Jump to content

Time for LL to implement Cloud computing?


Exavor Diesel
 Share

You are about to reply to a thread that has been inactive for 4346 days.

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

Recommended Posts

Firstly, if you don't know anything about cloud computing, I advise clicking X because this won't make a lot of sense otherwise!

I think it's time land gets an overhaul on Second Life. It has used the same basic concept since its inception and it doesn't appear to be progressing with modern technologies. The best LL has done is improve server hardware over the years, but that's about it. The concept has remained the same throughout the years. 

I think it is time to jump on the cloud bandwagon. I'm really surprised this hasn't been done yet! Like many on SL, I work in IT and despite my dislike for the overused term "cloud computing", I can see how cloud computing could benefit SL. Cloud computing would allow for scalability. This is an area where SL is currently lacking, or rather, it has too many limitations & constraints. Cloud computing would allow estate owners to define the size of their region, meaning enhanced scalability and flexibility. In addition, it could also reduce the cost of provisioning. 

Larger region  = more cloud server resources

Smaller region = less cloud server resources

With the current technology in use, estate owners ofen find themselves constrained. For example, there's many popular clubs on SL (I'm referring to clubs which have dedicated regions) These clubs are often laggy* and there's nothing the estate owner/club owner can do about it. Even if they added another region, it wouldn't resolve the issue because it wouldn't allocate more server resources to the affected region/area. Cloud computing could be the answer to this. And this is just one example which shows how the cloud computing concept could benefit Second Life. 

*Not referring to render lag.

Link to comment
Share on other sites

Linden Lab stuck their toe in that water a couple years ago........it was a total failure as far as participation was concerned.  Cloud computing relieves the users of needing the software locally and the need for higher spec'd machines.........a good move in theory.  But cloud computing removes the ability to create a 3D environment that SL is built on.  A program built and designed to run on HTML or Java is going to be a compromise between true 3D and 2D.........2.5D?  I tried it when LL experimented with it.........it was ugly and looked so much like those Zyngo games on Facebook it was scary.

Cloud computer has some growing up to do before it's ready for anything like SL.

Link to comment
Share on other sites

I am not sure which part of SL would be put into the cloud that isn't already in the cloud. The textures are hosted on amazon as far as i know, and since you mentioned that you did not mean renderlag i assume it is not the renderder or client part that is suposed to be in the cloud.

It appears you are asking for more flexible landsizes, something i believe they could implement quite easily. Making a big blank canvas in any engine i tried isnt a big deal. The issues start when you begin putting stuff and people on it. :matte-motes-agape:

More scalable power on sims would be nice, but i dont know much of that is possible with the SL architecture. Heroengine has a solution that suposedly is very scaleable. It also shuts down inactive worlds down and starts them up on-demand.

I'm not sure i would call it cloud computing myself tho, as my understanding of it means that rendering and client load would also have to be in the cloud so there is no need for a strong client. I believe it's not there yet. (see Peggys post)

Link to comment
Share on other sites

Each tme I hear someone mention "cloud" I feel like puking my guts out. The concept is neither new nor in any way shape or form revolutionary. I won't even get to the motivation of encouraging people to put all their private data into "The Cloud". People who are daft enough to use FakeBook will fall for every buzzword I guess.

It's already been said: A lot of SL is already "in the cloud". Putting simulators on distributed servers would work if SL wasn't such a rigid, inflexible structure. Whether it would improve performance remains questionable.

Putting the client onto a remote server (i.e. making SL accessible from anywhere with a browser) is doable and has been done before. There's quite a few challenges for that approach: If you render the data remotely and just stream the final result, you're adding latency and are consuming a fair bit of bandwidth. For some uses, that's acceptable. For others, not so much. Because of the nature of SL, you'd also need a lot of CPU/GPU power on the remote end. That's not cheap. OnLive is trying that approach, and they're not exactly doing stellar. Blue Mars tried that approach, and they're just about dead.

Distributed computing isn't the be-all, end-all of IT. It has its uses, but SL isn't quite it. "Cloud Computing". Yeah. Right. Glad there's suckers around every corner, else I wouldn't have a job.

Link to comment
Share on other sites

Inworld assets and static web assets are on S3, profiles and wiki are on EC2.

SImulators have unusual requirements, they are really simulators in function not only name. Mainly the physics and scripts are designed for an environment that runs 24/7, a conversion to on demand is not a clean fit for existing content.

A new type of regions or even off-grid spaces might work better with that, like a home or club that only goes online when avatars are there. These spaces would have limits. Content like breedable animal simlations, remote vendor services, inworld product updaters, anything that expects the sim to be running all the time would be a little or a lot borked. Many more use cases could work great in that environment. It might be painful to explain the differences to casual customers. It totally would add a new level of support hell for content creators who build for an always-on simulation.

Link to comment
Share on other sites

  • 3 weeks later...

3 reasons why I think LL should look on running the grid over a VM.

Scalable, Cheaper and more GREEN !

 

I really see no reasons to stick one SIM per server. It purely nonsensical and a great waste of ressources. The homesteads lacks of garanteed ressources and even if they arent supposed to be used as a real SIM we aren't really at ease on em.

Mesa

Link to comment
Share on other sites


Mesarim Vaher wrote:

 

[...]

I really see no reasons to stick one SIM per server. It purely nonsensical and a great waste of ressources. The homesteads lacks of garanteed ressources and even if they arent supposed to be used as a real SIM we aren't really at ease on em.

Mesa

There's more than one sim per server. Virtualization doesn't save ressources in this scenario, it wastes them because the VM eats ressources for the virtualization as well. It actually rarely saves ressources... it's just convenient for sysadmins (in many cases anyway).

Link to comment
Share on other sites

Hmm.. sure they run multiples instances on one machine but the result is sluggish. On the other hand, with a VM you can actually run a sim with few ressources and use more only when you need it. Convinient for admins but why is there is to be any advantages in a structure that have for aim to be as transparent as possible ? Lots of improvements are made in this field. That's the future. You can even run a VM in a VM nowadays.

Link to comment
Share on other sites

You may read up on how VMs work... you're thinking distributed computing, not VM. Different thing.

While it's possible to have a VM span multiple machines, it's not technically feasible for most applications. In essence, one VM runs on one host computer. Marketers often tout "hey if you need more CPU power, just get a bigger VM!", but that's pretty much bull**bleep**. A VM doesn't magically make everything run faster. In fact, if you have a fast server, you're better off NOT running applications in a VM if you need performance.

Distributed computing on the other hand does what you describe. It allows sharing the load of an application over multiple machines, even dynamically adjusting load distribution in some cases. That kind of application only works for specific scenarios and comes with its own drawbacks, namely that you need a very fast, low-latency, high-bandwidth backplane to connect it all.

Unfortunately, SL isn't designed with any dynamic scalability in mind beyond the ability to add/remove simulators at will. In the atomic unit "sim", there's no load balancing happening. It's one sim = 1 CPU (not the same as 1 sim = 1 server) and to some extent a sim can "steal" from other sims on the same server, but to what extent? No idea.

What you imagine does sound nice in theory. In practice, it's a whole can of worms and not even remotely trivial.

Link to comment
Share on other sites

Hmhm, distributed computing sure sound nicer than cloud computing but i think i am getting a bit out of topic. It's what I intended in my naïve and (a bit clouded) mind. some kind of cluster of blackboxes on which you could layer VMs transparently.

Their assets servers are pretty much scalable already as it's nothing less than a database and a file server. As for their simulator and physic engine it would mean to recode em for parallelism (or distributed, which is another thing) and due to the fact that the network is slower than a bus, I think it would require some kind of new hardware as well. Ive heard such thing exist thou. But hey, remotely or not, that's every developpers dream.

Link to comment
Share on other sites

You are about to reply to a thread that has been inactive for 4346 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...