30k concurrent players RO private server, is this possible?

[Pyromanix]

Hi all.

Recently, one of my work colleague claimed that back in 2005, he hosted RO private server, with at least 30000 players online at the same time. I find this very hard to believe, but some other of my colleague said it's possible because RO bandwidth requirement is very low...

Is this truly possible? Can Hercules or any other app support that many players?

What kind of server hardware needed for this?

Appreciate any input. I personally feel he is bluffing, but I lack any facts to say otherwise

 

[jaBote]

I guess that's possible since processor, memory and bandwidth requirements are (your friend's right on this one) quite low. Supposing 10KB of memory required for each user (don't actually know but it seems a pretty big estimation) that sums 30 MB of RAM needed just for handling the users, while just the map server on idle state used to consume around 500 MB RAM at that time. The bandwidth and processor requirements depends on what the people are doing on your servers.

What I find hard to believe, is to be able to gather that much people...

 

[Pyromanix]

Hmmm, while disregarding how can he get so many players to his server (if 30k players are online, actual number of subbed account going to be in hundreds thousands), the 1st thing that I find unbelievable is the bandwidth...

How much are bandwidth required per player if all of them are idle?

By the way, he said all the players are online mostly in few areas during Woe..

 

[jaBote]

How much are bandwidth required per player if all of them are idle?

I've never gone through the (quite complicated for me) client-server connection, but most client-server packets are just a few bytes in size. Dunno if a rough estimation of 1 kbps per user would be acceptable. You don't usually need bandwidth for most games since resources are almost always on client's side, only thing you need is speedy connections.

The thing that makes me more dubitative is, actually, the required processing power since it increases exponentially (if not factorially) depending on the amount of users connected on a same amount of maps and whatever they're doing on a given moment. And also, that the client side program lags if there are way too many sprites at once on screen.

 

[Pyromanix]

CPU requirement would be quite huge, not sure if that Pentium 4 is up to the task 

I haven't played RO for very long time, but from my short experiences just 30-40 people on screen alone is already start to tax on my frame rate. Especially in WoE when everyone is spamming meteors and such.

Which bring to the question, if the number of players increase, isn't server bandwidth also would increase exponentially ?

e.g let say there is 100 players in one area. One player AoE and hits 4 other players. The server then would need to inform other 95 players of the damage done, remaining HPs, etc. What if then actually there are actions done by more than one players affecting others at the same time. By right, the number of updates need to be sent out by the server would increase tremendously.

Thousands of players in combat in a single area with very high number AoE elements... I wonder and doubt if consumer-level PC can actually handle it.

Thanks for your insight. Appreciate it if there is any further arguments to support or oppose this claim, as this question is plaguing me for days 

 

[GmOcean]

What you're giving as an example is actually almost nothing. You're missing the bigger issue regarding that example. The PC itself, would be able to handle that and probably 100x that easily, the issue would lie with the bandwidth not the computer's computational abilities. Truthfully, the computer would have been able to process all of that information in about 0.3-1seconds time. The only visual lag you'd see would be from the time it takes the information to be sent to the end user via internet connection. Which is where the real root of the problem comes to play. The sever, with all it's bandwidth can only send out so much at a time, however in the end it's the connection that the user has to the server that will serve as the main variable here. Other's with a fast connection, higher ping, and faster computer, will receive that update anywhere from 10-200milliseconds before someone who would have a mid-level consumer PC back then.

This all holds true with current mmorpgs. The server can only do so much, after that it's the reliability of the end user. Which is why people talk about having 60fps and 80ping all the time. Because it's not the server that's stressing, but rather your personal comp.

Main reason being, the server doesn't have to run any 3d programs which take up a lot of memory, it just has to run a few calculations, which is way easier to do. Just like how you could copy your whole ragnarok folder, and paste it somewhere, this could take anywhere from 1second - 5mins on average. Now, thats about 2-4gigs of physical memory it had to copy over. Now let's say the physical memory needed for the server to maintain all actions of player to be 10KBs at the highest peak, which is a lot. That's still 10,000x less memory needed.

In the end, you have to remember not all 30,000 of those players are doing stuff. Heck most of them could have been merchants afk.

 

[Dastgir]

In short,

Server cam handle it, but you should have properly arranged everything, like an unindexed sql table can increase query time and can cause noticable lag to players.

But I doubt about client, since even official server dont have 30k players, no idea if client is made to handle that much players (and client will lag with low end pc too , and too many effects going on in client)

 

[GmOcean]

The client can handle 30k players, however it will lag if the PC isn't powerful enough. When RO was free to play during it's Open-Beta phase, there were 3 servers, Chaos, Loki and Sakray. Each server had 30k-45k players on at a time. Then, when p2p came around, it dropped to numbers 10x less, resulting in each server only having 3-6k on at a time, 6k during WoE, 3k on average. But yeah, back then PC's were crappy, just walking through Prontera (where all the vendors were) would lag you rediculously like 1-3second lag every step, if you even managed to make a step through that crowded place lol.

 

[reddozen]

Unless he was running aegis (official server)... back in 2005, no emulator could hold that kind of load. I had to direct a lot of changes through the character server in the eAthena base around that time just to get it stable for close to 1,000 players online. Freya was able to handle around 2500~3000 before it started getting unstable.

Would be hard to test current limits. Not near as many people play anymore, but I have seen p servers with up to around 5,000 players running emulators.

 

[Dastgir]

Unless he was running aegis (official server)... back in 2005, no emulator could hold that kind of load. I had to direct a lot of changes through the character server in the eAthena base around that time just to get it stable for close to 1,000 players online. Freya was able to handle around 2500~3000 before it started getting unstable.

Would be hard to test current limits. Not near as many people play anymore, but I have seen p servers with up to around 5,000 players running emulators.

I have saw somewhere Ind mentioning his server can handle 15k IndAI, (and IndAI should do same as players)

 

[reddozen]

I have no clue what "IndAI" is, but it would depend on how the simulation is taking place. cause "clones" of players takes almost nothing, but with actual player connections hooking through the char server and database servers, you have a different situation. I'm not saying it's impossible. I'm only saying that I don't know enough about the current state of emulators to prove you're correct.

Your biggest problem is the fact that Athena based emulators are single threaded and generally spegetti code.... There will eventually be a throughput wall that will be reached ona single threaded operation. That's why there's been so many projects to abandon it and modernize it. I used to have a build that me and a couple others made of a 3CEAM source that we converted to C++ and were threadding and modularizing the source on (it compiled and worked). That's a different story though.