Clustering technology status & proposal
Posted: Sat Mar 26, 2016 8:08 pmGreetings,
A few weeks ago, we presented the ongoing work regarding the hardware clustering development.
Video can be found here.
Since that day, what was presented in the video has been reworked (code refactoring, optimization) to match all the different use-cases, and the development team started working on group management between different server nodes and all the other interactions. For example, both nodes and master servers need to be aware of the current group, node of a given character, of the group attribution policy, mail, buffs, etc ... We had initially planned to finish this development for May. However, unforeseen events happened recently, and took a significant part of the developers' time. As a consequence, this clustering evolution won't be available for public testing until June. We know that with above 11k online players, lag starts to appear on the server in some areas, and may impact the in-game experience. So what solution do we have until that?
The most expensive hardware that we have right now is hosting the PVP, PTR, ISVV and 3 Development realms. We propose to the community to acquire more powerful hardware dedicated to the PVP realm. This new hardware would be 20% faster than the current one and 100% dedicated to the PVP realm only.
The price of this new hardware would be 360€ per month. Please contact Viper or Daemon in PM if you can afford it, and we will provide the OVH payment link. As a reminder, don't expect any in-game advantage for your support.
Benefits:
* The delay would be reduced on the PvP realm, which would allow us to increase visibility distance again when we are reaching our population peaks
* Both Dev and IsVV team will be able to work more efficiently, and process computationally expensive performance benchmarks (currently, we have to shutdown Dev & IsVV realms to reduce lag on the PVP realm, freezing development & ISVV progress)
* Once the clustering is available, we will be able to split workloads between this new server, and the current one
Our goal is to have the clustering technology ready for a public test for the AQ gate opening.
Best regards.
As a remember, below is the technology already developed to handle symbolic player caps:
* (2015/02) Instance, raid & battegrounds are handled in parallel
* (2015/03) MapReduce Paradigm at sub-map level
* (2015/04) Hardware changes
* (2015/05) Priorities between maps (instances, raids, BGs over continents)
* (2015/06) Priorities between players (in combat, or AFK)
* (2015/07) Priorities between player actions (spells, movements over chat or mail)
* (2015/08) Software Map clustering
Transverse works:
* (2015/03) Real time performance monitoring : dynamic performance parameters adjustments depending on server load
* (2015/03) Dynamic Respawn System for NPCs: The server analyzes the number of players playing each map in real time and adapts the respawn timers accordingly to fit with the rate on retail.
* (2015/12) Dynamic Respawn System extended to nodes: The server analyzes the number of players playing each map in real time and adapts the respawn timers accordingly to fit with the rate on retail.
A few weeks ago, we presented the ongoing work regarding the hardware clustering development.
Video can be found here.
Since that day, what was presented in the video has been reworked (code refactoring, optimization) to match all the different use-cases, and the development team started working on group management between different server nodes and all the other interactions. For example, both nodes and master servers need to be aware of the current group, node of a given character, of the group attribution policy, mail, buffs, etc ... We had initially planned to finish this development for May. However, unforeseen events happened recently, and took a significant part of the developers' time. As a consequence, this clustering evolution won't be available for public testing until June. We know that with above 11k online players, lag starts to appear on the server in some areas, and may impact the in-game experience. So what solution do we have until that?
The most expensive hardware that we have right now is hosting the PVP, PTR, ISVV and 3 Development realms. We propose to the community to acquire more powerful hardware dedicated to the PVP realm. This new hardware would be 20% faster than the current one and 100% dedicated to the PVP realm only.
The price of this new hardware would be 360€ per month. Please contact Viper or Daemon in PM if you can afford it, and we will provide the OVH payment link. As a reminder, don't expect any in-game advantage for your support.
Benefits:
* The delay would be reduced on the PvP realm, which would allow us to increase visibility distance again when we are reaching our population peaks
* Both Dev and IsVV team will be able to work more efficiently, and process computationally expensive performance benchmarks (currently, we have to shutdown Dev & IsVV realms to reduce lag on the PVP realm, freezing development & ISVV progress)
* Once the clustering is available, we will be able to split workloads between this new server, and the current one
Our goal is to have the clustering technology ready for a public test for the AQ gate opening.
Best regards.
As a remember, below is the technology already developed to handle symbolic player caps:
* (2015/02) Instance, raid & battegrounds are handled in parallel
* (2015/03) MapReduce Paradigm at sub-map level
* (2015/04) Hardware changes
* (2015/05) Priorities between maps (instances, raids, BGs over continents)
* (2015/06) Priorities between players (in combat, or AFK)
* (2015/07) Priorities between player actions (spells, movements over chat or mail)
* (2015/08) Software Map clustering
Transverse works:
* (2015/03) Real time performance monitoring : dynamic performance parameters adjustments depending on server load
* (2015/03) Dynamic Respawn System for NPCs: The server analyzes the number of players playing each map in real time and adapts the respawn timers accordingly to fit with the rate on retail.
* (2015/12) Dynamic Respawn System extended to nodes: The server analyzes the number of players playing each map in real time and adapts the respawn timers accordingly to fit with the rate on retail.
