Toggle System Apps
Toggle System Apps
This is the home page of a channel. It is similar to someone's profile "wall" in a social network context. Posts created by the channel are displayed according to the observer's viewing permissions.
Create a Post
If you have permission to create posts on the channel page, then you will see the post editor at the top.
Fri, 23 Nov 2018 04:43:00 -0800
wrote the following
Thu, 22 Nov 2018 15:10:45 -0800
So, ok i take it that D* is a** whole lot** lighter on energy consumption than, say, Facebook etc
In reality, this thread here is a bit like comparing taking a ride in a private jet vs. catching a train. Sure, if we only compare absolute numbers, then yes, diaspora consumes less energy than Facebook does, but if we compare the relative energy usage per users? Well, not even close.
I know that some "Facebook alternatives" market themselves as great "green" alternatives, but that's all marketing woo-woo. The reality is: Federated systems (not just diaspora, and in fact, not just federated social networks) make no ecological or economic sense, as the very model they are designed on makes it impossible for those systems to be efficient in any meaning of the word.
Let's look at
how Facebook works
first. They operate in about 11 datacenters, some of them in strategic locations that allows cooling with ambient air or natural water to reduce the energy required for cooling. They spin
, including using specialized GPU boards to run certain aspects of the calculation, a bit comparable to the custom FPGAs Google users to process video. They build these appliances to achieve the highest possible performance, which allows them to handle
tasks on one given node and thus keep the total number of nodes as low as possible. Estimating
monthly active users, this turns out to be roughly 227 million users per datacenter (yes, reality doesn't work that way, but let's just roll with that for now). I don't know the actual size of those data centers, let alone the number of machines in it, but there is only so much space you can use, so it's safe to assume they are using their resources to their fullest. All Facebook datacenters built in the last couple of years use 100% renewable energy sources, and some of the newest DCs
even leverage a system to turn their hot cooling liquid to heat up nearby cities
, which you can't really beat in terms of sustainability.
Now, let's look at diaspora. Contrary to Facebook, we don't do the processing in a few central datacenters, but we ask everyone to set up their own node. That's great for control and privacy, but
in every other sense. Even densely packed, large nodes like the one I maintain (Geraspora) operate leveraging
roughly a third of the available resources
most of the time. And they have to because there needs to be a fair bit of margin whenever there is a peak of users (like on Saturday evening), or just a processing heavy task like someone exporting their profile. However, when there is no such task, the machine effectively idles with 60% of its resources. Granted, CPUs can scale down, and PSUs also reduce their output a bit, they still use quite a substantial amount of power and are ultimately less efficient systems then they could be at 100% load. Contrary to large-scale systems (my favorite example here is Google, using idling Ad servers to render YouTube videos), I don't have anything else to process on that servers during the idle time, because I can't really ask smallpod.example.com to hand over some user data so that I can deliver their streams, so these CPU cycles to go waste.
Most diaspora podmins are private people without a large budget to spend on server hardware and co-location racks, so they rent their servers from hosters like Hetzner. While those hosters are cheap, they don't sell you actual server hardware, but rather PCs (don't believe me? Look at
, taken out of
A tour of Hetzner's Samrand Data Centre Park
by xneelo (Pty) Ltd. on YouTube
), resulting in even less efficient design. Desktop CPUs are not designed for constant workloads, and ATX cases are not designed for the most efficient cooling when stacked. PC hardware is generally not built for data center use, and there tends to be a high rate of hardware failure, resulting in more hardware replacements, creating more waste. So even if we could leverage all available resources somehow, or if we assume podmins use the free resources for other projects, it'd still be a far less efficient system than any
server could be.
One could argue that diaspora is still more efficient as we don't do complex computations for content recommendation etc., but I hardly believe this holds any merit. The thing is... yes, Facebook is doing
things, but they can't leverage magic spells either, so those
computations have to be efficient. That's the reason Facebook spins custom boards and custom chips because, at the end of the day, there are only so many cycles you can spend per user to avoid the system getting backlogged forever. I firmly believe that even Facebooks complex calculations use less actual power per user than diaspora serving a single user stream.
However, that's just my experience from working with large-scale messaging systems, and there is no way to prove that, so let's skip that and look at something else. If someone posts a thing on Facebook and it would be scheduled to be delivered to everyone in the network, Facebook would, effectively, only have to transmit that item to 10 other datacenters. With diaspora, ... not so much. Even as of right now, Geraspora would attempt to deliver a post to more than 400 nodes, spawning a connection for every single one of them, doing TLS handshakes, ... we don't have to get down to protocol implementations to realize that we create a lot of overhead, both concerning computation for cryptography, and concerning actual network usage.
So, uh, if you want to use social network with the smallest footprint on the environment, well, go and use Facebook.
Link to Source