Rogean
04-17-2014, 08:56 PM
After vast code improvements from Haynar and myself, we are looking to hold a stress test on the Velious Beta server this Saturday, around 9 PM Eastern.
The primary purpose of this is to test many code changes that have been made for bugs or other issues. We are also interested in seeing how the performance has increased, particularly for a raid situation.
To make the best use of this test, Nilbog and Alunova have escalated their efforts to bring Temple of Veeshan to a state that is also ready for testing. We will be utilizing this zone for much of the testing.
The first phase of testing will be a large group of players attacking a single, AE'ing mob. We would like to also test how many simultaneous player deaths affect the zone. We will start with between 100 and 150 players, and work up from there.
After that testing is completed, we will be splitting the guilds into three raid forces, for each wing of the zone.
What I am looking for is what guilds will be interested in helping us, and how many players they will be able to bring. Remember that boxing is allowed on Beta, but try to keep boxing to a managable number (I'd rather have as many real people as possible than half the raid force being boxes), so only box one or two other characters as necessary.
I will split and group the raids up depending on numbers and testing goals. The only thing I can say for certain is that Class C guilds will probably be raiding North wing.
For the rest of this post I'm going to go into details about some of the changes made, and what to look for.
Before I delve into the changes, I want to explain some of the performance measuring processes. It is almost impossible for us to attach any 'normal' type of performance gauging applications to our zone processes, as these will usually bog the zone down so much that it becomes unplayable. Therefor, we have opted to perform isolated profiling of various functions as we discover their slow execution.
We have already put in some minor checks for the zone process in general, when it is running slower than usual. This has already been helpful in identifying many issues, for example today a Kurns Tower AE group triggered our debug messages, and we are investigating why that zone in particular is having issues with pathing code. It does get tricky, however, since for example Kurns Tower has less mobs and half the pathing nodes of a zone like Sebilis, yet aggroing the entire sebilis zone does not cause anywhere near the same performance hit that Kurns is for some reason. Thus, the mad scientist (Haynar) will probably need to break the pathing code down into profiling sections to figure out why Kurns specifically is having issues.
So with that said, it is much easier for us to identify these issues with player help. The Inglourious Basterds group in kurns should have noticed how much they were lagging the zone and reported the issue and helped us identify the cause. Instead I had to investigate from work (Where I can't log into the game) and ask Derubael to send nasty words to members of that group. Not that it surprises me that IB continues to only care for themselves.
Anyways, moving on. If you guys have information regarding particular things that could be slowing a zone down, you are encouraged to report it so that we can profile that code.
As many of the raiders know, there were a lot of issues with players entering and leaving a zone. The functions that are processed during these actions were not very optimized. What was even worse was a player death. Believe it or not, moving all inventory items to a corpse was disgustingly tedious in it's previous form.
What I have done is recoded a lot of the multi threading sections of code to take as much database work off the primary zone thread as possible. Not just queries but a lot of information processing as a result of queries has been moved into a separate thread to be worked on, and then passed back. This has resulted in huge performance gains in these sections of code.
With any large change like this, there are bound to be issues. I've tested it as much as I can short on Beta, so I've managed to work most of them out. Part of this stress test is to identify if there are any remaining issues.
Keep an eye on all items in your inventory, particular during zoning and deaths. Make sure none of your items suddenly disappear, move locations, etc. Make sure looting your corpse is normal as well.
Let us know if you are crashing during zoning more than normally. (I don't want to hear from people that crash a lot normally anyways, only new cases).
There were also some changes to timers. There are two types of timers in code. Normal Timers and Persistent Timers. The former has been revamped, and is already live on the public servers. Not much testing needs to be done there. The latter... I completely ripped it out. The latter handles things like spell refresh, ability times, disciplines, etc. These have been consolidated into fields existing in the main profile that is saved with each character. These timers should be tested to make sure they are saving correctly across zoning.
The primary purpose of this is to test many code changes that have been made for bugs or other issues. We are also interested in seeing how the performance has increased, particularly for a raid situation.
To make the best use of this test, Nilbog and Alunova have escalated their efforts to bring Temple of Veeshan to a state that is also ready for testing. We will be utilizing this zone for much of the testing.
The first phase of testing will be a large group of players attacking a single, AE'ing mob. We would like to also test how many simultaneous player deaths affect the zone. We will start with between 100 and 150 players, and work up from there.
After that testing is completed, we will be splitting the guilds into three raid forces, for each wing of the zone.
What I am looking for is what guilds will be interested in helping us, and how many players they will be able to bring. Remember that boxing is allowed on Beta, but try to keep boxing to a managable number (I'd rather have as many real people as possible than half the raid force being boxes), so only box one or two other characters as necessary.
I will split and group the raids up depending on numbers and testing goals. The only thing I can say for certain is that Class C guilds will probably be raiding North wing.
For the rest of this post I'm going to go into details about some of the changes made, and what to look for.
Before I delve into the changes, I want to explain some of the performance measuring processes. It is almost impossible for us to attach any 'normal' type of performance gauging applications to our zone processes, as these will usually bog the zone down so much that it becomes unplayable. Therefor, we have opted to perform isolated profiling of various functions as we discover their slow execution.
We have already put in some minor checks for the zone process in general, when it is running slower than usual. This has already been helpful in identifying many issues, for example today a Kurns Tower AE group triggered our debug messages, and we are investigating why that zone in particular is having issues with pathing code. It does get tricky, however, since for example Kurns Tower has less mobs and half the pathing nodes of a zone like Sebilis, yet aggroing the entire sebilis zone does not cause anywhere near the same performance hit that Kurns is for some reason. Thus, the mad scientist (Haynar) will probably need to break the pathing code down into profiling sections to figure out why Kurns specifically is having issues.
So with that said, it is much easier for us to identify these issues with player help. The Inglourious Basterds group in kurns should have noticed how much they were lagging the zone and reported the issue and helped us identify the cause. Instead I had to investigate from work (Where I can't log into the game) and ask Derubael to send nasty words to members of that group. Not that it surprises me that IB continues to only care for themselves.
Anyways, moving on. If you guys have information regarding particular things that could be slowing a zone down, you are encouraged to report it so that we can profile that code.
As many of the raiders know, there were a lot of issues with players entering and leaving a zone. The functions that are processed during these actions were not very optimized. What was even worse was a player death. Believe it or not, moving all inventory items to a corpse was disgustingly tedious in it's previous form.
What I have done is recoded a lot of the multi threading sections of code to take as much database work off the primary zone thread as possible. Not just queries but a lot of information processing as a result of queries has been moved into a separate thread to be worked on, and then passed back. This has resulted in huge performance gains in these sections of code.
With any large change like this, there are bound to be issues. I've tested it as much as I can short on Beta, so I've managed to work most of them out. Part of this stress test is to identify if there are any remaining issues.
Keep an eye on all items in your inventory, particular during zoning and deaths. Make sure none of your items suddenly disappear, move locations, etc. Make sure looting your corpse is normal as well.
Let us know if you are crashing during zoning more than normally. (I don't want to hear from people that crash a lot normally anyways, only new cases).
There were also some changes to timers. There are two types of timers in code. Normal Timers and Persistent Timers. The former has been revamped, and is already live on the public servers. Not much testing needs to be done there. The latter... I completely ripped it out. The latter handles things like spell refresh, ability times, disciplines, etc. These have been consolidated into fields existing in the main profile that is saved with each character. These timers should be tested to make sure they are saving correctly across zoning.