Cost of Server for 200 Avatars


#1

So according to this blog post a HiFi domain was able to run 200 concurrent users. Taking a look at the article, it says five 72-core amazon servers were used.

Looking at Amazons server specs and prices I see that a 72 core instance is the x1 server. Scrolling down to hourly rates an x1 is listed at $14.672 per hour. There are also 1 year terms with the following prices:

  • No Upfront $0 Monthly $6599.93
  • Partial Upfront $57382 Monthly $1593.08
  • All Upfront $107879 Monthly $0

According to the blog post it took 5 of these to handle 200 avatars so here are the final costs for running a 200 avatar domain:

  • Hourly: $73.36
  • No Upfront $0 Monthly $32999.65
  • Partial Upfront $286910 Monthly $7965.4
  • All Upfront $539395 Monthly $0

These are ridiculous amounts of money and this is not economically viable for anyone.

How is anyone going to make $73 an hour from 200 avatars? Charge them all 40 cents an hour to stay logged in? (assuming 200 avatars are constantly paying to be in to your domain and you are not interested in making a profit).

The costs need to be less than 1% of these prices to even be close to viable. We can get 40 avatars for each of these 72 core instances. This needs to be more like 800 for each x1 before it becomes economically viable (four instances of 200 avatars for ~$329 per month each). Probably impossible.

IMO more than $1 per month per avatar slot is too much if Second Life is any indication of what people are willing to pay your business vs. the traffic expenses they generate by just being there.

Hopefully someone can debunk my numbers.


#2

I wouldn’t mind seeing the actual statistics from the test. Are they posted anywhere?


#3

HiFi is not SL. No one is forcing anyone to pay for peak usage for a month or year. I think the purpose of the stress tests are to see what is needed to accommodate large numbers of users when an event is happening.

If the transition between types of servers can be minimised and the LOD/level of interest can be fine tuned then the ramp up and down can be done to accommodate many use cases.

If I make a … comedy club or live music venue or …um… save the banana slugs charity event, I could start off with a server that would handle less than I expected, and then upgrade from there. And turn them off when it was done.

Anyway, I usually regret posting on forums but just my thoughts.


#4

Maybe not but using SL as a metric you can predict how much a person is willing to spend on something VS the amount of time they spend just standing around using up resources.

Yeah and that could give HiFi a nice monopoly on server hosting as well.


#5

Not sure what you mean. I was talking about an event that lasts 1 to 3 hours. What exactly are you talking about?

edit: as far as I know, HiFi is not doing server hosting


#6

Ahh… are you one of the people that wanted to rent domain servers?.. oops sorry.


#7

Yeah I set up a business (at some expense) to start renting out HiFi servers and the very next week HiFi rolls out there DigitalOcean servers. That makes me very suspiscious about them trying to compete with their own users every time they see someone trying to make some money by filling a need for the community. I guess the fastest way to get them to do something is to do it yourself then they will suddenly be hard at work.


#8

On DigitalOcean you pay when the droplet if turn off (weird… so actually you have no incentives to turn it off)
There is nothing stopping anyone to offer a better hosting service, containerize it, put it on AWS etc.


#9

Wybie and the banana slugs from Coraline.


#10

Currently trying to juggle the server and bandwidth calculations on my side as well. If you go AWS, you’re paying for equipment and bandwidth; if you are lucky enough to have hardware, then dedicated bandwidth is where the cost will be incurred. There really isn’t a great way to slice it; and honestly, it comes down to what you’re offering in addition to just the privilege to be “in - a - place”.

As far as I understand, the way it works in SL is you (an end user) must find a land-baron to then sell you a portion of a minimum section of a server (subletting) that is rented directly from Linden Labs.

I don’t know. I could be wrong. Personally, I’ve been pretty open about my projected business model. Not only are you (an end user) able to subscribe for access to our server, but you also get framework bonuses that are only available on our servers. I’ve been pretty loud and proud about not only offering micro-transactions for “stuff” (digital assets found on traditional marketplaces) but also things that can be done and experienced with said, “stuff”. We must stop thinking in terms of the way it worked in SL and think about ways to generate customers in new ways. Millennials don’t care about “stuff” they care about “experiences”.

I’ll be firming up production-level OAuth endpoints before the end of the year. At that time, we’ll start taking in users and listening to their feedback. Can’t wait to hear what you all have to say. If you’re paying for a product, shouldn’t you be able to provide feedback and requests to the developers? We do.


#11

If you can generate statistics that are not based on SL then that’s great but I think SL is a treasure trove of information about how virtual worlds work. Throwing all this valuable information away is kind of a waste. That’s a little like ripping up your resume and just starting from scratch, ignoring the years you put into your career.

I agree HiFi is not SL, but HiFi is a virtual world and so is SL. A lot of people here including Philip Rosedale learned a lot from it.


#12

In essence, I have to agree to some extent.

The argument isn’t about “Well, SL and Hifi are different.” That can be summarized as “Well, XX and Hifi are different.” but if both are social metaverse platforms, it will always be a comparison point.

So here’s my 5 cents on the topic: a major advantage that Hifi has over SL is that you can run a smaller server. At this time of speaking, a simple $15/mo server can handle a good load of users (I think about 20 or so without too much happening), which is often what some people want. As a domain (Hifi) and sim (SL) owner, the cost differences if your scope of user occupancy isn’t to be maxed is major ($15 vs $225 (Grandfathered + Prim bonus)). Considering that things are still small, this tends to work out okay.

Regarding ‘max capacity’ (SL’s 110 vs 200 as per the notes above), this is there things easily change, and understandably so. The cost of SL operation could easily be viewed as linear (due to sim tiers) while Hifi is exponential due to the underlining framework (audio streaming and kinematic data being sent to more and more people as the occupancy increases). I would even argue that if we took it down to match the limits of each-other (SL and Hifi at 110), the cost efficiency still lends itself to SL.

So here are the big golden questions:

  • What could be enabled to reduce server costs on Hifi’s end or could boost the performance of low powered servers (which would, in theory, boost performance on higher end, noting that effectiveness will lower due to how data in Hifi is handled)?
  • At what point would a Hifi server match SL’s cost (which is now $249/mo for a fresh sim) and what occupancy would that be? Would this theoretical occupancy limit be acceptable to most operators?

#13

I wonder how HiFi would do running off resource sharing pools like SONM or Golem? They are actually kind of perfect for this.

I haven’t looked into Golem much, but SONM uses distributed computing resources provided by “miners”. You can run it straight from a docker file. So if you served HiFI domains on the SONM network you would only be paying for resources you used. No need to upgrade/downgrade hosting packages and theoretically you wouldn’t need a centralized server at all. Although, what I would probably do is run domain server on a server I owned and only use SONM for assigment clients. Not sure how lag would be in this situation though (or cost).