As others have said, it’s a complicated question, but if you have the resources/wherewithal to run Ceph but don’t want to deal with co-location, you can get a bunch of storage servers from Hetzner and get a much better grasp on cost over S3.
For example, at 10PB with every object duplicated twice (so 20 PB raw storage), you’d need ~90 of their SX293[1] boxes, coming out to around €30k/mo. This doesn’t include time to configure/maintain on your end, but it does cover any costs associated with drive replacement for failure.
I’ve done similar setups for cheap video storage & CDN origin systems before, and it’s worked fairly well if you’re cost conscious.
I don't think you're in a bubble, but we've recently started evaluating Coder and have found the switch to a cloud-based IDE (especially when you add Progressive Web Apps to the mix for native keyboard shortcuts) has been extremely attractive. I'm finding myself more and more drawn to hosted IDEs where I don't need to worry about network performance or how the Docker VM on my Mac is eating up my battery life...
None of the servers on Packet.net include any bandwidth - it's all usage based if it leaves the datacenter. Internal bandwidth is free from what I can tell.
KEYPR | Application Engineer | Los Angeles | ONSITE - Relocation Provided | Full-Time
We're bringing the hotel experience into the 21st century. Think mobile check-in, big data analytics, and IoT devices (Nest, Sonos, etc) in hotel rooms.
Our offices are located in the heart of Downtown Los Angeles. We currently have about 20 engineers across 2 locations (LA & Kiev) working on everything from cloud services to custom hardware solutions. Our stack is mostly Python 3, Django and Angular.
It's not quite the same thing, but X.IO [1] from OTOY does a lot of the things mentioned, and can even run Steam directly without having to use in-home streaming (and we can take advantage of things like NvFBC to cut down capture time).
Going that method you lose the price advantage you'd get with the spot market, but a lot of the underlying complexity is removed.
It runs through a custom HTML/JS client over web sockets with a video codec designed to be decoded with low CPU overhead [1], or a native client [2] for better performance, especially with game streaming.
For anyone interested in the space, we also have https://x.io which wraps our streaming tech up with a bunch of other stuff to make running an app as easy as making an API call.
I'm curious what performance issues you were seeing in Firefox. I'm doing quite a bit of work with the same stack, and have been able to get sub-10ms decode times with Broadway.JS in all the major browsers. As long as you use transferable objects to pass data around, I haven't had any problems using web workers as decode threads either.
I'm sorry, I actually mixed up two different experiments regarding the performance issues (I've been playing with this stuff for a looongg time and have many different prototypes). Allow my to clarify. :)
The performance issues with using Broadway.js were more "general", in that the experience varied wildly, from amazing to unusable depending on the browser/device/etc)... Not necessarily related to Websockets + Webworkers.
The Websocket issue on the other hand, was for a more recent experiment where I was playing with the idea of bridging NoVNC over WebRTC data channels.
It's not so much a performance "issue" as much as a "possible area for optimization"... Though, every time I'm playing with noVNC or Broadway.js in a FF tab on my i7 laptop, it pretty much renders FF pretty laggy in all other tabs. I imagine offloading as much of the processing to workers as possible would be the best approach to lessen the effect — though, I'm not much of a frontend / JS dev.
All I have to say is "Wow". I just switched our boot drives in us-west-2 over, and our initialization time was cut in half (~7 minutes to ~3 minutes).
As someone who launches a lot of instances based on user demand, I'm very excited EC2 has finally addressed the glaring speed issues with EBS volumes. This brings boot times in line with the original GCE boot times, which were stellar.
Of course, time will tell if the new General Purpose volumes can hold up as more users come onto the system, but for now I'm impressed.
Another vote for Digicert for corp certificates - their support team is top-notch and if you opt to chain to Entrust you get browser support back through 1999, if you need to support older devices.
For example, at 10PB with every object duplicated twice (so 20 PB raw storage), you’d need ~90 of their SX293[1] boxes, coming out to around €30k/mo. This doesn’t include time to configure/maintain on your end, but it does cover any costs associated with drive replacement for failure.
I’ve done similar setups for cheap video storage & CDN origin systems before, and it’s worked fairly well if you’re cost conscious.
[1] https://www.hetzner.com/dedicated-rootserver/sx293/configura...