Goblin Squad Member
|
So GTA Online just released and of course the servers are kaploot like every online game in the history of Man. I'm not even mad about it anymore. It's become an acceptable and expected situation.
It just seems odd that in today's world a problem that has plagued this industry for a decade hasn't been fixed.
Does the industry have some technical breakthrough in the pipeline so our grandkids will be able to play Pathfinder 4 on launch day?
|
Goblin Squad Member
|
...hasn't been fixed.
In the eyes of some of the tech and finance folks, it's not broken. For so many games, there's more demand on Day One than at any subsequent time, and there's a significant cost without significant benefit in meeting that demand.
Ten to thirty days after launch, the game will need far fewer servers, connections, support personnel, etc. The cost of providing them on Day One isn't the only cost; one has to reckon with the cost of shutting down the excess capacity as well.
Our friend Xeen has it right: if one doesn't try to hype a game to the skies, and one controls access, one may find the enviable space of excess unmet demand, rather than excess unused capacity. Not many have aimed for that target, and we'll see how well GW hits it.
|
Goblin Squad Member
|
So GTA Online just released and of course the servers are kaploot like every online game in the history of Man. I'm not even mad about it anymore. It's become an acceptable and expected situation.
It just seems odd that in today's world a problem that has plagued this industry for a decade hasn't been fixed.
Does the industry have some technical breakthrough in the pipeline so our grandkids will be able to play Pathfinder 4 on launch day?
No battleplan survives the first few minutes of battle. You can test all you want, but nothing can ever fully prepare you for launch day.
What you can do if launch days give you so much angst is to exercise the ancient and nearly forgotten arts of patience and delayed gratification.... and wait until Launch Day + a week.
Blasphemy! :)
|
Goblin Squad Member
|
Another thing to keep in mind is the cost of setting up, maintaining and operating servers in addition to ensuring enough bandwidth is available from providers. This costs money. And most companies look to turn a profit. There is a balance to be struck between the Average Heavy Load and the Maximum Heavy Load. The Average Heavy Load is a more stable number to be expected in evenings and on weekends when more people are home from work/school and is the price point most companies try to estimate. Maximum heavy loads such as from Launch or other super special events are ultimately so rare that the Return On Investment in server technologies can be difficult to justify. Many Online Games just don't see the same Peak Simultaneous Users as they do the first couple days after launch. Those that do have numbers to figure out how much additional server capacity to invest in.
Most of these games are out to make money first and to entertain you second.
That being said, I like the slow ramp-up being offered by Goblinworks for PFO. Being a first-monther AND knowing the EE game is going to be fairly simple compared to the OE game makes it easy, my opinion may be different if I had to wait for OE.
|
Goblin Squad Member
|
Again, consider the flexibility the Cloud can provide. And many Cloud vendors have high speed internet connection. Server is not the only consideration. IF you only have a T1 link, it does not matter how many servers you have. Designing with option to spin up some cloud instances to handle peak usage enables putting off server purchase until needed for longer term support.
e.g. holiday times may see more traffic with players home from school or work.
Lam
|
Goblin Squad Member
|
| 1 person marked this as a favorite. |
Cloud is a great solution for some, but not all. The Cloud has become some buzz-word sold as cure-all for everything that ails us. There are disadvantages that come with the Cloud as well. I highly advocate that companies would look at it as a potential solution and evaluate it compared to other options. Every option should be considered objectively.
|
CEO, Goblinworks
|
| 1 person marked this as a favorite. |
My experience is that most companies don't have the money and time to do real stress tests. I'll include us in that group; we'll be winging it (on the assumption that our starting cadre is so small compared to the tech that we should not have any bizarre unexpected problems not detected in Alpha).
Simulating load on an MMO is a nontrivial task. REALLY simulating it is a VERY nontrivial task. (A real simulation would involve setting up client connections from all over the world, having them running on many different kinds of hardware and software, different versions of OSs and video card drivers, connecting via ethernet, wireless, etc.)
The things that crush most online services are CONCURRENT constraints with NONLINEAR response times. That is, a system service works fine with 10 connections, or 1,000 connections, but at 1,001 connections, something catastrophic happens. An easy example is a system that can cache in system RAM transactions up to 1,000 accounts, but at the 1,001st account it has to start virtualizing data to the hard drive; and suddenly performance on that system drops by a factor of 1,000x.
Until you figure out where these nonlinear changes are you really don't understand your system at load.
So in the real world, you design for the base case (what you think load will be during most of the time you use the system) and then you try to simulate that load to see if the system performs as expected. And on launch day, you expect that stuff will break.
Until and unless you have so much extra cash that you can spend ludicrous money on high-stress testing, that's usually the best approach.
|
Goblin Squad Member
|
What you can do if launch days give you so much angst is to exercise the ancient and nearly forgotten arts of patience and delayed gratification.... and wait until Launch Day + a week.
Agreed. When PFO launches I'll just be glad to have something to play. I'm fully expecting it to be buggy, and I'm fully ok with that. If I wasn't, I'd just wait an extra week or month for them to sort out the bugs.
|
Goblin Squad Member
|
@Carissa Merz - game development, like cooking, is messy, chaotic, and requires one to be willing to adapt to unexpected complications. :) The Beta period is not about bug fixing. We expect the game to be stable, not bug free. There will be bugs to fix in any MMO for as long as the game operates. What we don't expect to do is any server wipes after we begin the Beta. We also don't expect to have operational problems where the load on the servers exceeds their capacity. That's part of our core strategy of working from a small start and growing in a controlled manner.
Nov. 27, 2012 on Pathfinder Online: A Fantasy Sandbox MMO.
A post from "Goblinworks Inc" on this issue from the Kickstarter probably sums it up best.
|
Goblin Squad Member
|
There are different services that have different temporal impacts. Downloads of updates come to mind. Biggest of these is the initial load. The download site may be better situated where it does not impact game loads in either servers or the pipes to the internet. For time of expected heavy download, place that on remote download service. then transfer back to local machines when peak goes away (DNS is your friend).
Ryan is right about testing what loads servers can handle. It would even be useful under EE to reduce the number of servers to see where the breaking point is (say 1001 clients per server). Then one applies an engineering factor to back off that level, bringing on more servers as use arrives above that design point. Commonly used factors are 40%, 50%, 80%, and 90%. business case argues for higher percentage loading? The predictability/unpredictability may argue for lower numbers. The risk is also a factor. This is a game so consequences is not major risk. If this was a health system, it would argue for lower percentage.
Ultimately this is a business decision on acceptable risk and engineering implementation to match that risk. At some point there are needs for more servers or new clients are denied (business case).
And Cloud providers are not equal in cost, service or flexibility. To make most effective use of cloud for temporary load balance (whether 1 hour or two weeks) requires designing for that possibility. If you design for that capability and don't use it, there are development costs. If it is not designed in, then there can be craziness trying to add servers.
Note, if GW does stress testing where they try to break system under EE, they should warn in advance. We do know we are guinea pigs.
Lam