Hacker News Reader

Comment by zzzcpan | Hacker News Reader

Curious, why don't you compare with known "non-cloud" hosting companies too, that provide vps services, like ovh, hetzner, leaseweb, etc.?

EDIT: Anyone cares to explain his reasons behind a downvote?

View on HN

Does anybody care about the answer to your question? In all likelihood, the answer is: because that information is even more work to collect.

Your post reminds me of the burden of proof fallacy, in which you create work for other people by asking questions you could easily answer yourself.

If you or anybody posts the comparison you requested, it will surely be upvoted.

I did, over here: https://news.ycombinator.com/item?id=13074325

Thanks for posting a note in this thread - I would have missed it otherwise!

I don't believe he directly requested a comparison. What he asked was why the commenter above him chose not to. You seem to acknowledge that in your second sentence, but then go on to state that his comment is reminiscent of asking questions one could answer themselves - asking a commenter why they chose (not) to do something doesn't strike me as answerable by a second or third party.

There are also many less known VPS providers with truly great deals. I've been using these two:

NodeServ: $1.25/month = 50GB HDD, 512MB RAM, 1 core, 1TB bandwidth, location in Jacksonville.

Host.us: $6/month = 150GB HDD, 6GB RAM, 4 cores, 5.12TB bandwidth, location in Dallas.

Both deals found on LowEndBox.

Your lesser known VPS providers probably have lower tier bandwith, more crowded hypervisors, old virtualization tech or old hardware.

That's not to say getting a 8GB openVZ vps for $4 a month isn't an amazing deal, but just that there are caveats.

I just had a look through their ToS for the catch.

For the most part it's reasonable, but there's a freaking litany of reasonable things you're not allowed to run, including IRC, audio/video streaming, game servers, and so on.

Why on earth do you sell me X block of resources for Y$/month if you're going to tell me what I can and can't do with them? Surely unreasonable use would be covered by resource limits already in place?

There are mainly 2 reasons:

* they oversell, so they assume that only small fraction of web servers will consume 100% of resources, while almost every torrent client will consume 100% of the bandwidth. There is nothing wrong with overselling hosting within a reasonable margin, but most of the people here want to run more than a LAMP stack on the server. * they get too much admin overhead replying to Tor "abuse" letters etc., so they just decided to deal with it in the simplest way possible.

I guess both factors contribute equally.

Most VPS providers already limit the options to run Tor nodes or SMTP servers in one way or another. However forbidding things like IRC and audio streaming is quite unusual and I wonder how oversubscribed their bandwidth must be on these hosts.

I doubt CPU or RAM allocation is the issue here given AWS already have a good CPU time credit system to manage it.

In the old days (90s-2000s), allowing IRC bots opened yourself to being a DoS target and general receiver of harassment complaints from perceived social abuses that happened in the chat rooms. I assume things haven't changed that much.

All major IRC networks hide your IP address when you connect now (mode +x), but I bet the perception still exists. Most of these ToS'es are thoughtless copypasta from other services.. every now and then there's a Show HN from a new hosting company that has absurd nonsense in their terms, and the creator gets suitably chastised for it.

I don't really care. My $6/month deal beats any $5/month deal from the major players, and by a huge margin. I recently tested the internet speed on it, and I got 850 real Mbps out of the promised 1Gbps channel, which is good enough for me. I can give Memcached 1GB of RAM and not worry about killing everything else.

I have a bunch of sites running on it without stepping on each other, and I doubt that would be the case on AWS / Google / DO.

You're telling me that your 150GB HDD beats a 20GB SSD? Even though they're both going to sit at 10GB free for years.

All the critical things should sit in RAM anyway. The SSD will beat the HDD if you read/write to disk heavily. But not if you need space.

If I need an SSD, there are options too, though DigitalOcean is indeed one of the best if you need a cheap US-based server. If the location doesn't matter, EU, Russia, Ukraine have some great deals.

Example: $4.6/month = 40GB SSD, 1GB RAM, 3 cores, unlimited traffic @200Mbps.

https://firstbyte.ru/vps-vds/kvm-ssd/

LowEndBox and LowEndTalk are awesome, been following for years. CyberMonday got me KVM with 8GB RAM for $10.

Possibly because (except for Leaseweb), they're all European and of limited usefulness as a result. Linode, Vultr, DO and AWS all have numerous regions around the world.

They're also typically leased for at least a full month and can't be spun up/down on demand like you can with these services.

Plus they focus on large (>16GB) dedicated servers.

Leaseweb is in Amsterdam.

It's also in the US, Germany, Hong Kong and Singapore: https://www.leaseweb.com/platform/data-centers

Oh wow. They must have expanded over the years then.

That's simple. When DO came on the scene it was the only one of its kind. Now there are 4 almost identical services: DO, Linode, Vultr, and Lightsail. On the surface those seem the same because of the near linear pricing and similar allocation of resources. The ones you listed aren't even close. Each of those may have some but they don't have all of what makes the DO model so useful to some of us:

1. Mission-critical/Production ready reliability and communication (all maintenance and issues)

2. No unexpected termination of instances / Reasonable warning & mediation

3. Not overprovisioned / little concern of noisy neighbors

4. Tier 3/4 redundancies

5. Strong American coverage (each DC with Tier 3/4 level services)

6. No setup fee on new instances

7. 1-minute provisioning (simple creation of instances / no ticket needed for deleting resource)

8. Programmatic IaaS management including provisioning, DNS, and images

9. Quality resources - mostly Xeons not ARMs, local SSD not Ceph

10. Huge backing - they're not closing tomorrow & I wanted a #10

While OVH, Hetzner, Leaseweb seem like nice services, particularly for needs in Europe, I can't build an American-centric service on those, set it and forget it nearly as easily or worry-free as with DO/Linode/Vultr/Lightsail.

I expanded it with those companies:

Price breakdown vs DigitalOcean, Vultr, Linode, OVH, and Online.net / Scaleway:

$5/mo

    | Provider               | RAM   | Cores | Storage    | Transfer |
    | ---------------------- | ----- | ----- | ---------- | -------- |
    | LightSail              | 512MB |     1 |   20GB SSD |      1TB |
    | DO                     | 512MB |     1 |   20GB SSD |      1TB |
    | VULTR                  | 768MB |     1 |   15GB SSD |      1TB |
    | Hetzner (virtual)      |   1GB |     1 |   25GB SSD |      2TB |
    | OVH                    |   2GB |     1 |   10GB SSD |      ∞TB |
    | Scaleway (virtual)     |   2GB |     2 |   50GB SSD |      ∞TB |

$10/mo

    | Provider               | RAM   | Cores | Storage    | Transfer |
    | ---------------------- | ----- | ----- | ---------- | -------- |
    | LightSail              |   1GB |     1 |   30GB SSD |      2TB |
    | DO                     |   1GB |     1 |   30GB SSD |      2TB |
    | VULTR                  |   1GB |     1 |   20GB SSD |      2TB |
    | Linode                 |   2GB |     1 |   24GB SSD |      2TB |
    | Hetzner (virtual)      |   2GB |     2 |   50GB SSD |      5TB |
    | OVH                    |   4GB |     1 |   20GB SSD |      ∞TB |
    | Scaleway (virtual)     |   8GB |     8 |  200GB SSD |      ∞TB |
    | Online.net (dedicated) |   4GB |     2 |  120GB SSD |      ∞TB |

$20/mo

    | Provider               | RAM   | Cores | Storage    | Transfer |
    | ---------------------- | ----- | ----- | ---------- | -------- |
    | LightSail              |   2GB |     1 |   40GB SSD |      3TB |
    | DO                     |   2GB |     2 |   40GB SSD |      3TB |
    | VULTR                  |   2GB |     2 |   45GB SSD |      3TB |
    | Linode                 |   4GB |     2 |   48GB SSD |      3TB |
    | Hetzner (virtual)      |   4GB |     2 |  100GB SSD |      8TB |
    | OVH                    |   8GB |     2 |   40GB SSD |      ∞TB |
    | Scaleway (dedicated)   |  16GB |     8 |   50GB SSD |      ∞TB |
    | Online.net (dedicated) |  16GB |     8 |  250GB SSD |      ∞TB |

$40/mo

    | Provider               | RAM   | Cores | Storage    | Transfer |
    | ---------------------- | ----- | ----- | ---------- | -------- |
    | LightSail              |   4GB |     2 |   60GB SSD |      4TB |
    | DO                     |   4GB |     2 |   60GB SSD |      4TB |
    | VULTR                  |   4GB |     4 |   45GB SSD |      4TB |
    | Linode                 |   8GB |     4 |   96GB SSD |      4TB |
    | Hetzner (virtual)      |  16GB |     4 |  400GB SSD |     20TB |
    | OVH                    |   8GB |     2 |   40GB SSD |      ∞TB |
    | Scaleway (dedicated)   |  32GB |     8 |   50GB SSD |      ∞TB |
    | Online.net (dedicated) |  32GB |     8 |  750GB SSD |      ∞TB |

$80/mo

    | Provider               | RAM   | Cores | Storage    | Transfer |
    | ---------------------- | ----- | ----- | ---------- | -------- |
    | LightSail              |   8GB |     2 |   80GB SSD |      5TB |
    | DO                     |   8GB |     4 |   80GB SSD |      5TB |
    | VULTR                  |   8GB |     6 |  150GB SSD |      5TB |
    | Linode                 |  12GB |     6 |  192GB SSD |      8TB |
    | Hetzner (virtual)      |  32GB |     8 |  600GB SSD |     30TB |
    | Hetzner (dedicated)    |  64GB |     8 | 1024GB SSD |     30TB |
    | OVH                    |   8GB |     2 |   40GB SSD |      ∞TB |
    | Scaleway (dedicated)   |  32GB |     8 |   50GB SSD |      ∞TB |
    | Online.net (dedicated) |  64GB |     8 | 1500GB SSD |      ∞TB |

Gist available here: https://gist.github.com/justjanne/205cc548148829078d4bf2fd39...

It's must be said Hetzner have much cheaper offers on their auction: https://robot.your-server.de/order/market

There also have no setup cost for these dedicated servers.

They're also worryingly keen on non-ECC machines at the low end though

The OVH numbers are wrong. For example in the public cloud for $ 80 you get 60GB RAM, 4 vcores and 400 GB disk (or 200 GB SSD no raid).

Could you link the offer? I couldn’t find any in their VPS or Dedicated offers better than the 16$ VPS (which I then used for the 40$ and 80$ tier, too)

https://www.ovh.com/us/cloud/instances/prices.xml

I edited the gist, but sadly can’t edit the post anymore. Thanks, btw. I completely missed those.

Thank you for this.

Well, you are comparing shitty atom cores to E5 cores. These are just some random numbers without context.

Well, I’m just extending the previous’ posters list.

But Amazon will lower your CPU quota as well, if you use it for too long, so it’s not like the numbers of Amazon themselves even really mean anything.

The same story with storage performance – Amazon’s is horrible, but it’s network-attached.

"EDIT: Anyone cares to explain his reasons behind a downvote?"

I downvoted you because you talked about your downvotes.

Don't interrupt the discussion to meta-discuss the scoring system.

Write your post and live with the results.

My question is why does no one talk about prgmr?

Do you have experience with prgmr? It's reliable enough for a small project?

I've used it and am happy with it. It's reasonably reliable, but has more downtime than the big providers. I wouldn't run mission-critical software that needs 5 9's uptime on it, but for anything else it's fine, certainly for personal projects. They're transparent with any outages, so you can check up on the outage history on their blog: https://prgmr.com/blog/

It's a small business run by a few people (though it's been around for 10 years, so a pretty stable one), which has the pros and cons that go along with that. The tech staff is good, techies who know what they're doing and generally assume that you do also. So if you send a request or problem report, you aren't going to get a form reply that asks if you tried turning it off and back on again. But it's just a handful of people, so if there's a major issue, fixing things is pretty manual and slower than at places that have armies of 24/7 devops staff.

One specific thing I really like about it: it gives you SSH access to a proper text console, in case you want to install a custom OS, recover a broken install, etc. Most VPS providers give you console access, but most do it via VNC in the browser, which is not my favorite way to do sysadmin work.

> in case you want to install a custom OS

Do they let you do that? They don't say that in the purchases page.

Also, how is uptime relative to vultr?

Yeah, you have a choice of using their prebuilt disk images, running one of the officially supported OS installers from the console menu, or downloading your own installer. The list of prebuilt images and supported installers is here: http://wiki.prgmr.com/mediawiki/index.php/Distributions

I haven't used vultr so can't comment on that.

[Disclaimer: I work at prgmr.com]

You can install a custom OS. But it can be difficult to use an installer we don't provide right now because we only allow serial console access, not VNC. This means most installers won't work out of the box. Worst case you can dd an image to the disk using ssh from the rescue image.

FYI we don't do overage charges right now. For network, if we can't throttle your traffic down then we will shut your service off.

Our blog is a little misleading these days in that for downtime for individual servers, we started emailing customers directly rather than posting to the blog. This is because we want to make sure customers see the downtime notice. We also got confused responses sometimes to the blog wondering whether a given service was affected or not and if we email directly there is no such confusion.

I think our worst case downtime barring about 5 services this year has been the following:

* 0.75 hour network outage, unplanned - 2016-03-16 (gave proportional credit)

* ~2.5 unplanned downtime due to hardware failure requiring new components - 2016-04-03 (gave 15% month credit)

* 2.6 hours downtime from start of maintenance window, planned due to security upgrade - 2016-07-23 (gave proportional credit)

* 2 hours or less downtime, planned due to security upgrade - sometime around 2016-09-01 (gave proportional credit)

* 1.5 hour network outage, unplanned - 2016-09-09 (gave proportional credit)

* 1.3 hour network outage, unplanned - 2016-11-06 (gave proportional credit)

* 2.04 hours downtime from start of maintenance window, planned due to security upgrade - 2016-11-18 (gave proportional credit)

This is a total of up to 12.69 hours downtime over the year so far, assuming downtime started at the beginning of maintenance windows (it usually started after.) Of that 6.05 hours, or less than half, was unplanned.

So far this year there's been about 336 days or 8064 hours. 12.69/8064 is 99.84% uptime overall, which is significantly lower than we would like. For some servers the uptime has so far been significantly better in that there were no hardware failures, one of the security upgrades was unnecessary, and the turnaround time for the remainder of the security upgrades was much faster than for this particular server.

For this particular server, the largest downtime contributors were security upgrades and network outages in that order. For network downtime, we got around to setting up our second upstream but there's a number of single points of failure we should take care of in 2017. There is also some additional scripting we should probably do that would cut down on the network downtime a lot, such as automatically taking down BGP if connectivity beyond the first hop is lost.

For the security update downtime, I think our most realistic bet right now is to get ourselves on the latest version of Xen once it comes out. That will hopefully have a stable implementation (not a technology preview) for live patching.

It's probably since the ones listed above are the typical VPS to go to when it comes to cheap hosting. The ones you mentioned are lesser known.

Reply on news.ycombinator.com