Latency and Packet Loss
A multiplayer game operating over the internet has to deal with several adverse factors that aren't present when developing single-player or LAN-only multiplayer games.
According to wikipedia, 200 ms of input lag is the lower threshold at which people typically notice lag.
Different game genres and platforms allow for different acceptable latency budgets:
- VR games are highly sensitive to latency: 7-20ms.
- Fighting games and shooter games with twitchy gameplay are the next most sensitive: 16-150ms latency before the user starts noticing lag regardless of frame rate.
- RTS games are on the other side of the latency tolerance spectrum: values up to 500ms can be acceptable and not degrade the overall experience.
Round Trip Time without frame calculation:
When a PC or console "pings" the server, it sends an ICMP (Internet Control Message Protocol) echo request to the game server, which then answers this request by returning an ICMP echo reply.
The time between sending the request and receiving the answer is your ping to the game server. This means that with a ping of 20ms, it takes data 10ms to travel from the client to the server, as the ping is the round-trip time of your data.
Higher ping values mean that there is more delay or lag, which is why you want to play on servers with low pings, as that is the basic prerequisite for games to feel snappy and responsive.
One of the adverse factors is Latency, which in the context of games means the amount of time between a cause and it's visible effect. An example can be a click on a button and a message popping up in response to said click.
Excessive latency that causes noticeable delay between cause and effect is typically referred to as Lag.
While we can minimize the perceived impact of latency, we can't get rid of it - latency is just a fact of life that has to be taken into account.
There are both network and non-network related components of latency.
Non-network latency is a serious issue and can eat up a large chunk of our ~200ms latency budget. This means that for online games input lag, and specifically the part that is largely under our control (render pipeline related lag) should be minimized.
- Input sampling delay: The time it takes for the input device to recognize that it has been activated and the time it takes for the game to detect that change
- Render pipeline delay: GPUs don't perform draw commands immediately, instead they batch them to be performed later on
- Vsync: Prevents an artifact known as screen tearing by locking the GPU to the vertical blanking interval of the monitor
- Display processing delay: Display devices typically process the incoming signal in some ways (such as deinterlacing and noise cancellation), which adds to latency
- Pixel response time: LCD screen pixels physically take time to change their brightness
All in all a sum of non-network latency contributors can be called Input Lag - the time it takes user input to be rendered on screen (without any network communications being involved).
Processing delay: This is the time it takes for the router to read the packet header and to figure out the next machine that should receive the packet and then to forward it to the appropriate interface.note
This delay also includes any encryption, NAT, network mapping and other work that the router might be doing. Typically this delay is low, but it's total effect can accumulate the more hops the packet takes).
Transmission delay: This is the time it takes to push the packet bits onto the physical link, irrespective of the distance to the destination. it's proportional to the packet length in bits. Transmission delay is greatest at the "last mile" of the internet - the parts of the network that goes to the end users. Fewer but bigger packets sent would reduce the amount of bits being spent on headers (in proportion to the bits used for the game data), thus reducing your transmission delay somewhat.
Queueing delay: Becasue a router can only process a limited number of packets at a time, if it receives more packets that it can handle - these packets get queued in the receive queue. In a similar way, a network interface can also only output a single packet at a time, so if after the packet is processed the appropriate network interface is busy, then the packet will land in the transmission queue. Queueing delay can be a significant contributor to overall latency.tip
Typically fewer large packets perform better than many smaller packets because typical routing requires examining only the header of the packet.
Propagation delay: This is the time signal spends traveling through the physical medium. It can't be less than the speed of light divided by the distance between source and destination, which is:
Propagation Delay = 0.3-ns * meters-to-travel
In practice that means that even under ideal conditions, it would take at least 50ms for a packet to go from Montreal to Australia, and approximately 15ms to cross the USA east to west.
This type of delay is optimizable in a client-server game - by placing servers closer to your players you effectively reduce propagation delay. In case dispersing servers isn't an option for gameplay reasons (for instance when your players NEED to be able to play with as little latency across, say, the entire US) an even more expensive and performant option is possible - a so-called edge network can be built.funfact
Riot games famously had to go that route for their League of Legends, building their own network that peered with ISPs across North America to provide the fastest connection possible and reducing network hops as much as possible.
In the context of networking it's valuable to consider the combination of network latency factors which is typically referred to as Ping or Round Trip Time (RTT).
Round Trip Time (RTT)
RTT is the time it takes for a packet to travel from one host to another and then for a response packet to travel back. This includes both the two-way sum of network latency factors and, as it contributes to how quickly the server can send out a response, the frame rate of the remote host,.
Since the traffic is unlikely to travel at the same speed in each direction, the RTT is rarely exactly the time it takes for a packet to go from one host to another. Regardless, it's a common practice to approximate one-way travel time by dividing RTT by 2.
Another complication arises from the fact that for any two hosts, the RTT time between them isn't a constant. It varies over time, normally hovering around a certain average value. The components of network latency can also vary over time, causing the RTT to deviate from the expected value. This deviation is called Jitter.
Jitter is the rate at which ping changes over a period of time. It can affect RTT mitigation and also make packets arrive out of order if it causes the router to send packets through different routes, which can cause an older packet to arrive before a newer packet.
Another danger that can befall our packets apart from being delayed is being lost entirely. Packet loss can be a greater problem than even Jitter.
Packet Loss, apart from degrading and potentially making the game unplayable, can contribute to the overall delay the user would experience. This occurs when packet loss heavily affects the data that must be delivered reliably and forces the game to resend data, multiplying delays.