You are viewing a single comment's thread from:
RE: Steemit Update
I very much agree with you @anyx that there needs to be more decentralisation of the back end and I congratulate you for launching a full node.
I also think that witnesses need to be owning their own hardware for witness nodes rather than renting it on the centralised "cloud". While @anyx's Xeon server with 512 Gb cost you CAN$16,000, its is possible to create a 128Gb witness & API server using HEDT motherboard and sub $500 CPU for around US$2000.
Owning your own witness server is somewhat problematic.
Other than bandwidth issues, (unless you have fiber based internet).
A power outage, internet issues (which I'm sure are common not only in Israel) and even a piece of hardware that stops working can make you miss far too many blocks, and cause small malfunctions all across the blockchain, not so small if you're in the top 20.
Maybe it would be best if different witnesses would host their server on different server providers, but it seems there are many providers to begin with.
Those are my 2 cents on the issue.
Thanks. I've addressed these issues in more detail here. https://steempeak.com/@apshamilton/properly-decentralising-steem-and-cutting-costs-by-witnesses-running-their-own-servers-and-apis
All these things can happen at a centralised hosted solution which is fundamentally more vulnerable because of its centralisation.
The whole point of decentralisation and having 20 witnesses and 100+ backups is that there is a quick failover to the next witness.
Also, for each witness having HEDT servers each with a once off capital cost of $2000 rather than an annual cost of $2000 forever means you can have more backup nodes and more redundancy.
It simply does not make sense to be renting when you can buy for around 1 year's rent.
A server provider usually operates in several countries, so unless someone is cutting an underwater internet cable, a witness can reactivate their server within 24 hours (much less if he/she is more experienced).
While a server in their home/office can easily be shut down for 3+ days, if their ISP is shitty in their area.
If there was an infrustrcture to allow witnesses to share their servers with each other (if that's even possible), to could allow more decentralisation without the drawbacks I meantioned. But I don't know how feasible such a project is.
Posted using Partiko Android
I’m not suggesting no redundancy. A witness should obviously have their backup server node in another location with another internet provider. Also it’s easy to have have a backup internet provider at your home or office. Everything an outsourced provider can do can be done for much less cost yourself when you are talking about single digit number of servers, which is what decentralized solutions involve.
Posted using Partiko iOS
That's unless we have someone as huge as YouTube's Linus Tech Tips on @dtube who runs servers themselves...
ned would have to give up most of his steem for any real whale to come in and start dealing with the costs ;)
While you're right, you can have redundancies like a UPS(duh), a generator for back-up power and do the same thing for internet connections.
Witnesses typically scale up over time. If you're a backup, you can front a fixed cost of less than $1000 and run it on your home internet. This is reasonably stable enough nowadays. When you get higher in the witness ranks you can deploy more nodes and improve redundancy and performance.
Missed blocks from one entity on the network won't impact things significantly anyways, we have built in redundancy at the system layer (the blockchain itself) so it really isn't a practical concern.
HEDT are a great way to go for limited API support (you can't fit all the api's, but you can serve a large amount of requests locally and route the ones you can't serve appropriately).
Notably, you can also build a consensus machine (e.g. 'witness node') for less than 1000$. My purchase price for a full api was overkill -- we can bring the costs down by optimizing for tasks rather than "do it all super fast on the same giant machine".