AirenSoft / OvenMediaEngine

OvenMediaEngine (OME) is a Sub-Second Latency Live Streaming Server with Large-Scale and High-Definition. #WebRTC #LLHLS
https://airensoft.com/ome.html
GNU Affero General Public License v3.0
2.54k stars 1.06k forks source link

Discussion: Hosting providers for OME #481

Closed alexthedamager closed 1 year ago

alexthedamager commented 3 years ago

I saw a lot of mention of DigitalOcean in #468 and wanted to mention a few possible alternatives for OME users.

Keep in mind that if you are looking at the built-in GUI CPU usage graphs on Linode or Vultr - they report CPU usage per core (ie. 100% usage on a 2 vCPU instance = 50% overall usage), whereas DO's built-in graphs report overall CPU usage (ie. 100% usage on a 2 vCPU instance = 100% overall usage).

Does anyone know if there are hosting providers that have guaranteed network out speeds?

@fcqpl know you were looking into low-cost hosting options, so you may be interested in this.

mpisat commented 3 years ago

I got a dedi server from ikoula for 44.09 EUR a month after 30% lifetime discount, I7-9700K (8 real cores, Quicksync enabled!), 960GB NVME, 64GB Ram. Network speed is pretty good in the EU.

MoZyo commented 3 years ago

OVH :)

basisbit commented 3 years ago

OVH: Beware traffic costs: 100$ per TB for their Singapore servers.

Vultr: Beware traffic costs: 10$ per TB for North America and Western Europe; 25$ per TB for Singapore and Japan; 50$ per TB for Australia and South Korea

Hetzner: 1$ per TB of traffic after using the initial 20TB of free traffic included with each VM per month, or use a dedicated server with "unlimited" included traffic. They occasionally kick very big customers if they don't play fair. Typical reliable (lowest) network speed for Hetzner Cloud VMs: 400Mb/s, for their dedicated servers: 650Mb/s

IONOS: unlimited free traffic, but their VMs have very unreliable network upload speeds (often below 150Mb/s), do not use. Their dedicated servers (hourly billed) have mediocre network upload speed stability up to ~ 500Mb/s

DO: 10$ per TB of traffic for not-included traffic, or 5$ per TB for included traffic (order bunch of unused shared resource VMs when you are close to the end of the month to turn 10$/TB traffic into 5$/TB traffic), and all their servers have very unreliable network upload speeds, especially in Asia and USA. Sometimes 800Mb/s, sometimes 150Mb/s. Upload speed is independent from "droplet" size, thus I'd suggest using many smaller instance sizes instead of a few big ones.

Linode: 10$ per TB over too much traffic usage, but they kick customers who use a lot of traffic on a lot of servers. Also rather expensive for the amount of CPU resources.

Also, avoid those companies who just rent colo space at big data centers and then resell that. They usually have too much oversubscription on their resources, or don't have good peerings with the big T1 ISPs.

stale[bot] commented 2 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

naanlizard commented 1 year ago

@basisbit do you have any recommendations for an edge server with moderate bandwidth usage (~30TiB/mo) in the SEA/Asia region? Data seems ridiculously expensive in the region

basisbit commented 1 year ago

Not really. Maybe try Oracle Cloud or Vultr or try to find a dedicated server provider that bills traffic per TB and which buys enough transit from (preferably all) big Tier 1 providers of that region.

mpisat commented 1 year ago

@basisbit do you have any recommendations for an edge server with moderate bandwidth usage (~30TiB/mo) in the SEA/Asia region? Data seems ridiculously expensive in the region

Singapore has the best connectivity to SEA but don't expect cheap BW :) Maybe you get 3 of those for 30TB? https://www.ovhcloud.com/en-sg/bare-metal/rise/rise-apac-1/

basisbit commented 1 year ago

Sorry, but OVH's Singapore location is a pita to use. I do not recommend it. Also, their pricing per TB is just a rip-off or them saying they don't want customers in that region.

On a side note, I also have to revert my suggestion of trying Oracle - their business processes are just absolutely terrible, especially regarding sign-up process, trying to get quota canges to be able to create some more servers and trying to get any support that doesn't just tell you "Sorry, but we can't help you". It seems Oracle just doesn't want people to actually use their service.

With Vultr, the problem seems to be that they randomly kill your VMs when you start to actually use them a lot, and their support does not know why the more than just lightweight usage gets falsely detected as DDoS.