Hacker News new | past | comments | ask | show | jobs | submit login
Ask HN: What is the hardware of Hacker News?
90 points by r2dnb on June 12, 2016 | hide | past | web | favorite | 38 comments
I'm curious about the hardware used to run Hacker News. My guts tell me that despite its popularity it is running on modest hardware, but I'd like to confirm.

It's (edit: almost) the same since Nick posted https://news.ycombinator.com/item?id=9222006.

I'd be curious what AWS/DigitalOcean setup would be able to run HN. Any educated guesses?

Playing around with the calculator, you're roughly on track with an r3.x4large - most of the Linux options happily bundle far less RAM.

Getting 9TB of EBS on top of that puts the bill at $1516 per month. None of that accounts for traffic, of which we know nothing about.

The thing is in an Amazon world, you could optimise the setup for an Amazon deployment. For example, you'd probably move parts of that data, like database snapshots, onto S3.

Those costs are still nothing to sneeze at, and very likely suboptimal.

3 months ago traffic was about 3 million page views per day.


It just goes to show that cloud prices are somewhat of a ripoff if you have the knowhow to use CDNs and setup a single server in a colocation center. That machine can be built for less than 5k.

Of course -- if you know your traffic well ahead of time. If you start growing quickly, it's not quite as simple. (And if you don't have the traffic/growth anticipated, you wasted a bunch of money).

You also aren't accounting for redundancy. If the machine totally blows up, how quickly can you build another? Or do you spend +5k to have a second sitting as a standby? If the hardware does fail (whether it's a single drive in RAID or the whole system), who fixes it (or deals with getting it fixed)? For that matter, is the person who built or sourced the hardware's time worth $0?

Cloud IS more expensive, but it's not as simple as comparing to the up-front cost of one piece of hardware.

This is true but cloud can still be a redundancy mechanism for a colocated server.

With EBS you can't just match size, since the IOPS and latency matter far more.

What, 9TB disk space? And 128GB RAM?

How much of this is actually utilized? O.o

Most of it is redundancy. Either duplicate copies, parity, or previous snapshots (off of which the backups are based).

I think the whole HN dataset is well under 100GiB, and maybe even under 30GiB.

So in theory you don't need any spinning rust, it should all fit into RAM ;)

I like my storage to persist across power failures :)

That is an absolutely beast of a motherboard. 1.5TB of ram!

As an aside, how do you do to find such old posts ? Is there a search function I haven't heard of ?

To add to what CarolineW posted, I remembered that the comment existed so I went to HN search and entered "by:kogir gc". (I then had to manually switch the results from stories to comments, though.)

I'm assuming you know of the search box at the bottom of every HN page - what more do you think is being used? Why do you think something more is needed?

As an aside, I went to look at some of your old posts and got this:

    Secure Connection Failed

    The connection to the server was reset
    while the page was loading.

        The page you are trying to view cannot
        be shown because the authenticity of
        the received data could not be verified.
        Please contact the web site owners to
        inform them of this problem.

Thanks a lot, no I never noticed the search bar.

Thanks the website has been taken down.

what is the connection it is being served on?

Does it use the Racket webserver?

wait.. how much RAM? 8x16 = 128gb?

Yep. Most used by the ZFS ARC, the rest by Racket and HN itself.

Thanks a lot

I'm curious about the total byte size of all HN content.

The mirror I keep of all HN items (posts, comments, jobs, polls, etc.) in JSON format is at 5.32GB uncompressed.

Cool! Did you mirror everything from the API? Does it go back to the first test posts or whatever? :)

Yeah, it came from the API which goes back to item id 1. It's all here if you are interested: http://silo.lukeheuer.com/mirrors/news.ycombinator.com/ Post items are organized by month so you can check out the byte size progression over HN's lifespan.

IIRC, 2 servers - 1 production, 1 standby, running FreeBSD.

Detailed specs have been written before, try searching.

Hrm it's behind a web app firewall (server:cloudflare-nginx) so harder to tell.

My bet is on AWS, but I'm also rather curious about this.

You're rather mistaken my friend. Hacker News existed a long time before AWS. It has 2 servers - 1 production, 1 standby, running FreeBSD.

FreeBSD is the OS of choice for them.

Perhaps I misunderstand your remark, but AWS (2006) predates Hacker News (2007). According to Wikipedia, that is.

EC2 was only in limited beta back in 2006. It wasn't in full production until 2008.

While we are at it: what language is HN written in?

I believe it is Arc, a lisp dialect that Paul Graham developed


Is Arc used for anything else besides powering HN?

Not really, it is still being "developed."[O] But some HN clones or forums built like HN use it.

[O] http://arclanguage.org

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact