Thank you for shedding light to this key topic, @cryptographic .
As actifit witness, we had our witness running previously on hetzner back in steem days, but as of hive, we moved it to privex team, their service is quite reliable. 
We do not prefer running on our own home connection as the potential for Internet loss is at stake, so from a service continuity and reliability basis this is more optimal. 

If someone is kind enough to direct me to instructions on how to run a node. I will gladly build one and run it.

I'm running two witness servers (primary & backup) on PCs sitting right next to me as I write.
I'm in the process of spinning up another backup witness on an older MacBook Pro which will provide battery backup and the ability to switch to other internet (wifi / 4G) if my home fibre internet was to go down.

Once I get that working I'll convert my primary witness to an API node and Hive-Engine Node.

In terms of reliability my main issue has little kids pressing buttons and rebooting the PC but I've changed the settings now so the buttons don't do anything. 

Since I got the backup working 2 weeks ago I've only missed one block.

I run 4 witness nodes in 4 separate locations/countries.

2 are hosted in 2 different Cloud Service Provider's datacenter, mainly because they provide reliability and high availability for power and connectivity, with SLA. This is a key factor when you are a top 20 witness.

The 2 others are hosted in different secured professional IT rooms, which I have physical access to, one of them being under my sole control and ownership.

1. I won't be able to find reliable internet connection at homes in India and thus will not be possible for me to host the witness on own hardware.

2. Getting high availability (different from decentralization) is quite easy with available software as @@blocktrades had mentioned.  Find multiple locations, get a DDNS, get a heart beat tool (there are many) + get your domain name registered in say https://handshake.org/ and done.

Both internet connections I have access to are abysmal (10 mb down 3 mb up each) - therefore I can't host my server at either of those places. Fibre/faster internet is not available in either place either, so I can't do that.

My witness node is currently hosted on a Hetzner Dedicated Server which I do not have physical access to unfortunately (but at least it has decent internet)

Ich habe eine solide Internetverbindung, so kann ich der Server zu Hause aufstellen , das einzige Problem zurzeit sind die Stromkosten  ;)

I rent a server from Privex and have full faith that if something did happen there, @someguy123 and his team would be on point and handle it with discretion and swift action.

I don't have physical access to my witness server, since my internet connection is not reliable and/or fast enough (I have a 20/5 Mbps connection with a 300 GB/month limit (it's capped at 5/1 Mbps after)) to run it at home. 

If something extreme happened and I couldn't host my witness at any server provider, I could get a witness up and running on my personal PC in under 3 hours, and I could get my own hardware & better internet connection in about 1-2 weeks. 

But unless something really extreme happens,  I prefer renting servers since it's much faster to upgrade/switch to a new provider and they have more reliable power/internet than most households (and it's cheaper).

An article worth passing along...
@newparadigmtt, @eyeofthestrom, @fulltimegeek

> _"I firmly believe that all the resources we currently have available that are not dealing with mission critical issues should be immediately directed towards this urgent task of getting our witnesses set up on independently owned servers to which those witnesses have PHYSICAL ACCESS"_

That doesn't make much sense. 
There are pros and cons related to running a witness node "on-premises".
In most cases, for a top witness, running a node that way is might be a bad idea.

> Are you currently running your witness node on hardware you personally own and have physical access to?

Yes. No. Maybe.
I have variety of nodes (some of them are witness capable, i.e. backup witness nodes):
I own them, I lease them, I rent them, I have them on-premises, I have them collocated in data centers, in some cases I have direct physical access, in some cases I can have a physical access with some time and effort, and in some cases I can't even get close to them.

My current scenarios:
| Owned | On-premises | Physical access |
|:-----:|:-----------:|:---------------:|
|   0   |      0      |        0        |
|   0   |      0      |        1        |
|   0   |      1      |        1        |
|   1   |      0      |        0        |
|   1   |      0      |        1        |
|   1   |      1      |        1        | 

None of my witness nodes run on AWS, GCP, or Azure (but it's nothing wrong to run them there).
And I don't touch Alibaba Cloud even with a poop-stick.

When I saw the mention here I figured this was going somewhere else. That IPFS document was unacceptable to have been passed and promoted as quality by Jack from Twitter without further review. I understand how it happened but it still pisses me off. We can't undo it and I'm not about to cry over it publicly or go back to them with corrections. Hive has everything going for it.

Regarding the witness question: We've always been on our own various servers. During the HF and immediately after we were cycling through our 5 different backup witnesses + main (although having them all crash in sequence in one day was not fun). Now we're down to fewer due to the need to upgrade the servers. But they are all decentralized, hosted in different areas of the world. The reason for that is lessons learned during the blackout in the early 2000s that took out the power grid in our area and kept it off for about a week. We also realized the difference in datacenter upkeep and tech physical response time between countries and services with the pandemic. 

Having physical access to the main witness is not a good idea for most people. That would mean it's running out of their house or out of their office or a local datacenter. We tried it (local datacenter) and it was sub-optimal, mainly due to sharing the same infrastructure issues as the datacenter. Backup machines, however, can run out of a house or some local building very well. 

The issue with some witnesses is that running a node is challenging, paying for it is challenging, and some like to start off with the 'in the box' step by step guide or to buy a ready-made one from Privex (who does a great job managing them). Where some witnesses upgrade once they learn more others just become complacent and shift their attention elsewhere. None of this is a simple matter to go through.

(Just to be clear here; a witness is not one server. It's one main server and at least one backup server or secondary server, usually more than one. It takes a long time to replay. You need to manage snapshots. There's a lot involved there.)

My witness nodes are running in datacenters where I don't have access... but what I have is a stable network and redundant electricity and these are major factors for running services 24/7.

I kind of sort of agree more is better, but honestly only 21ish witnesses are actually getting used at any given time.  API servers are a different beast, since you actually have to point your applications to them.  I'm more worried about the 21 witnesses we have voted in not being able to sustain a DDoS, and how it would play out if multiple top witnesses are DoS'd.  I whole-heartedly agree that any top witnesses running on underpowered or vulnerable hardware need to be weeded out.  At some point they'll have to cross the threshold from hobbyist to institutional grade.  We need the witness votes to age out to keep constant pressure on them to stay ahead of the curve.

I am preparing to run 1x Witness node and 1x RPC node from performance hardware (on-premises) connecting through Starlink (With secondary ISP on Optic Fiber for backup). 

Hive.StarlinQ.Space 

I would say I am within weeks/days to have it up and running. As I am following available documentations, I am still facing roadblocks, so I take note of what needs to be addressed for more clarity.  Lack of documentation should not be a bottleneck for Hive's distribution. I am part of a team designing complex strategies and Hive has been identified as a key component of a Global Space Force strategy. Plans are under development to streamline Hive's evolution. To be continued...

Yes

Although there are big advantages to running witness servers with experienced and reliable hosting providers: DDoS protection, network uptime, power uptime. As long as there is sufficient distribution of witness servers across hosting provider companies, I think the network is more robust using top tier professional hosting services.

Thanks for the great summary about the current situation here on HIVE. I am currently running my witness on a server I do NOT have physical access to. But it is running on a small data center and not on one owned by a big company. So I don't really see any shortcomings in relation to running it on a server I have physical access to. Additionally I have a backup witness running on my computer at home, in case the other server has problems, or if the other server should be shutdown for some reason. I am using the server as my main witness, because of the much higher reliability. Also if someone was to attack my witness server, I guess they could also be attacking it if it was running at my home. Once again - I don't really see any shortcomings running my witness node on a small data center owned by a small company, in relation to running it at home. What do you think? 

Hosting a node at home is not ideal, in my area broadband connections are crappy, we have several blackout a year due to weather conditions. Renting server in a data center in the country (Australia) is very expensive compared to EU or USA.

We just need more witnesses with more diversity of providers. But a backup witness is not paid well enough to attract new witnesses especially when going up the ranks is a slow process.

I have my own server sitting next to me that serves as a backup, and my main server is being hosted on Privex. Both work very well, but I keep paying the server bill because I like the freedom it affords me in handling things like forks and replays.

Let's see what we hear from our witnesses. Even though some are not running on decentralized physical servers they own, how can attackers take over them ? They would still need lot of hive stake to control, right ?