I’m in the process of planning to move a well settled forum to discourse and am thinking about the hosting. I do use Hetzner VPSs a lot and am quite happy with it but didn’t yet run a larger discourse. I wonder if it’s worth thinking about splitting services into multiple VPSs or go with a bigger one instead.
Forum is just shy of 1m posts, 15k users, 100k+ PI/day.
Any hints on hosting this category of forum would be appreciated.
On Blender Artists we have about 4.2M posts and 250k pageviews per day. When we migrated to Discourse 1.5 years ago we started on a medium-sized VPS on Digital Ocean, but that didn’t cut it at all. We have since moved to a dedicated server we got at the Hetzner server auction (E5-1650V2 Xeon with 64GB of RAM and 2x 240GB SSD for about 60 Euros a month). We currently use about 50% of our RAM so the database is mostly cached and this machine performs very well.
I wouldn’t recommend a VPS for your forum size, you’ll most likely be upgrading very soon.
I’m not a big fan of going dedicated with such projects. Especially with Hetzner server auction servers. Depending on the age of the server type they might not have replacement parts if something fails. I’ve had this in the past and enjoy running most of my projects on VPS nowadays. Way less downtime and more flexibility. Thanks for that tip anyways.
And it’s not so much the cost that drives me so if going dedicated I’d probably go for a regular server but splitting services onto multiple VPS is something that I’d like to investigate first for scalability. Maybe someone’s got a hint on that.
I’m pulling this up again as we’re slowly getting into moving the thing to discourse. Currently trialing the migration process. Maybe someone has additional insights / experiences in that forum size category? I was reading @sam posting about splitting the DB off to another machine, I wonder if it’s worth going down that route.
How big are your longest topics? Do you plan to support megatopics in the future (>10k posts) - it’s stuff like that which will obliterate server performance before anything else.
Well there are a few that are in the ballpark of 5k-15k posts and they are growing but there really are a few only. But interesting to hear that this is a pain point in server performance. It’s definitely an option to think about those topics and maybe close some and open a category instead to have issues discussed in separate threads.
I’m interested in this too. I’m writing a grant proposal that will include a forum that could have up to 7,500 users, though I have no idea how active they’d be. I obviously need to make a guess at how much hosting for that would be, but have less than no clue, so I would be very grateful if folks could share their experiences and costs.
If the budget allows it, you could say you’re going to spend $300/month for business hosting. There are plenty of ways that you could actually solve the problem for less than that if you get the grant.
Yes, I can certainly use a ballpark figure… the question I couldn’t answer was how big is the ballpark? Using the Discourse business hosting would be a very good move.
@pfaffman’s suggestion is very good if you don’t have any idea what your relevant statistics are likely to be.
I’m somewhat mystified why you are asking for others to invest some time when you don’t even know what ballpark you’re playing in.
Ballpark means:
an area or range within which an amount or estimate is likely to be correct.
The two main methods for estimating hosting requirements both require that you pin down a forum statistic that can be related to the drivers (e.g. CPU to process transactions, RAM size to load database) or constraints (e.g. storage space) of hosting plans. If you can’t do this then you won’t get very far with what you’re asking here:
apply an established rule of thumb for hosting
find a forum with similar characteristics to use as your model
The reason I don’t know what my stats are likely to be is that this is not my project, and the project leader hasn’t given me any numbers because they don’t know yet themselves. Of course my question is a “how long is a piece of string” question, but as you saw, Jay came up with an excellent answer despite my vagueness.
Sometimes, vagueness doesn’t preclude a helpful answer.
I’ll also point out that a lot of grant applications require you to predict the unpredictable and provide budgets based on those predictions. That’s a sad fact of how grant tend to work, so before you criticise me for not having figures to hand, try applying for a grant and see how easy it is to come up with a five-year budget without having the slightest clue what take-up for the project will be.
With computer systems, take your original figure and multiply by four because they are consistently intransigent and can’t be sweet talked into cooperating.