Jump to content

So, funny story...


You are about to reply to a thread that has been inactive for 1809 days.

Please take a moment to consider if this thread is worth bumping.

Recommended Posts

I've been playing on a private MMO lately that supports about 8000 players. They rent 5 servers. 2000 people per server plus a login server. Two of the servers are often full with a queue. 

Monthly cost for one of those servers, at 8 cores and 32 gigs of RAM, is 160 USD per month. Often dozens if not hundreds of people clustered in an area with little to no server lag. 

Gives some perspective when you're looking at 100 people lagging around for 295 in SL. 

Link to comment
Share on other sites

can do it for even cheaper. I have a OpenSim server running on a laptop with 4G RAM and a i3. Which can handle 2300 agents simultaneously. The server-side agent handling is the easy part

the only hard part server-side is physics simulation and maintaining client interaction with this. When hand the physics simulation off to the clients then it gets even easier.  As from the server's pov it's then not any different to a combined chat and page server

the real world issues that requires answers aren't caused by the servers really. The issues are caused by the client devices, particularly assets individualised by users. Asset management on the client is the hard part when have a streaming asset service which the client interacts with

Link to comment
Share on other sites

SL region performance is constrained in a few ways.

These numbers are silly .. but you get the idea. Just counting the avatars (no stuff), each one generates a certain amount of data, multiply that up for all the avatars and that's a lot of UDP traffic per target avatar. For simplicity sake, we have 40 people, each generating 1kb per second of updates (movement, etc), means one avatar needs 40kb per second to see everyone move, that's 1600kb per second total for the region. Lets hope this isn't a shopping event with 80 people on 5 regions. 80 avatars x 5 regions = 400 people .. 400kb of data per second to each one .. x 400 people = 160,000kb per second.

How many regions are on this one sim, how much network connectivity does each one get, is there any quality of service voodoo in play .. probably.

What happens when you have 5 people wearing animated tails that use alpha switching at an event with 80 people?

Have we forgotten that the region has to also manage a list of everything every individual avatar can see, and all the other stuff like chat, and stream all that on top of the basic update data, yup!

48 minutes ago, Mollymews said:

I have a OpenSim server running on a laptop with 4G RAM and a i3. Which can handle 2300 agents simultaneously.

It might be able to handle that many, but can you send the required amount of data down the pipe .. 2300 agents .. 1kb each, 2300kb needs to be sent to every avatar per second to keep them all upto date ..that's 5,290,000kb per second total. Before you have sent a single avatar any information about anything they can actually see.

I'm going to say, fat chance. 

58 minutes ago, Mollymews said:

the real world issues that requires answers aren't caused by the servers really. The issues are caused by the client devices, particularly assets individualised by users

This hot mess makes the serve-side juggling look easy, but for realism, lets do it on a region with 200 vendors.

Is the lag .. 

On the server?

On the client?

On the cheap router your isp sourced from the lowest bidder?

On your wifi / ratty old cheap cat 5e cable snaked around the edge of your room ?

On your computers $2 integrated network adapter?

  • Like 4
Link to comment
Share on other sites

4 minutes ago, CoffeeDujour said:

It might be able to handle that many, but can you send the required amount of data down the pipe .. 2300 agents .. 1kb each, 2300kb needs to be sent to every avatar per second to keep them all upto date ..that's 5,290,000kb per second total. Before you have sent a single avatar any information about anything they can actually see.

I'm going to say, fat chance.

it's the number of requests at any given moment that are the issue, not so much how many connections there are. There can be up to 65535 simultaneous TCP connections per server. Requests are queued and it takes time to process them. The bigger/better the hardware/pipes the faster the queue is processed and responses sent

Link to comment
Share on other sites

7 hours ago, Gadget Portal said:

Incorrect. 32 gigs of RAM is 32 gigs of RAM. An i7 is an i7. I'm talking purely about renting hardware. 

.. for a platform where everyone already has a copy of all the assets, all the physics aer client side, all the server has to do is play piggy in the middle with a very optimized agent information protocol and keep everyone on the same page.; And that's if all communications go via the server.

You are comparing a sports car to a municipal dump truck and suggesting they are similar because wheels are still round.

  • Like 7
Link to comment
Share on other sites

17 minutes ago, CoffeeDujour said:

.. for a platform where everyone already has a copy of all the assets, all the physics aer client side, all the server has to do is play piggy in the middle with a very optimized agent information protocol and keep everyone on the same page.; And that's if all communications go via the server.

You are comparing a sports car to a municipal dump truck and suggesting they are similar because wheels are still round.

Still irrelevant (although in many of todays games almost as much information gets exchanged as in SL, including custom textures, dynamic objects, user made scripts, character data, chat, and so on).

I'm talking about the hardware being rented. It doesn't matter if it's hosting pictures of your grandma's dog. Everything I can find indicates that SL servers are running a fraction of what I listed in my original post, and hosting multiple simulators on them.

Edited by Gadget Portal
  • Haha 2
Link to comment
Share on other sites

This like comparing world of Warcraft to second life but imagining both running on the same hardware. One is a client side installed game where every modification and every character type, effects and physics are all already stored and downloaded ready to go. On a game like this the data being sent is minimal by comparison as most of it will be in the form of text updates to show movement or spell effects etc etc.

The other is mostly client side but rather than being pre downloaded it's constantly downloading and uploading. A spell effect in SL requires the textures, particles, physics, script effects etc etc to be downloaded together as opposed to being readily available. Those who have said the comparisons are basically asinine are correct. SL has to download and upload at any given time to show the end user the changes, modification, new people etc etc for that particular region. 

Even if you compare SL to 7 Days to Die which is a custom mod heavy game, all of the game data is downloaded before you enter the world so all the "mods" are already available on your computer for that particular server. You are also discounting that in most MMORPG games the models used are optimised in ways models on SL just are not. You could be wearing a pair of pants right now that has millions of polys where hundreds would do.

Its not really a comparison. Look at Sansar, HiFi, Sinespace they are even newer and they still suffer from some of these issues as they are asset heavy worlds that handle their assets and data differently to your average thousands of players MMO

Edited by ItHadToComeToThis
  • Like 3
Link to comment
Share on other sites

7 hours ago, Gadget Portal said:

Everything I can find indicates that SL servers are running a fraction of what I listed in my original post, and hosting multiple simulators on them.

The only information we have about LL's data centre is so out of date the machines referenced will have been scrapped by now.

The SL service is not just the bare region, and that alone would not give you a usable end product. SL has billions of assets, all of which are available to users of your region via the CDN.

  • Like 3
Link to comment
Share on other sites

On 5/10/2019 at 10:19 AM, Gadget Portal said:

32 gigs of RAM is 32 gigs of RAM. An i7 is an i7.

gettyimages-805054770-170667a.jpg.1cde26bf2119cec73995f90408895789.jpg

Well, there's a Bloomfield i7, a Lynnfield i7, a Gulftown i7, a Sandy bridge i7...

download.jpg.c42cb56d3261642f4c0a8846d9bf1ccc.jpg

.. a Kaby Lake i7, an Amber Lake i7, a Coffee Lake i7, a Whiskey Lake i7...

And then the RAM. There's the 400Mhz/3200Mbps DDR SDRAM, the 800Mhz/6400Mbps DDR2 SDRAM...

 

Edited by Arduenn Schwartzman
  • Like 1
Link to comment
Share on other sites

You are about to reply to a thread that has been inactive for 1809 days.

Please take a moment to consider if this thread is worth bumping.

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
 Share

×
×
  • Create New...