more cleanup and some CR feedback fixes

This commit is contained in:
Brad Hefta-Gaub 2017-02-23 13:09:39 -08:00
parent f61e16fccf
commit 8cb8d686ec

View file

@ -80,13 +80,6 @@ int AvatarMixerSlave::sendIdentityPacket(const AvatarMixerClientData* nodeData,
static const int AVATAR_MIXER_BROADCAST_FRAMES_PER_SECOND = 45;
// only send extra avatar data (avatars out of view, ignored) every Nth AvatarData frame
// Extra avatar data will be sent (AVATAR_MIXER_BROADCAST_FRAMES_PER_SECOND/EXTRA_AVATAR_DATA_FRAME_RATIO) times
// per second.
// This value should be a power of two for performance purposes, as the mixer performs a modulo operation every frame
// to determine whether the extra data should be sent.
static const int EXTRA_AVATAR_DATA_FRAME_RATIO = 16;
// FIXME - There is some old logic (unchanged as of 2/17/17) that randomly decides to send an identity
// packet. That logic had the following comment about the constants it uses...
//
@ -156,6 +149,20 @@ void AvatarMixerSlave::broadcastAvatarData(const SharedNodePointer& node) {
// setup a PacketList for the avatarPackets
auto avatarPacketList = NLPacketList::create(PacketType::BulkAvatarData);
// Define the minimum bubble size
static const glm::vec3 minBubbleSize = glm::vec3(0.3f, 1.3f, 0.3f);
// Define the scale of the box for the current node
glm::vec3 nodeBoxScale = (nodeData->getPosition() - nodeData->getGlobalBoundingBoxCorner()) * 2.0f;
// Set up the bounding box for the current node
AABox nodeBox(nodeData->getGlobalBoundingBoxCorner(), nodeBoxScale);
// Clamp the size of the bounding box to a minimum scale
if (glm::any(glm::lessThan(nodeBoxScale, minBubbleSize))) {
nodeBox.setScaleStayCentered(minBubbleSize);
}
// Quadruple the scale of both bounding boxes
nodeBox.embiggen(4.0f);
// setup list of AvatarData as well as maps to map betweeen the AvatarData and the original nodes
// for calling the AvatarData::sortAvatars() function and getting our sorted list of client nodes
QList<AvatarSharedPointer> avatarList;
@ -170,7 +177,6 @@ void AvatarMixerSlave::broadcastAvatarData(const SharedNodePointer& node) {
avatarList << otherAvatar;
avatarDataToNodes[otherAvatar] = otherNode;
}
});
AvatarSharedPointer thisAvatar = nodeData->getAvatarSharedPointer();
@ -191,8 +197,87 @@ void AvatarMixerSlave::broadcastAvatarData(const SharedNodePointer& node) {
return glm::max(nodeBoxHalfScale.x, glm::max(nodeBoxHalfScale.y, nodeBoxHalfScale.z));
},
[thisAvatar](AvatarSharedPointer avatar)->bool{
return (avatar == thisAvatar); // ignore ourselves...
[&](AvatarSharedPointer avatar)->bool{
if (avatar == thisAvatar) {
return true; // ignore ourselves...
}
bool shouldIgnore = false;
// We will also ignore other nodes for a couple of different reasons:
// 1) ignore bubbles and ignore specific node
// 2) the node hasn't really updated it's frame data recently, this can
// happen if for example the avatar is connected on a desktop and sending
// updates at ~30hz. So every 3 frames we skip a frame.
auto avatarNode = avatarDataToNodes[avatar];
if (avatarNode) {
const AvatarMixerClientData* avatarNodeData = reinterpret_cast<const AvatarMixerClientData*>(avatarNode->getLinkedData());
if (avatarNodeData) {
quint64 startIgnoreCalculation = usecTimestampNow();
// make sure we have data for this avatar, that it isn't the same node,
// and isn't an avatar that the viewing node has ignored
// or that has ignored the viewing node
if (!avatarNode->getLinkedData()
|| avatarNode->getUUID() == node->getUUID()
|| (node->isIgnoringNodeWithID(avatarNode->getUUID()) && !getsIgnoredByMe)
|| (avatarNode->isIgnoringNodeWithID(node->getUUID()) && !getsAnyIgnored)) {
shouldIgnore = true;
} else {
// Check to see if the space bubble is enabled
if (node->isIgnoreRadiusEnabled() || avatarNode->isIgnoreRadiusEnabled()) {
// Define the scale of the box for the current other node
glm::vec3 otherNodeBoxScale = (avatarNodeData->getPosition() - avatarNodeData->getGlobalBoundingBoxCorner()) * 2.0f;
// Set up the bounding box for the current other node
AABox otherNodeBox(avatarNodeData->getGlobalBoundingBoxCorner(), otherNodeBoxScale);
// Clamp the size of the bounding box to a minimum scale
if (glm::any(glm::lessThan(otherNodeBoxScale, minBubbleSize))) {
otherNodeBox.setScaleStayCentered(minBubbleSize);
}
// Quadruple the scale of both bounding boxes
otherNodeBox.embiggen(4.0f);
// Perform the collision check between the two bounding boxes
if (nodeBox.touches(otherNodeBox)) {
nodeData->ignoreOther(node, avatarNode);
shouldIgnore = !getsAnyIgnored;
}
}
// Not close enough to ignore
if (!shouldIgnore) {
nodeData->removeFromRadiusIgnoringSet(node, avatarNode->getUUID());
}
}
quint64 endIgnoreCalculation = usecTimestampNow();
_stats.ignoreCalculationElapsedTime += (endIgnoreCalculation - startIgnoreCalculation);
if (!shouldIgnore) {
AvatarDataSequenceNumber lastSeqToReceiver = nodeData->getLastBroadcastSequenceNumber(avatarNode->getUUID());
AvatarDataSequenceNumber lastSeqFromSender = avatarNodeData->getLastReceivedSequenceNumber();
// FIXME - This code does appear to be working. But it seems brittle.
// It supports determining if the frame of data for this "other"
// avatar has already been sent to the reciever. This has been
// verified to work on a desktop display that renders at 60hz and
// therefore sends to mixer at 30hz. Each second you'd expect to
// have 15 (45hz-30hz) duplicate frames. In this case, the stat
// avg_other_av_skips_per_second does report 15.
//
// make sure we haven't already sent this data from this sender to this receiver
// or that somehow we haven't sent
if (lastSeqToReceiver == lastSeqFromSender && lastSeqToReceiver != 0) {
++numAvatarsHeldBack;
shouldIgnore = true;
} else if (lastSeqFromSender - lastSeqToReceiver > 1) {
// this is a skip - we still send the packet but capture the presence of the skip so we see it happening
++numAvatarsWithSkippedFrames;
}
}
}
}
return shouldIgnore;
});
// loop through our sorted avatars and allocate our bandwidth to them accordingly
@ -219,223 +304,107 @@ void AvatarMixerSlave::broadcastAvatarData(const SharedNodePointer& node) {
// NOTE: Here's where we determine if we are over budget and drop to bare minimum data
int minimRemainingAvatarBytes = minimumBytesPerAvatar * remainingAvatars;
bool overBudget = (identityBytesSent + numAvatarDataBytes + minimRemainingAvatarBytes) > maxAvatarBytesPerFrame;
/**
qDebug() << "Budget debugging:"
<< "numAvatarDataBytes:" << numAvatarDataBytes << "\n"
<< "identityBytesSent:" << identityBytesSent << "\n"
<< "remainingAvatars:" << remainingAvatars << "\n"
<< "minimumBytesPerAvatar:" << minimumBytesPerAvatar << "\n"
<< "minimRemainingAvatarBytes:" << minimRemainingAvatarBytes << "\n"
<< "overBudget:" << overBudget << "\n"
<< "overBudgetAvatars:" << overBudgetAvatars;
**/
bool shouldConsider = false;
quint64 startIgnoreCalculation = usecTimestampNow();
quint64 startAvatarDataPacking = usecTimestampNow();
// make sure we have data for this avatar, that it isn't the same node,
// and isn't an avatar that the viewing node has ignored
// or that has ignored the viewing node
if (!otherNode->getLinkedData()
|| otherNode->getUUID() == node->getUUID()
|| (node->isIgnoringNodeWithID(otherNode->getUUID()) && !getsIgnoredByMe)
|| (otherNode->isIgnoringNodeWithID(node->getUUID()) && !getsAnyIgnored)) {
++numOtherAvatars;
shouldConsider = false;
const AvatarMixerClientData* otherNodeData = reinterpret_cast<const AvatarMixerClientData*>(otherNode->getLinkedData());
// make sure we send out identity packets to and from new arrivals.
bool forceSend = !nodeData->checkAndSetHasReceivedFirstPacketsFrom(otherNode->getUUID());
// FIXME - this clause seems suspicious "... || otherNodeData->getIdentityChangeTimestamp() > _lastFrameTimestamp ..."
if (!overBudget
&& otherNodeData->getIdentityChangeTimestamp().time_since_epoch().count() > 0
&& (forceSend
|| otherNodeData->getIdentityChangeTimestamp() > _lastFrameTimestamp
|| distribution(generator) < IDENTITY_SEND_PROBABILITY)) {
identityBytesSent += sendIdentityPacket(otherNodeData, node);
}
const AvatarData* otherAvatar = otherNodeData->getConstAvatarData();
glm::vec3 otherPosition = otherAvatar->getClientGlobalPosition();
// determine if avatar is in view, to determine how much data to include...
glm::vec3 otherNodeBoxScale = (otherPosition - otherNodeData->getGlobalBoundingBoxCorner()) * 2.0f;
AABox otherNodeBox(otherNodeData->getGlobalBoundingBoxCorner(), otherNodeBoxScale);
bool isInView = nodeData->otherAvatarInView(otherNodeBox);
// start a new segment in the PacketList for this avatar
avatarPacketList->startSegment();
AvatarData::AvatarDataDetail detail;
if (overBudget) {
overBudgetAvatars++;
_stats.overBudgetAvatars++;
detail = AvatarData::NoData;
} else if (!isInView && !getsOutOfView) {
detail = AvatarData::NoData;
nodeData->incrementAvatarOutOfView();
} else {
const AvatarMixerClientData* otherData = reinterpret_cast<AvatarMixerClientData*>(otherNode->getLinkedData());
shouldConsider = true; // assume we will consider...
// Check to see if the space bubble is enabled
if (node->isIgnoreRadiusEnabled() || otherNode->isIgnoreRadiusEnabled()) {
// Define the minimum bubble size
static const glm::vec3 minBubbleSize = glm::vec3(0.3f, 1.3f, 0.3f);
// Define the scale of the box for the current node
glm::vec3 nodeBoxScale = (nodeData->getPosition() - nodeData->getGlobalBoundingBoxCorner()) * 2.0f;
// Define the scale of the box for the current other node
glm::vec3 otherNodeBoxScale = (otherData->getPosition() - otherData->getGlobalBoundingBoxCorner()) * 2.0f;
// Set up the bounding box for the current node
AABox nodeBox(nodeData->getGlobalBoundingBoxCorner(), nodeBoxScale);
// Clamp the size of the bounding box to a minimum scale
if (glm::any(glm::lessThan(nodeBoxScale, minBubbleSize))) {
nodeBox.setScaleStayCentered(minBubbleSize);
}
// Set up the bounding box for the current other node
AABox otherNodeBox(otherData->getGlobalBoundingBoxCorner(), otherNodeBoxScale);
// Clamp the size of the bounding box to a minimum scale
if (glm::any(glm::lessThan(otherNodeBoxScale, minBubbleSize))) {
otherNodeBox.setScaleStayCentered(minBubbleSize);
}
// Quadruple the scale of both bounding boxes
nodeBox.embiggen(4.0f);
otherNodeBox.embiggen(4.0f);
// Perform the collision check between the two bounding boxes
if (nodeBox.touches(otherNodeBox)) {
nodeData->ignoreOther(node, otherNode);
shouldConsider = getsAnyIgnored;
}
}
// Not close enough to ignore
if (shouldConsider) {
nodeData->removeFromRadiusIgnoringSet(node, otherNode->getUUID());
}
quint64 endIgnoreCalculation = usecTimestampNow();
_stats.ignoreCalculationElapsedTime += (endIgnoreCalculation - startIgnoreCalculation);
detail = distribution(generator) < AVATAR_SEND_FULL_UPDATE_RATIO
? AvatarData::SendAllData : AvatarData::CullSmallData;
nodeData->incrementAvatarInView();
}
if (shouldConsider) {
quint64 startAvatarDataPacking = usecTimestampNow();
bool includeThisAvatar = true;
auto lastEncodeForOther = nodeData->getLastOtherAvatarEncodeTime(otherNode->getUUID());
QVector<JointData>& lastSentJointsForOther = nodeData->getLastOtherAvatarSentJoints(otherNode->getUUID());
bool distanceAdjust = true;
glm::vec3 viewerPosition = myPosition;
AvatarDataPacket::HasFlags hasFlagsOut; // the result of the toByteArray
bool dropFaceTracking = false;
++numOtherAvatars;
const AvatarMixerClientData* otherNodeData = reinterpret_cast<const AvatarMixerClientData*>(otherNode->getLinkedData());
// make sure we send out identity packets to and from new arrivals.
bool forceSend = !nodeData->checkAndSetHasReceivedFirstPacketsFrom(otherNode->getUUID());
// FIXME - this clause seems suspicious "... || otherNodeData->getIdentityChangeTimestamp() > _lastFrameTimestamp ..."
if (!overBudget
&& otherNodeData->getIdentityChangeTimestamp().time_since_epoch().count() > 0
&& (forceSend
|| otherNodeData->getIdentityChangeTimestamp() > _lastFrameTimestamp
|| distribution(generator) < IDENTITY_SEND_PROBABILITY)) {
identityBytesSent += sendIdentityPacket(otherNodeData, node);
}
const AvatarData* otherAvatar = otherNodeData->getConstAvatarData();
// Decide whether to send this avatar's data based on it's distance from us
glm::vec3 otherPosition = otherAvatar->getClientGlobalPosition();
float distanceToAvatar = glm::length(myPosition - otherPosition);
if (shouldConsider) {
AvatarDataSequenceNumber lastSeqToReceiver = nodeData->getLastBroadcastSequenceNumber(otherNode->getUUID());
AvatarDataSequenceNumber lastSeqFromSender = otherNodeData->getLastReceivedSequenceNumber();
// FIXME - This code does appear to be working. But it seems brittle.
// It supports determining if the frame of data for this "other"
// avatar has already been sent to the reciever. This has been
// verified to work on a desktop display that renders at 60hz and
// therefore sends to mixer at 30hz. Each second you'd expect to
// have 15 (45hz-30hz) duplicate frames. In this case, the stat
// avg_other_av_skips_per_second does report 15.
//
// make sure we haven't already sent this data from this sender to this receiver
// or that somehow we haven't sent
if (lastSeqToReceiver == lastSeqFromSender && lastSeqToReceiver != 0) {
++numAvatarsHeldBack;
quint64 endAvatarDataPacking = usecTimestampNow();
_stats.avatarDataPackingElapsedTime += (endAvatarDataPacking - startAvatarDataPacking);
shouldConsider = false;
} else if (lastSeqFromSender - lastSeqToReceiver > 1) {
// this is a skip - we still send the packet but capture the presence of the skip so we see it happening
++numAvatarsWithSkippedFrames;
}
// we're going to send this avatar
if (shouldConsider) {
// determine if avatar is in view, to determine how much data to include...
glm::vec3 otherNodeBoxScale = (otherPosition - otherNodeData->getGlobalBoundingBoxCorner()) * 2.0f;
AABox otherNodeBox(otherNodeData->getGlobalBoundingBoxCorner(), otherNodeBoxScale);
bool isInView = nodeData->otherAvatarInView(otherNodeBox);
// this throttles the extra data to only be sent every Nth message
/**
if (!isInView && !getsOutOfView && (lastSeqToReceiver % EXTRA_AVATAR_DATA_FRAME_RATIO > 0)) {
quint64 endAvatarDataPacking = usecTimestampNow();
_stats.avatarDataPackingElapsedTime += (endAvatarDataPacking - startAvatarDataPacking);
shouldConsider = false;
}
***/
if (shouldConsider) {
// start a new segment in the PacketList for this avatar
avatarPacketList->startSegment();
AvatarData::AvatarDataDetail detail;
if (overBudget) {
overBudgetAvatars++;
_stats.overBudgetAvatars++;
detail = AvatarData::NoData;
} else if (!isInView && !getsOutOfView) {
detail = AvatarData::NoData;
nodeData->incrementAvatarOutOfView();
} else {
detail = distribution(generator) < AVATAR_SEND_FULL_UPDATE_RATIO
? AvatarData::SendAllData : AvatarData::CullSmallData;
nodeData->incrementAvatarInView();
}
{
bool includeThisAvatar = true;
auto lastEncodeForOther = nodeData->getLastOtherAvatarEncodeTime(otherNode->getUUID());
QVector<JointData>& lastSentJointsForOther = nodeData->getLastOtherAvatarSentJoints(otherNode->getUUID());
bool distanceAdjust = true;
glm::vec3 viewerPosition = myPosition;
AvatarDataPacket::HasFlags hasFlagsOut; // the result of the toByteArray
bool dropFaceTracking = false;
quint64 start = usecTimestampNow();
QByteArray bytes = otherAvatar->toByteArray(detail, lastEncodeForOther, lastSentJointsForOther,
hasFlagsOut, dropFaceTracking, distanceAdjust, viewerPosition, &lastSentJointsForOther);
quint64 end = usecTimestampNow();
_stats.toByteArrayElapsedTime += (end - start);
static const int MAX_ALLOWED_AVATAR_DATA = (1400 - NUM_BYTES_RFC4122_UUID);
if (bytes.size() > MAX_ALLOWED_AVATAR_DATA) {
qCWarning(avatars) << "otherAvatar.toByteArray() resulted in very large buffer:" << bytes.size() << "... attempt to drop facial data";
dropFaceTracking = true; // first try dropping the facial data
bytes = otherAvatar->toByteArray(detail, lastEncodeForOther, lastSentJointsForOther,
hasFlagsOut, dropFaceTracking, distanceAdjust, viewerPosition, &lastSentJointsForOther);
if (bytes.size() > MAX_ALLOWED_AVATAR_DATA) {
qCWarning(avatars) << "otherAvatar.toByteArray() without facial data resulted in very large buffer:" << bytes.size() << "... reduce to MinimumData";
bytes = otherAvatar->toByteArray(AvatarData::MinimumData, lastEncodeForOther, lastSentJointsForOther,
quint64 start = usecTimestampNow();
QByteArray bytes = otherAvatar->toByteArray(detail, lastEncodeForOther, lastSentJointsForOther,
hasFlagsOut, dropFaceTracking, distanceAdjust, viewerPosition, &lastSentJointsForOther);
}
quint64 end = usecTimestampNow();
_stats.toByteArrayElapsedTime += (end - start);
if (bytes.size() > MAX_ALLOWED_AVATAR_DATA) {
qCWarning(avatars) << "otherAvatar.toByteArray() MinimumData resulted in very large buffer:" << bytes.size() << "... FAIL!!";
includeThisAvatar = false;
}
}
static const int MAX_ALLOWED_AVATAR_DATA = (1400 - NUM_BYTES_RFC4122_UUID);
if (bytes.size() > MAX_ALLOWED_AVATAR_DATA) {
qCWarning(avatars) << "otherAvatar.toByteArray() resulted in very large buffer:" << bytes.size() << "... attempt to drop facial data";
if (includeThisAvatar) {
numAvatarDataBytes += avatarPacketList->write(otherNode->getUUID().toRfc4122());
numAvatarDataBytes += avatarPacketList->write(bytes);
_stats.numOthersIncluded++;
dropFaceTracking = true; // first try dropping the facial data
bytes = otherAvatar->toByteArray(detail, lastEncodeForOther, lastSentJointsForOther,
hasFlagsOut, dropFaceTracking, distanceAdjust, viewerPosition, &lastSentJointsForOther);
// increment the number of avatars sent to this reciever
nodeData->incrementNumAvatarsSentLastFrame();
if (bytes.size() > MAX_ALLOWED_AVATAR_DATA) {
qCWarning(avatars) << "otherAvatar.toByteArray() without facial data resulted in very large buffer:" << bytes.size() << "... reduce to MinimumData";
bytes = otherAvatar->toByteArray(AvatarData::MinimumData, lastEncodeForOther, lastSentJointsForOther,
hasFlagsOut, dropFaceTracking, distanceAdjust, viewerPosition, &lastSentJointsForOther);
}
// set the last sent sequence number for this sender on the receiver
nodeData->setLastBroadcastSequenceNumber(otherNode->getUUID(),
otherNodeData->getLastReceivedSequenceNumber());
// remember the last time we sent details about this other node to the receiver
nodeData->setLastBroadcastTime(otherNode->getUUID(), start);
}
}
avatarPacketList->endSegment();
quint64 endAvatarDataPacking = usecTimestampNow();
_stats.avatarDataPackingElapsedTime += (endAvatarDataPacking - startAvatarDataPacking);
}
}
if (bytes.size() > MAX_ALLOWED_AVATAR_DATA) {
qCWarning(avatars) << "otherAvatar.toByteArray() MinimumData resulted in very large buffer:" << bytes.size() << "... FAIL!!";
includeThisAvatar = false;
}
}
if (includeThisAvatar) {
numAvatarDataBytes += avatarPacketList->write(otherNode->getUUID().toRfc4122());
numAvatarDataBytes += avatarPacketList->write(bytes);
_stats.numOthersIncluded++;
// increment the number of avatars sent to this reciever
nodeData->incrementNumAvatarsSentLastFrame();
// set the last sent sequence number for this sender on the receiver
nodeData->setLastBroadcastSequenceNumber(otherNode->getUUID(),
otherNodeData->getLastReceivedSequenceNumber());
// remember the last time we sent details about this other node to the receiver
nodeData->setLastBroadcastTime(otherNode->getUUID(), start);
}
avatarPacketList->endSegment();
quint64 endAvatarDataPacking = usecTimestampNow();
_stats.avatarDataPackingElapsedTime += (endAvatarDataPacking - startAvatarDataPacking);
};
quint64 startPacketSending = usecTimestampNow();