interpreting Dashboard data

KevinB

Hi,

I am reading the dashboard real time data. I can see values of SocketServer.UdpDisconnectedPeersByClientPerSecond and SocketServer.UdpDisconnectedPeersPerSecond. However, it seems like SocketServer.UdpDisconnectedPeersByManagedPerSecond and SocketServer.UdpDisconnectedPeersByTimeoutPerSecond is always zero even though it's not. we are using server version 3.0.37.3631. Eventsentcount and eventsentpersec seems to be always zero as well.

also, what is the appropriate range of commandsResentPerSecond? i am constantly seeing 1-3 commands resent per second. it can hit 5-7 every now and then. oh btw, the avg cpu usage is just below 10%, so I know it's not cpu bound.

there's also some reading for Enet Queue Avg. It used to be zero, but now we are constantly 1 and occasionally 2. is that the incoming or outgoing queue? the Business Queue Avg tends to always stay zero.

K

[Deleted User]

Hi Kevin,

regarding disconnects:
- "Managed" disconnects are the ones where you explicitly call peer.Disconnect() in your (managed code) server-side application.
- "Server" disconnects are disconnects triggered by Photon's native code - especially disconnects after an exception occurred, you should see a log entry for each one in the Photon-Default---. log (or Photon-Loadbalancing---.log, if you use that one).

It is very likely that these counters are 0.
Although, the data might not be 100% exact because of the way the "per second" counters are calculated.

EventSentCount & EventSentPerSec are not set in 3.0 - that was a bug, it will be fixed in the next Server SDK Release. (Coming soon, I hope!)

It is nearly impossible to say what an acceptable range of Resent commands is. (A "resend" happens if you send reliable commands in UDP and don't receive an ACK for it after a certain timeout).
First, it depends on the number of sent commands in total. If the resend rate is <1%, everything is perfectly fine, and Photon & your clients should be able to work smoothly with 5-10% as well. If resend rates are significantly > 10%, you might want to have a closer look.
You'll see the lowest rate of Packet loss (and therefore, resends) if client & server run on your local machine, its' in general also low on a LAN, higher on public internet and you'll see the highest resend rates if you have mobile (WLAN / 3G / ... ) clients. So there is no definitive answer to your question.

The ENet queue is the incoming queue.

You might be interested in the "photon_perfcounters.pdf" in the doc folder of your Server SDK - it contains a brief documentation for Photon's Windows Performance Counters (from which the dashboard counters are read).

KevinB

Thanks Nicole,

I am pretty sure the application has called peer.Disconnect() and my server log shows that there're ManagedDisconnect and TimeoutDisconnect. However, it's not shown on the dashboard data of the game server. meanwhile I can see TimeoutDisconnect on the dashboard data of the master server. my code is based on the loadbalancing example, wonder if there's something wrong with my GameApplication code, although I don't seem to find these anywhere in the code.

K