Goblin Squad Member
|
In "Blogs and Q&A Videos: Submit questions here" and other threads people have asked about servers. That level of discussion has not been public. Security through obscurity does not work, but in actual roll out I see this as changing. I expect in alpha this will be single computer in the US Pacific northwest.
I had proposed in private mail concepts with 3 (or 4 with hypercube core) level architecture which would support client interfaces around the world, possibly using cloud centers to adjust to demand. Interface computers connect to middle ware which isolates the core (core only speaks to middleware). Initially this can be implemented on one computer (but no security isolation) and later layers become multiple computers, but one distributed server. I do not know their Unity tool to know if that is possible. molding function calls as sockets does not work, but messages as asynchronous sockets does.
Having a single location can have internet bottleneck which is reduced with multiple client interfaces.
A single server does not mean a single machine. Hyper cube architecture allows for data farm in excess of 2^64 (though 2^64 is a lot -- 4 ska bytes). Sun Microsystem 1990's sales slogan was something like "The network is the system.".
Goblinworks Executive Founder
|
In "Blogs and Q&A Videos: Submit questions here" and other threads people have asked about servers. That level of discussion has not been public. Security through obscurity does not work, but in actual roll out I see this as changing. I expect in alpha this will be single computer in the US Pacific northwest.
I had proposed in private mail concepts with 3 (or 4 with hypercube core) level architecture which would support client interfaces around the world, possibly using cloud centers to adjust to demand. Interface computers connect to middle ware which isolates the core (core only speaks to middleware). Initially this can be implemented on one computer (but no security isolation) and later layers become multiple computers, but one distributed server. I do not know their Unity tool to know if that is possible. molding function calls as sockets does not work, but messages as asynchronous sockets does.
Having a single location can have internet bottleneck which is reduced with multiple client interfaces.
A single server does not mean a single machine. Hyper cube architecture allows for data farm in excess of 2^64 (though 2^64 is a lot -- 4 ska bytes). Sun Microsystem 1990's sales slogan was something like "The network is the system.".
Security through obscurity might not be effective, but security through publicity is worse.
Each additional layer adds latency to the overall system, which is bad. I'm also confused about what security is gained by isolating one system from the others by means of a limited interface- surely if the computer running the middleware is compromised, then a MITM situation exists for all users.
|
Goblin Squad Member
|
Any computers compromised is a problem. WIth IP filters and other features middleware talk to interface and core; and core talks to middle. Middle is a bastion that can only best be attacked when clients are overtaken and the nature of the protocol determined and holes found. Once one interface is understood, any middle layer is at risk. But that is notice that a problem exists. Layers do add latency to the extent that unique data must be communicated. Local data is held locally and used local while relayed upstream.
Brute attack of middle layer can be a problem but only if IP connections are allowed. Many cloud models will refuse connections by IP at the service firewall. THis does become an issue for the cloud provider, but they do have resources beyond what you can have.
And the size of the pipe is another form of latency. Having everyone on one pipe can cause other issues as I saw a long time ago (one computer using 2/3 of the company interface in what was then thought to be a big interface). The client interface does not need to have the whole model, but needs to update within themselves and pass to middle. Middles may only have subsets. Updates of nascent middle data to core need to be considered. Does core need every hit point or can this be handled at middle ware. THis argument may be too late for this tools. But subsets of this may work. I DO NOT KNOW YOUR TOOL. That is not important. How has dev planned for more than one interface but still one server. I DO NOT NEED TO KNOE. I expect you have looked at this. It may be that the interface do not even need to be out side UP, but they should have independent wide pipes to different vback bones.
Secure sockets is another latency. but is improve you security.
What is known this is classic problem. Within this problem domain, not all servers need to know all data. One can envision client servers divided by user domain of hex domain. THis needs to be modeled by use. middleware needs distribution around the net so as a character moves from a core 37 or 55 hexes to another with overlap they are handed to another interface or middle server (how and when you do this is drivern by load).
The issue is not exactly how you initially implement (though 3 layers in on foodie is not all that expensive). Does the design expand to multiple patters and multiple servers as load goes. Core should be about capturing what is REAL. Middle is about ditrubutin parts of real differentiate by HEX, or lat/lon. interface is dispersed interface on hex and or lat lon (can you afford multiple hex interface servers in Asia or do you have differing hex servers aroung the work, each with robust pipes.
If you think that you are safe because this is not public. That only survive your first ALPHA connection.
And security through publicity, imagine 4000 people trying to tell you where the holes were in your code, vs one toon working his way in an not telling you.
|
Goblin Squad Member
|
Tricksee,
There is the potential for many different types of security breaches, so increasing the scale globally increases the complexity and required security overheads (theoretically.)
Having a single "server" or server farm reduces complexity and overhead initially but does introduce a single-point-of-dependency which is inherently fragile.
I assume that they will provide the best security they can at the physical and network layers, look at some appropriate safeguards at the application layer and throw a beasty layer 3/5 firewall for the in-between.
I also assume that their code integrity would be the highest priority, followed by server availability with hacking or cheating down the priority list, but I have been known to be wrong in the past ;-)
Hard to say how they are handling it though as we have little information, but a useful conversation to explore options.
