonna send %llu", updated_size);
sent = sendMigrationData(updated_size, 2, stage2_round, mig_speed, mig_timeout);
- } catch (xbt_ex& e) {
+ } catch (const xbt_ex&) {
// hostfailure (if you want to know whether this is the SRC or the DST check directly in send_migration_data
// code)
// Stop the dirty page tracking an return (there is no memory space to release)
sg_vm_stop_dirty_page_tracking(vm_);
return;
}
- double clock_post_send = s4u::Engine::get_clock();
+
+ remaining_size -= sent;
+ double computed = sg_vm_lookup_computed_flops(vm_);
+
+ clock_post_send = s4u::Engine::get_clock();
if (sent == updated_size) {
- /* timeout did not happen */
- double bandwidth = updated_size / (clock_post_send - clock_prev_send);
- threshold = bandwidth * max_downtime;
+ bandwidth = updated_size / (clock_post_send - clock_prev_send);
+ threshold = bandwidth * max_downtime;
XBT_DEBUG("actual bandwidth %f, threshold %zu", bandwidth / 1024 / 1024, threshold);
- remaining_size -= sent;
stage2_round += 1;
mig_timeout -= (clock_post_send - clock_prev_send);
xbt_assert(mig_timeout > 0);
-
- } else if (sent < updated_size) {
+ XBT_DEBUG("mig-stage2.%d: remaining_size %zu (%s threshold %zu)", stage2_round, remaining_size,
+ (remaining_size < threshold) ? "<" : ">", threshold);
+ updated_size = get_updated_size(computed, dp_rate, dp_cap);
+ remaining_size += updated_size;
+ } else {
/* When timeout happens, we move to stage 3. The size of memory pages
* updated before timeout must be added to the remaining size. */
XBT_VERB("mig-stage2.%d: timeout, force moving to stage 3. sent %llu / %llu, eta %lf", stage2_round, sent,
updated_size, (clock_post_send - clock_prev_send));
- remaining_size -= sent;
-
- double computed = sg_vm_lookup_computed_flops(vm_);
updated_size = get_updated_size(computed, dp_rate, dp_cap);
remaining_size += updated_size;
break;
- } else
- XBT_CRITICAL("bug");
+ }
}
}
try {
XBT_DEBUG("Stage 3: Gonna send %zu bytes", remaining_size);
sendMigrationData(remaining_size, 3, 0, mig_speed, -1);
- } catch (xbt_ex& e) {
+ } catch (const xbt_ex&) {
// hostfailure (if you want to know whether this is the SRC or the DST check directly in send_migration_data code)
// Stop the dirty page tracking an return (there is no memory space to release)
vm_->resume();
}
}
-static void onVirtualMachineShutdown(simgrid::s4u::VirtualMachine& vm)
+static void onVirtualMachineShutdown(simgrid::s4u::VirtualMachine const& vm)
{
if (vm.get_impl()->is_migrating_) {
vm.extension<simgrid::vm::VmMigrationExt>()->rx_->kill();
/* wait until the migration have finished or on error has occurred */
XBT_DEBUG("wait for reception of the final ACK (i.e. migration has been correctly performed");
- simgrid::s4u::MailboxPtr mbox_ctl = simgrid::s4u::Mailbox::by_name(
+ simgrid::s4u::Mailbox* mbox_ctl = simgrid::s4u::Mailbox::by_name(
std::string("__mbox_mig_ctl:") + vm->get_cname() + "(" + src_pm->get_cname() + "-" + dst_pm->get_cname() + ")");
delete static_cast<std::string*>(mbox_ctl->get());
tx->join();
