rare disconnect/crash due to packet spam

I defrag regularly and run CCleaner at least once a week. A friend helped me with some settings and explained that program to me many months ago.

I am very careful about what I put on this computer and make sure I know what it is doing.


Edit:
I even had the Windows "prefetch", or whatever that indexing of frequently used programs is called, disabled many months ago which increased my performance a lot. I know that even the standard programs the operating system has can be bad. My friend helped me optimize.
This was all before this crash bug ever started.

1) OK.
2) Hm. So no specific titles, just games in general?
3) How is it nitpicking to ask for clarification of something that was worded rather vaguely?

Actually that pretty much is how it works.

The server says Item a is here in state x
When you perform some action the client sends the details of what action is performed which is then applied to the state of each affected object. That is performed on the server.
Some things are indeed calculated on the client - particularly ragdoll physics and power animations - this is why different users will see different animations but nothing which can affect other users.
If you attack a mob using Propel for example player A might see a Sofa flying towards it whilst Player B sees a car - the selection of which animation is clientside but whether it hits, how much damage it takes, how far it gets knocked back are calculated serverside.
Similarly when you get a poor connection and suffer lost packets the location of all objects is updated from the details sent from the server - this is commonly known as rubberbanding (the client displays where it believes everything is until the server tells it where they actually are).
Once the data is received from the server there is no comparison - the objects locations displayed by the client are just set to the details held on the server.
Pretty much any other option would allow cheatbots to be set up.

Welcome to the forums.

Then why does my character, which has been afk not moving for 5 minutes, move when somebody else, who is "item b not here in state x", "collides" with my character?

If my character really is "item a in this spot in state x" then there is no way "item b" can become "item b in a's spot in state y".


As for Propel, I believe it shows the same object for everyone. The only thing anybody else notices different is what happens after it strikes the target, or after it misses. It still tosses the same item but the item moves in different ways when it is "loose" for each client. At least it used to work this way.
That is easily proven by just asking what object somebody sees when you use it.



My whole point was that the client can move your character and the server can say "you are here" and the client has to obey.
So when you send a command to use a power, the server receives the command and then sends back either "activate power" or "can't do that yet", causing exponential lag due to needing a send and reply upstream and downstream, instead of "a certain competitor" that eliminates the downstream "permission" step which cuts lag in half and ensures exponentially better performance.

Seriously, they're making more lag than necessary by attempting to ensure everyone see the same thing and failing because everyone lags more and makes everything much more out of sync. The results, such as health changes, can't be cheated anyway.
For example, if I were to fire a power at another player in PVP with my client side in absolute control and my client said "HIT" then I would still not have actually hit the other player unless their client side allows it. If they lose the packet that says the power was a hit, or the data is incorrect(because of hacking or cheating another way or just a bug), then they will not take the hit and will not lose to me.
You can't cheat when both sides are separate instances of the client because you don't have control of the other side.


The server just needs to "pass it along", sharing necessary information with everyone within the same zone or instance, especially considering that even NPC and event tracking is pre-programmed and only changes when there is an observer present.
Just run through Perez Park in the hellion section during a slow time on the server to see what I mean. You will see and hear the "Girlfriend from Hell" NPCs summoning their pets because you got in range for them to be rendered and tracked in real time.(Edit: They do this because nobody else was in range within a certain time-frame to trigger the rendering and tracking before you.)

It's like the old saying "if a tree falls in the forest and nobody is within earshot, does it make a sound?".
In this case, nothing in game actually happens unless someone is there as witness.
Yes, many events are timed, but they still don't actually proceed unless there are players within the area to witness it. You won't see Rikti or Zombies or Nemesis pop from an invasion unless there is a player there for them to trigger. Then, the server starts tracking them for that player and sending the copies of that information to new entrants into the area.
That is why you encounter a lag spike and/or graphical stutter when entering an area with a lot going on. You don't load it until you are in range to track it.


So, it's not necessary to "ask the server for permission" to move or activate powers.
1) It's always tracking whether you are connected or not anyway.
2) You won't receive NPC data or any other data if you aren't connected to be able to send your own data, and that means you can't fight anything anyway. It would be like the old "lost sync" bug that made it so you were able to move around a zone but unable to activate any doors or see any NPCs or players past a certain distance from your load-in point.
3) Even if it has the "ask permission" step, there are still obvious "lag effects" such as damage and animations being out of sync and other characters and NPCs appearing to teleport all over the place as the client gets packets out of order for an instant.

"Asking permission" doesn't work any better than not doing so, and usually works worse due to the back and forth exponential lag(lag is always present since no communication is instantaneous, at least until we figure out quantum entanglement, so even the slightest lag is multiplied).
In this case, it really is "better to ask forgiveness than permission" when it comes to online game code.

I was rather amused by the edit that was made in my post. But then, I generally don't try to tell the Moderator's how to do their job. They have a boss that already does that.

Of course, if people didn't break the forum rules then the Moderator's wouldn't have to edit their posts.

Or they could take the ham-fisted approach some did previously of just deleting the entire post instead of editing them. It would likely save them time to do that.

When the other person moves their client passes the move action to the server which then resolves the conflict exactly the same as if they had attacked a foe and knocked them back or performed any other action affecting another object. Once it has calculated where all affected objects are - in this case including your character it will send this data back to both players clients which will then render the result.

I'll admit I haven't checked whether the object is the same or not but the principle remains valid.
A demorecord session would show it one way or the other if you see objects in the files from that which are each of the propelled objects rather than a single generic propelled object it should render the same.
If its something that must display the same for all users it will be set serverside actions where this is not required can safely be calculated clientside
Ragdoll physics animations are definitely calculated clientside - they are compuationally expensive and not abusable so can and should be offloaded to the client,
Power effects are serverside. - Could be abused if offloaded to the client so need to remain serverside.
The choice of activation animation for powers which can show different animations which each generate an identical result could be done clientside as could character stances - I'm not 100% certain whether it is though. Computationally cheap but not abusable so doesn't really make a lot of difference where the decision is made.


When you attempt activate a power the client sends a message to the server to say you have done so. This does not activate the power - it merely tells the server you wish to do so. If the power can be activated the server will calculate the effect and update the client with the changes to be rendered. If it can't then it will clear any queues action and queue the requested action and then when it can be activated calculates the changes and updates the client. There is no permission step. This cannot be offloaded to the client as it would be abusable and also the expensive part is the download from server to client which on almost every home internet connection will have a greater bandwidth. Apart from the potential for abuse offloading this stage to the client would require rendering information for multiple objects to be passed from the client to the server and then back out to everyone else's client, greatly increasing upload bandwidth without reducing download requirements.

You have just perfectly explained the way this would allow cheating.
If player A attacks player B doing X damage but player Bs client says I'm just ignoring any damage I take then the worst that can happen to player B is a draw - well until they get found out and get banned.
The calcuation of the damage done and whether it has hit is done on the server for good reasons.
The data passed from the client to the server is minimal - all it needs is Player A fires power X with Target Player B.
To try and do this client side would require the client to get the details of both the power being activated and of all the targets it may hit. You might hold details of the power but you aren't going to hold details of every possible target - i.e. every mob at every level, every power which could affect them and also the characters for the every player in the entire playerbase as a result you would need to make remote database queries to obtain that information, then calculate the result, then send an update query back to the serverside databases. Then the server could pass that along to all clients - this last stage being exactly the same as now.



So why are you advocating doing so?
As I have said there is no ask permission stage

Client asks server to perform action

If the request doesn't reach the server it doesn't happen
If it isn't an allowed action it doesn't happen
If it fails it doesn't happen

If it succeeds the server calculates the effect and updates the status of all affected objects.

Whilst this is happening the server repeatedly sends the current status of all objects visible to each client to those clients which can see it.

If that update doesn't reach one client it will still reach the others and the client that missed it will get updated the next time the statuses are sent out.
If the client misses too many updates you will get a map server disconnected warning and eventually will get disconnected.

What I am saying is that the game's physics do not apply to player on player collisions.

Do you see a super speeder run into a player and see both players drop dead and go to the hospital or the one being hit get launched at half their combined velocity for quite a distance?
No, you don't.


One character is not moving and "owns that spot". The server "knows they own that spot".
But for some reason, the character that "is on the spot" moves "just a little" when another player comes too close.

Obviously, either the server is not "master" then or it has a problem actually tracking static objects, which includes players that are sending no movement commands whose client side is not determining movement(and it wouldn't move them when another player hits them anyway due to the fact that player collisions are not calculated for physics simulations).




Consider the standing character a concrete wall when it comes to the game's code. Why does the concrete wall move 1ft when being hit by a 60mph other player if the code says "the concrete wall wins" and should not move?



Edit:
Either the server really does have a confirmation step for client side commands or character tracking is broken.
Either possibility is not good for differing reasons.


Edit 2:
If a stationary player is supposed to be moved by a moving player colliding with them then that can be abused to push afk players into aggro.
Therefore, that should not be possible, and thus physics do not apply. They still move though, for some reason that I can only attribute to bad code or a bug.

What decides whether it is "allowed or not" and "succeeds or fails"?

It is not the client side, because lag would not affect powers then and certainly would not make 1000 ping seem like 2000+ and would likely allow cheating by forcing every attack to always hit for crazy damage(with a simple code change client side, a hack) because it wouldn't have the server say "that data is incorrect".

The server has to calculate "is this action allowed?" and then all the other data that determines success or failure then it sends that info back to the side performing the action and every other side that needs to track the action before the action can go through on all sides at the same time(hypothetically since it doesn't work that well in practice with this system).

This is why when we lose connection to mapserver we can't move or activate powers, the same with the "lost sync" bug described above that still allows us to move but not attack.






If the client did not have to ask the server for permission then the client could calculate everything and powers would never fail to activate when your client shows them as recharged. FYI, I know you have seen powers that look recharged but won't activate. That proves me correct.
I'm done explaining how server-client communication works.

That's because we are playing a Superhero game where actions are supposed to be cinematic and not a snooker table simulator where I would expect accurate physics to be used to calculate the effects of collisions including conservation of angular momentum and kinetic energy.
It used to be that players could not move other players which caused problems when someone went AFK in a doorway or narrow passage. That was changed to allow pushing and this is what you see when two superspeeders collide. They both reach the same point and then one or both push the other before determining final position.

No it isn't accurate physics but then again neither would be modelling all collisions as perfectly elastic. In the real world objects deform or fragments converting kinetic energy to sound and heat reducing the overall kinetic energy and momentum of both (all) objects involved in the vast majority of cases. This reduction is exaggerated for cinematic effect in the game.

Yes I hate the way pedestrians are far too good at pushing and cannot be pushed back as much as the next person. To say nothing of trucks being able to drive through each other.

My point was that character tracking is obviously broken.
1) We can't stay in one place if we want to and somebody pushes us.
2) We can't push other players far enough to matter if they're in a doorway, without a lot of repeated bashing.
3) It just creates more lag.

Correct

more like turn a 150ms ping into 6 or 8 seperate 150ms pings with a much greater amount of data being transferred in both directions as the seperate queries are made on the database.

exactly - which is why that validation needs to be performed serverside where there is no ability to change how it works.

Yes

No - the client does not need to know anything about why it failed or even that it did.
All it needs to know is the current status of the objects it needs to render - how they reached that state is irrelevant.


No - the client will estimate where it expects you to be and render your character there if it misses a few updates. When it receives the status of objects from the server it will display them according to that - that is your "lost sync" effect (commonly known as rubberbanding)

You can appear to move but as you cannot send you cannot activate powers.
Anything already queued will activate but as you are not receiving updates you cannot see this.

If it continues to not receive updates as to the status of objects then the client recognises that the problem is more severe and that it can no longer expect to recover - at this point it will stop estimating position and give the Mapserver warning

At this point there is no longer any point in estimating position so you cannot move in the client.

If it continues further then the client will disconnect from the server and drop you back to the login screen. As the server does not know the client has disconnected you get the message that the account is still logged in upon reconnection.


No I've seen powers where the client hasn't registered the activation which has happened - not where they haven't activated.
It doesn't prove you correct at all just that you don't understand how client server systems work.[/QUOTE]


Good - feel free to try again once you understand it.

No.
The object thrown at the enemy is rendered locally and is completely random. I *HAVE* seen it be the same item for two people on occasion, which is to be expected for a random selection. But over 90% of the time, each person sees something different. And this is the way it has always worked.

Maybe some testing should be done on that.

If true, that really messes up any attempts to say something funny when you toss a specific item, such as a chair and saying "have a seat!" or a lamp and saying "lights out!" or any other funny quip.

It's true, and it's been true for as long as I've been playing. Archie and I play sat at side-by-side desks, and we noticed the local determination of the Propel object pretty quickly.

No testing is needed. I just told you how it works. This is based on actual, eyewitness experience. This was discovered shortly after the power came into existence and is widely known.

And, yes, sadly, it ruins attempts at humor. My personal favorite is when I defeat an enemy by hitting them with a coffin. Sadly, a teammate might see them being hit with a roll of pink insulation - not sure how much that would hurt.....

Well, one difference right off is the quality of the netgraph between the two screenshots. The first one doesn't look TOO bad, but the second one is just horrible with those solid red patches. This suggests that your connection was way worse at the time of the second screenshot.


They were posted. It seems they only showed up in the Patch Notes section on the main site, and not in the Announcements section in the forums.


On a side note, the taskbar in both images show the NC Launcher being active. You really should close that. It consumes a terrible amount of resources and certainly isn't going to be helping the situation.

Up front, I admit that I have very little specific knowledge on how packets work in general and with CoH.

However; from what I noticed on your screen shots, your wireless connection was always between 1 - 2 bars, which shows very poor signal strength.

My recommendation, if possible, would be to try using your PC on a wired connection - at a friend or relative's house, to rule out that it's your connection itself causing the problems.

If you still experience the same issues on a wired connection, then you know it's either your PC or server side. If the problems go away, I think you can safely assume it has to do with your connection.

/2 inf

[ @Zombie Fryer ][ @Zombie Smasher ]
| Home Server: Virtue |

Twitter: @ZFLikesNachos Save City of Heroes (Titan Network) [Successful "The Really Hard Way" runs: 4] [Click ^]

I get almost the same amount of packet loss if I'm right next to the router. It is more than enough speed for my internet service and very reliable. I normally play the game with under 200 ping, even when I'm not the only one using the network.

Also, I have played with 4000 ping many times before, due to high load areas and/or my family hogging bandwidth.
I don't have problems with crashing under high lag situations. It just seems to be random, or more dependent on circumstances than lag or hardware.



I don't have any other option than the situation I am in right now.
I don't have this issue on other games. I never had this issue on that "800lb gorilla" competitor, and didn't even have it in CoH before incarnates, on this same computer.

I don't even have the issue solo when in high load situations, just teamed with incarnate level shifted team/league leaders, and still many seem to not cause the issue while some will cause it every mission.

First, the first screenshot was from at least 1 minute after what caused the problem with the power trays.

Second, the second time was when I had flown over a portion of the mission map with a lot of enemies(54x8 difficulty) which were just starting to fire on a teammate. That was the first time it loaded that area of the map, including those enemies.
I had the red netgraph spikes due to packet spam, which invariably caused me to lose a packet that was essential and crash the game.



So, this proves the "packet spam problem" this game never had before issue 19 while showing that it somehow is related to the default power trays going blank.

The most important thing, though, is that I was fully expecting to crash after the first screenshot, before it even, but I did not. For some reason, the game did not go into "2 minute count-down to crash" mode and instead lasted at least 4 minutes at which time I exited the mission again which fixed the trays and never crashed or had any issue again for that entire mission.
Somehow, the game was able to not go into what I think was a "cascade failure" after whatever caused the tray issue.





Edit:
Actually, the red blocks on the netgraph in the second picture were from the "crash in progress".

I have witnessed the netgraph "speeding up" during times of yellow or red spikes. It will "spam spikes" filling what would normally take even up to one minute to march across the netgraph disaply in under 5 seconds.
It's as if time speeds up right before the crash and the netgraph is recording a hell of a lot faster than it usually would.

So, that wasn't "red spikes before the crash" but actually "red spikes during the 1 second crash" that ended up looking like 10-15 seconds.



I think that could be a symptom or a cause of the problem.
The game seems to respond to lag by actually attempting to communicate with the server more often and faster. It would be like you not receiving a reply message from someone you know must have received it and, assuming the worst from the silence, started spamming them apology messages or questioning messages and just plain begging for them to respond.
The game is expecting a message and when it doesn't receive it right on time then it spams requests for it.

At least, that is my guess at an explanation for why the netgraph will suddenly accelerate when there is severe lag for even an instant.

Maybe I would have normally had the "lost connection to mapserver" message at those moments, but I instead "stay connecteD" and get a spammed netgraph instead.
That's still a problem with the game, if that is what is happening. I guarantee you I have seen a lot less "lost connection" messages(almost none) since issue 18 and a lot more crashes since the update with the BAF trial and level shifts.