open All Channels
seplocked Features and Ideas Discussion
blankseplocked Discussion about fighting lag
 
This thread is older than 90 days and has been locked due to inactivity.


 
Pages: 1 [2]

Author Topic

Aineko Macx
Posted - 2009.12.19 10:06:00 - [31]
 

Originally by: tx eight
That has nothing to do with Python, given a solar system is one or more OS processes, they can be run in a VM can (that is, OS-level virtual machine, think Xen, KVM, VMware). Such container can be moved to another hw node in seconds.

There are multiple solutions at different levels for the problem. There is the one you mention, moving an OS-level VM around, but also the two I mentioned: Moving the process instances (Python VM) or using handovers to move players to a different node. With an outside view it's impossible to tell which is more feasible.

Quote:
On the issue of spreading a solsystem over several cores, I can only think of offloading all geometry (positions, velocity, collision detection) to GPU/idle CPU cores via OpenCL, and grafting a low-latency message passing IPC onto stackless.

Position and speed updates might be parallelized, collision detection only if you group the objects interacting (bumping). But unfortunately position, velocity calculation and collision detection cannot be run in parallel to fight calculation, as these parameters have a large impact on the fighting and therefore must be calculated in a serialized way.

Quote:
Unfortunately this implies a rewrite of a scale of several years. Plus blades don't typically have GPUs.

I agree, also on the problems arising from platform choice. Blades don't have GPUs, but since the trend is merging CPU and GPU, we might see that one day...

Juliette Leblanc
Gallente
Posted - 2009.12.30 04:37:00 - [32]
 

Originally by: tx eight
Originally by: Solbright
CCP's choice of cooperative threading within Python is the real stopper here, not the language itself.


I think this was and still remains the only choice that allows hundreds-thousands of interacting agents with acceptable latency and code complexity.
No, it isn't. There are several languages that support this and Snow Leopard on the Mac even added a similar facility at the OS level.

But when EvE was designed the "hardware trend" was to have a single execution engine with ever increasing speed. A few years later Intel changed the rules and started a trend where processor speed was actually reduced but you had multiple processors and each processor had multiple execution cores.

EvE however had already been developed.

ElderFather
Posted - 2009.12.30 16:54:00 - [33]
 

My thoughts are related to the lag issue but not on the server/client hardware side but a software patch for lag issues while undocking.
My PC lags and I have lost 2 ships on trade routes while undocking from station; (I'm webbed and scrambled, I have lost my shields and my armor is red-lining before I even saw the stars in space or the pirates)!
Idea Include a feature to activate mid and low slot defensive modules before undocking procedure completes (i.e. the undocking sequence has several steps before entering space, simply add "checking" and "activating defensive systems" to this sequence; activate shields or armor modules for example).
Now the undocking checklist is complete; fly safe!

An optional fair mod for all undocking pilots is to extend station shields to a 5 km (or some appropriate) range. Now hopefully the damage to undocking ships will not occur until AFTER the client pc can see space!
This at least gives (non-pvp'ers or lagging client PCs) a fair chance.

Seth Ruin
Minmatar
Ominous Corp
Circle-Of-Two
Posted - 2010.01.13 11:49:00 - [34]
 

Originally by: Juliette Leblanc
Originally by: tx eight
Originally by: Solbright
CCP's choice of cooperative threading within Python is the real stopper here, not the language itself.


I think this was and still remains the only choice that allows hundreds-thousands of interacting agents with acceptable latency and code complexity.
No, it isn't. There are several languages that support this and Snow Leopard on the Mac even added a similar facility at the OS level.

But when EvE was designed the "hardware trend" was to have a single execution engine with ever increasing speed. A few years later Intel changed the rules and started a trend where processor speed was actually reduced but you had multiple processors and each processor had multiple execution cores.

EvE however had already been developed.
HPC and multithreaded applications have been around a hell of a lot longer than EVE. I would be extremely surprised if CCP designed their HPC cluster, paid the big bucks for the hardware, software, and support from Microsoft, but failed to utilize it to its fullest.

The recent multi-core paradigm shift is only in reference to consumer computers.


Pages: 1 [2]

This thread is older than 90 days and has been locked due to inactivity.


 


The new forums are live

Please adjust your bookmarks to https://forums.eveonline.com

These forums are archived and read-only