Make "performance" indicate min performance over 30 minutes

Racer1

On the server browser, the current performance shown appears to be only either a "current" or very recent performance indicator. It needs to show the minimum performance over the last 30 or even 60 minutes (map loads/new round hiccups can be ignored). I want to know the worst I can reasonably expect, not what it is right now.

Alternatively, giving the worst performance over the last 3 rounds played (including the current one) would suffice.

SamusDroid

This would mean querying over 100 servers for performance which would be performance intensive, over 60 seconds.

ZEROibis

A better solution would be for them to fix the performance tank these last few builds have had.

Ghosthree3

This would mean that if the server did hitch even for 1 second and only one second for some reason. It would report a tickrate of 5 or something.

HeatSurge

Ghosthree3 wrote: »

This would mean that if the server did hitch even for 1 second and only one second for some reason. It would report a tickrate of 5 or something.

... which is about 100% of servers in late game. And could easily be solved by some kind of elimination of outliers or weighted averaging algo.

Omega_K2

It should be more of an avarage. Not min, min is gonna be bad for pretty much any server, i.e. when loading maps, or at round end/start.

Racer1

@SamusDroid - Using "min over the last 3 games", each server would need to keep only 3 numbers (one per game), and then only send the smallest of those 3 to the client when requested.

@Ghostthree3 - I agree that is a problem. However, hitches are nonetheless a problem, and should be reported as such. If all servers suddenly showed a performance of under 50 as a result, wouldn't that be telling indicator as to a systemic problem? Alternatively, if half of servers showed this, then you could look at other factors, such as server config, player count, map, etc.

@Omega_K2 - While average would be nice, typical averages would almost always be 95+, because usually the server is usually at near 100% performance. The problem is when there are server stutters, etc. which drop performance to <75 for a few seconds every so often. "Average" hardly reflects this. (That said, I do agree that average would be nice to have and helpful, but I did not suggest it due to the extra overhead necessary to implement).

SamusDroid

This would just make server performance worse

xDragon

If done right it wouldn't have any impact on server performance, but seeing as the tickrate('performance') is part of NS2's query responder there is no easy way to fix this that's not engine level.. so its doubtful it will get changed at this point. It could also be done using averages and a custom server key in lua, and since the server already samples its performance every 60 seconds there would be little to no perf impact. You could also just look for frames that exceed a certain delta time and keep a running total against some constant.

Racer1

@xDragon Ahh too bad it can't be a mod. I agree that if done right it would have no performance impact, especially since the calculation only occurs once a minute.

Right now, with 99% of servers with a "100" performance, the metric is basically useless.

Omega_K2

Racer1 wrote: »

@Omega_K2 - While average would be nice, typical averages would almost always be 95+, because usually the server is usually at near 100% performance. The problem is when there are server stutters, etc. which drop performance to <75 for a few seconds every so often. "Average" hardly reflects this. (That said, I do agree that average would be nice to have and helpful, but I did not suggest it due to the extra overhead necessary to implement).

I wasn't clear, I had an avarage over the last minute or so in mind, in that case you'll notice drops more easily.