Hetzner Cloud
hetzner.com
hetzner.com
/ducks....
I use them heavily for their cheap object storage + backups.
But even for cheap, dumb storage situations they sometimes give "interesting" responses to support questions.
To be fair, I contact them maybe 3 to 4 times a year (always related to hardware / network down issues). So, my experience may not be representative.
> When contacting OVH support in French, I generally receive very good support. I sometimes wonder how good the response is when they receive the same question in English.
Still using AWS for Apps which rely on cloud features, e.g. SES/RDS/etc but for static servers Hetzner is now our goto.
Super exciting to see them entering the cloud space and offering easy snapshots + backups, should open it to hosting more stuff on there.
The one difference is noticeable latency from their DC in Germany vs the instant response times I was getting from AWS's N.Virgina DC. Would obviously love it if Hetzner could open a DC in the US.
Having everything in the EU makes it much easier for them and is a selling point for many customers.
Quebec appears to be the better choice for cheap renewable energy.
Not sure if that helps with your latency problem or not, but options.... :)
They should start operations this year
Since a few weeks ago, Nvidia changes their driver ToS, and any new versions of the driver are not permitted to be used in datacenters (except for crypto mining).
See more here: https://news.ycombinator.com/item?id=15983587
Considering this is HN, it's a pretty safe assumption that people are running these side projects to turn them into startups.
Hardware fails, and part of having a dedicated server is that you are directly on top of said hardware and have to be prepared for it to fail.
One commenter wrote "Hetzner has an excellent reputation for rock-solid engineering" and as a counterpoint I wanted to tell my anecdote which opposes that "rock-solidness". So for that I get downvotes :)
In retrospect yes, I agree that we had to be prepared for such event, to have failovers and so on.
So I'm still sold vs. AWS, and I have to run some nodes, and I think great, I can put the business credit card on a cloud account and run stuff. But then Google refuses to let me sign up because every time I try it says "Unable to verify with this phone number.". This could be because I can only access Google via proxies and they are not in the country of my phone number (China or HK). So perversely I am left in a situation where to get Google to allow me to give them money I have to first open a proxy compute node with another Google account in the jurisdiction my phone number is in, install a proxy, then access them through that. But you'd think they probably block their own IPs, so that probably won't work.
So Hetzner, maybe you got my business.
Having some kind of network topology (like OVH's weathermap) would help with that, esp where to position servers that rely on certain peering points.
"How are the Data Centers connected to each other?
The two Data Center Parks are both connected to Frankfurt (FFM) and each other with dark fiber. Thus, a redundant loop is formed, which ensures the availability of a Data Center should one of the connections fail. The n10 Gbit/s connections provide ample bandwidth between the Data Centers.
The bandwidth of the connections between Nuremberg-Frankfurt, Nuremberg-Falkenstein and Falkenstein-Frankfurt are at least 120 Gbit/s. Through the Frankfurt location data is transported to the peering partners at DE-CIX and also to the uplinks Noris, GLBX, Aixit, AMS-IX, Init7 and Level3. At the Nuremberg location there are connections to Noris, KPN, Init7, Level3 and N-IX.
In each Data Center several Juniper EX Core switches, each with 64x 10 Gbit/s ports, are operated and bundle the streams of the Data Center to the n10 Gbit/s backbone and then over the various uplinks. "
And here is a list of their peerings: https://www.hetzner.com/unternehmen/rechenzentrum/
Let's just hope that they get more peering partners over the medium term. Peering with Telekom would be very helpful, pings from within Germany can be close to what you get from the US East Coast (from my not representative tests).
I think any real effort at enshrining net neutrality as law instead of a gentleman's agreement needs to include provisions that prevent residential / small business ISP's from throwing their weight around with peering agreements at the detriment of their customers.
-------------------------------------------------
nench.sh v2017.06.01 -- https://git.io/nench.sh
benchmark timestamp: 2018-01-23 20:03:56 UTC
-------------------------------------------------
Processor: Intel Xeon Processor (Skylake, IBRS)
CPU cores: 1
Frequency: 2099.998 MHz
RAM: 1.9G
Swap: -
Kernel: Linux 4.4.0-112-generic x86_64
Disks:
sda 19.1G HDD
CPU: SHA256-hashing 500 MB
3.249 seconds
CPU: bzip2-compressing 500 MB
5.604 seconds
CPU: AES-encrypting 500 MB
2.970 seconds
ioping: seek rate
min/avg/max/mdev = 32.9 us / 63.2 us / 4.71 ms / 22.5 us
ioping: sequential read speed
generated 22.2 k requests in 5.00 s, 5.42 GiB, 4.44 k iops, 1.08 GiB/s
dd: sequential write speed
1st run: 391.96 MiB/s
2nd run: 311.85 MiB/s
3rd run: 315.67 MiB/s
average: 339.83 MiB/s
IPv4 speedtests
your IPv4: 94.130.181.xxxx
Cachefly CDN: 81.40 MiB/s
Leaseweb (NL): 134.69 MiB/s
Softlayer DAL (US): 6.10 MiB/s
Online.net (FR): 88.06 MiB/s
OVH BHS (CA): 7.12 MiB/s
IPv6 speedtests
your IPv6: 2a01:4f8:1c0c:xxxx
Leaseweb (NL): 92.31 MiB/s
Softlayer DAL (US): 4.22 MiB/s
Online.net (FR): 79.54 MiB/s
OVH BHS (CA): 6.79 MiB/s
------------------------------------------------- -------------------------------------------------
nench.sh v2017.06.01 -- https://git.io/nench.sh
benchmark timestamp: 2018-01-24 04:11:01 UTC
-------------------------------------------------
Processor: Intel(R) Xeon(R) CPU E5-2650L v3 @ 1.80GHz
CPU cores: 1
Frequency: 1799.998 MHz
RAM: 992M
Swap: -
Kernel: Linux 4.4.0-112-generic x86_64
Disks:
vda 25G HDD
CPU: SHA256-hashing 500 MB
4.450 seconds
CPU: bzip2-compressing 500 MB
7.930 seconds
CPU: AES-encrypting 500 MB
2.435 seconds
ioping: seek rate
min/avg/max/mdev = 37.2 us / 62.3 us / 15.8 ms / 67.2 us
ioping: sequential read speed
generated 26.4 k requests in 5.00 s, 6.44 GiB, 5.28 k iops, 1.29 GiB/s
dd: sequential write speed
1st run: 619.89 MiB/s
2nd run: 775.34 MiB/s
3rd run: 774.38 MiB/s
average: 723.20 MiB/s
IPv4 speedtests
your IPv4: 165.227.96.xxxx
Cachefly CDN: 183.84 MiB/s
Leaseweb (NL): 11.74 MiB/s
Softlayer DAL (US): 6.29 MiB/s
Online.net (FR): 11.81 MiB/s
OVH BHS (CA): 34.46 MiB/s
IPv6 speedtests
your IPv6: 2604:a880:800:xxxx
Leaseweb (NL): 12.01 MiB/s
Softlayer DAL (US): 4.41 MiB/s
Online.net (FR): 23.50 MiB/s
OVH BHS (CA): 36.51 MiB/s
-------------------------------------------------I'll quote a different customer's opinion (on those forums) that summarizes my experience using their service: "Whenever I've tried to use Scaleway for something serious, I regretted bitterly".
My experience was essentially this: Scaleway advertises that you can have virtual servers with up to 10 150GB SSD volumes, so I signed up and created a virtual server with two 150GB volumes. After several attempts over three days, the server wouldn't start at all; it never actually started.
Scaleway's support told me that "it's because there are no available nodes matching your configuration." and "If our stock is low then there is more chance that the only free nodes are "default" ones, with 200GB available only". The solution, according to their support, would be to "keep trying to start the server until it works".
To add to this, Scaleway started billing me for the (unused) volumes attached to a server that cannot be started because, according to their support, they didn't have the resources to provide it.
So, I certainly didn't feel assured that they are able to provide what is advertised on their website (virtual servers with up to 10 ssd volumes). My key takeaway from this experience was that if you have a platform on Scaleway and you need to add servers to cope with growth/features, they may not have the capacity/stock to provide it. Even if you have only 1 server, which was my case, they may be unable to say when you'll be able to start the server on a physical instance.
My experience with Hetzner is/was great, but as a customer it's impossible to rely on a service provider that extinguishes a whole service line in a few months and forces me to move a whole infrastructure at a whim...
It would be great to have an EOL warning at least a few months before a service line is discontinued, just to provide enough time for customers to move critical/production workloads to another infrastructure provider.
Check out our pricing:
https://www.ssdnodes.com/pricing/
Happy to answer questions if there's any.
I'm sure you'll find some customers at those prices, but it does not compare with what Hetzner is offering here just because of that.
In most cases our annual price is equivalent to 1-3 months of a competitor's monthly price. You mentioned AWS so I'm going to use them as an example, even though it's not an apples to apples comparison. Their m5.xlarge is similar to our X-Large plan (16GB RAM + 4vCPUs). I'll ignore that it doesn't include storage or bandwidth, for simplicity. The m5.xlarge (with no storage or bandwidth) costs $140/month on-demand, which exceeds the annual price of the X-Large plan. If you reserve an m5.xlarge for 1 year (paid upfront) with AWS, the cost is a little less than $1K, which is close to 10X the cost of our X-Large.
If customers cancel their server with us after the 14-day refund period, we provide a prorated credit for the remainder of their billing cycle[0]. So if someone cancels after 2 months, they get the remaining 10 months as credit. They can use that towards renewals, new services, etc. I totally respect it's not for everyone, but the cost savings can be substantial for those who are able to use our services.
We are also expanding to Europe in the near future.
https://blog.ssdnodes.com/blog/comparison-vultr-vs-digital-o...
Provisioning is already impressively fast, especially if you're coming from EC2 where it feels like an intern has to press a button or something to get an instance online.
full disclosure: am working for hetzner cloud
However, what keeps me on Linode is that they have a London datacentre with excellent peering. I've seen 9ms pings, 150 miles away. Makes for seriously fast websites (in concert with good development processes).
Hetzner responds in 31-35ms from here. That might be Good Enough™ for many applications but it's not as special.
I've never been this tempted to jump ship though. Hopefully some of these features will become industry standard.
That's because they fleece you with the exchange rate. You're paying a many-points worse exchange rate, even against a stabilised market average.
For one recent actual example, Paypal recently gave me 1.339 USD for 1 GBP in a transaction. The raw market rate that day was 1.421. The week mean was ~1.410. Paypal already holds this cash, they're not exchanging jack on the open market.
A pre-pay FX card —shop around— would have given me $1.40/£ on that date.
Paypal work like a bookmakers. They cover the entire spread with very little risk. Nothing on Paypal is really free.
Very different obviously if you have APIs that require low latency.
This way you can run for example a FreeBSD or Arch Linux VPS for under 3 Euro per month. :)
Anyone know if the rescue system is available on the cloud offering as well? (So alternate OS can be installed)
And do the "cloud" have the same painful setup with the gateway outside the subnet?
Currently I'm running a CX20 (primarily as mail server and for smaller private projects), thinking of switching to the CX21 (more RAM, less storage, cheaper) or the CX31 (way more RAM, more storage, price per month is around 2€ higher than the CX20). If someone has benchmarked them against the old offerings, I'd be happy to see the result. :)
URL: https://www.scaleway.com/pricing/
The problem with Scaleway is that they are working at capacity and are almost always out of stock.
They are expanding now (and are hiring!), https://blog.online.net/2017/11/13/scaleway-enters-a-new-gro...
The big difference between Hetzner and all the rest of the cloud servers, is that they start with 2GB RAM for their entry level packages.
Their setup is interesting and I really wanted to like it (I did like the pricing), but it was a dreadfully slow experience. And those Atoms really suffered when benchmarking HTTPS connections compared to almost any other provider. Not even close.
A vCPU means nothing without considering the underlying CPU being used, and whether or not it's a dedicated core or not.
Two Intel Avoton cores, quite inferior to the Xeons that power Hetzner's vCPU. (assuming they are in stock)
I wrote this in detail on another reply, but Scaleway's "customer support" forums will give you a good grasp of what you're in for if you buy their service.
Are you also working on docker hosting and dbaas?
I'm a german development contractor and long-term Hetzner customer. These are the only two missing features, causing us to still operate our own systems.
Regarding the DSGVO: Do you support encryption of the filesystem?
Regarding the DSGVO: Do you support encryption of the filesystem? -> Hetzner Cloud servers are fully virtualized. So the customer can even fully encrypt the whole server. --Katie, Marketing, Hetzner Online
I don't think they have understood, what Cloud Computing means.
All WHOIS protection services I’ve seen until now either ignore fax, phone and letter, or just respond with "this method of contact is ignored, please use the email"
Typical example: https://www.rockit-internet.de/impressum/
> Companies and online media outlets, such as newspapers, magazines and journals, wishing to operate within any of these countries are obliged to make this information easily available to users. This requirement also applies to legal entities registered outside of the German-speaking realm that compete on the respective German, Austrian, or Swiss markets.
[0] - http://www.it-recht-kanzlei.de/nicht-eu-ausland-impressumspf...
[1] - https://www.1and1.com/digitalguide/websites/digital-law/a-ca...
EDIT: added a source in English
To clarify - saying this as someone who was forced to blacklist their IPs on more than one occassion, not as someone who ran into tainted IPs as their customer.
The only organization that caused issues is Microsoft / Outlook.com, but they seem to block every IP address by default. I was able to unblock myself by filling out a form and waiting a few hours for all my servers so far.
I've used Hetzner for two years and I never had any issue regarding blacklisted Ip addresses.
Emailing abuse@hetzner.de yields:
Dear Sir or Madam,
Thank you for your email.
In order for us to process your request, we ask that you
please fill out the form at the following link:
https://abuse.hetzner.de/
Please use this form for all future complaints.
Kind regards
Abuse team
Still, even with this permission and several hours in, the box is still spamming, customer is "investigating" and Hetzner has done all they could to handle this incident.
https://support.microsoft.com/en-us/getsupport?oaspworkflow=...
Check your /24 neighborhood first though, one of your neighbors might be sending spam:
It's the first link in the delisting section: http://go.microsoft.com/fwlink/?LinkID=614866&clcid
I also recommend signing up for SNDS.
They have an automated abuse-handling system that sends your complaint to the spammer and asks them politely to stop. This often works for small-time operation on a single IP address, and for compromised servers, but it's completely inadequate for dealing with large customers and rogue resellers spamming over large netblocks, and against darker-coloured-hat spammers engaging in reprisals.
Replying to the automated mails can get you through to a human, but their ops are stuck on a limited script so they're not really any more use than the automated systems. They have no interest in investigating wider patterns of abuse than a single spam from a single IP.
I regret having to blacklist but it's not worth the time and frustration of trying to engage with them.
(Not talking about all of Amazon, just SES - AWS for example is a different animal.)
That’s what I’m trying to get at here.
We have also had good experience with Hetzner South Africa's abuse department.
We do see plenty of spam from OVH however. On one occasion when we reported several IPs to them for sending Russian spam, they did not reply.
> Above-average raised floor system
What's strange is that only RAM and disk space increase linearly with the price, not the CPU, nor the traffic. That creates an incentive to buy small instances and getting a lot more for your buck.
That I am presently not one is entirely Hetzner's own doing: Some years ago, testing out their offerings (which were good!), at one point I deleted my last VPS, thinking I'd come back in a few days and actually start getting serious.
Not to be! Virtually the very second I deleted the instance, an email pinged in: No active machine, account deleted. Which sent me over the border to the French guys (Whose interface to this day is a riddle to me, every time I haven't worked with i for a week or two. I can read the French version, it's not a language thing, just a general UI disaster).
For 35€ a month I get 32GB RAM, 8 vCPUs, 240GB SSD + 20TB Traffic. The comparable 40$ plan for Linode offers 8GB RAM 4 CPU Cores 48GB SSD and 3 TB Transfer (See: https://www.linode.com/pricing)
Am I missing something or is Hetzner just way more competetive?
Edit: Their Cloudhosting control panel is top notch.
The whole architecture has been redeveloped from scratch. We have also designed and developed an entirely new control panel called the Hetzner Cloud Console. It is and will be actively worked on to make it as easy to use as possible for our users. Feedback is highly appreciated :) Also the new cloud servers are billed hourly instead of monthly, as the old VPS were.
Full disclosure: I work for Hetzner Cloud as UX Designer.
No matter what filter settings (including default) I see 0 results.
I have to say the UX is beautiful and easy to use. Also, the setup was easier, and performance better than any other provider I was testing. I'm sticking with hetzner.
OVH had basically the same price but their heavy investments in new DCs have made them more expensive. Since Hetzner doesn't invest a lot (at least in new buildings) they don't need to raise prices for that.
→ OVH VPS Cloud 2, 2 cores 4GB RAM: 23.79€
→ OVH VPS SSD 2, 1 core, 4GB RAM: 9.51€
→ OVH VPS SSD 3, 2 cores, 8GB RAM: 17.84€
For me, OVH's unlimited traffic does not explain the price difference. I will highly consider switching to Hetzner.
EDIT: correct meaning
Meanwhile, in Spanish I would mix up cebolla vs cepillo vs cabello vs caballo.
Approaching a woman in a bar: "I like your onion."
No, you’re missing nothing. Yes, DO and especially AWS and GCP are about an order of magnitude more expensive than Hetzner, OVH, and similar providers. Much more for traffic.
It’s always been like this, the cloud providers simply have much larger profit margins.
1. What would be the downside of hosting my Apache server on Hezner instead of AWS? Will it affect load times, website SEO, downtime?
2. Hosting Mysql Server on CX31 on Hezner instead of AWS? If my Apache server is on Ec2 and I make a connection to a Mysql server on Hezner will it affect performance of my sites because they're hosted on different places?
I have a very basic understanding of networking, so appreciate any comments from someone who knows about these.
2. For sure of you are using AWS in US: Traffic across the atlantic will incur at least ~65ms latency per roundtrip.
Can you comment if the location of my IP will affect my website SEO wrt to USA visitors?
If that was the case, all the companies that are hosted in the US would already be punished in Europe.
That said, noone knows the rules for Googles Ranking algorithm, so I guess you will have to try.
1. Pings within Europe can be a bit longer than to other providers, they lack peering with some providers. However, nothing noticeable (10-20ms max) and behind cloudflare TTFB wasn't longer than with other providers. Keep in mind that this is central Europe, ~15ms more from the US compared to Google Cloud BE or OVH in France and around ~20-25ms longer latency to the US compared to UK/Ireland. I've never experienced downtime but follow their status on twitter and they sometimes have (announced) maintenance for vhosts, think they don't live migrate customers. But I've also experienced that with AWS.
SEO is dependent on response times and the provider shouldn't make a difference.
2. That shouldn't be an issue and they peer with Amazon in Frankfurt. Expect 20ms round trip per db query (from a European DC). Can make a difference for websites with a lot of small queries. I would keep webserver and database closer together and rather keep Apache also on Hetzner.
It is something I can get from Packet.net, but not from Linode or DI (Although Linode promised that it was in their pipeline, I have yet to see it), OVH does offer this but I only experienced pain with their interface and service.
https://blog.simos.info/a-closer-look-at-the-new-hetzner-clo...
(not affiliated with Hetzner)
disclosure: am part of the frontend team of hetzner cloud
With Google, Microsoft Facebook as well as Amazon leading the way in layering out cables across continent, having a server in US means you have a high probability most ISP in the world has tuned the networking route around those traffic. The same cant be said for EU.
Nevertheless, I have made very good experience with their hardware, network and customer support - so they are also suitable for professional customers.
Hetzner got nasty mails from a couple of "victim" sites, forwarded them to me, helped me clean house and everything was fixed and airtight within hours.
I think it's to be expected that a few self-managed root hosts will be poorly managed (like mine was) or abused. But as far as I can tell, Hetzner does a credible job of taking care of problems.
This ( http://eur-lex.europa.eu/eli/reg/2016/679/oj ) replaces all data proyection laws in the EU.
Must comply now, but sanctions won't be in effect until may 25th.
That is not technically correct, and the difference matters.
The EU regulation forces it's members to implement their own local laws in alignment to the EU regulation. The EU regulation can get a member state into trouble if not implemented in time or correctly. They describe a set of "baseline" data protection laws for the whole EU, but the members can still have stronger local rules. Only the local law is what can get a resident company in trouble.
Germany has (had for a long time) very strong data protection laws that are still going to be stronger than the EU regulation.
* By "comes in", I mean begins being enforced. There's been a 2 year grace period as advance notice.
We host our cloud instances in our own data centers in Nuremberg and Falkenstein. And we operate our data centers in accordance with ISO 27001 guidelines while also adhering to strict German data protection regulations.Otherwise I fail to see why Finland makes sense, unless you need cheaper energy for applications that don't need low latency.
It does not have any options and it looks like it is not active yet. It would be great if they offered network storage through Ceph.
Are they just lowering their prices or am I missing out on features ?
e/ am part of the team working on hetzner cloud
The whole architecture has been redeveloped from scratch. We have also designed and developed an entirely new control panel called the Hetzner Cloud Console. It is and will be actively worked on to make it as easy to use as possible for our users. Feedback is highly appreciated :) Also the new cloud servers are billed hourly instead of monthly, as the old VPS were.
Full disclosure: I work for Hetzner Cloud as UX Designer.
https://wiki.hetzner.de/index.php/Hetzner_Rescue-System/en
From there you can do pretty much what you desire.
https://nixos.org/nixops/manual/#opt-deployment.hetzner.crea...
If you want security, use dedicated servers.
If you're hardcore, you can also do what a friend of mine did: Use the provided Linux system (Debian, Ubuntu, there's a bunch of choices) to fiddle its own boot loader, possibly re-partition the disk, upload your favorite system and reboot into that. You're root, you can do what the hell you want; at worst you might end up bricking your system to the point where you need to re-install from the rescue system.
[ed: never mind - 20tb included for the cheapest box. That's a bit of a departure from earlier. That's 60 mbps sustained 24/7 for a month. That's probably hard to meet with a full regular web stack running on a single tiny vm..
Oth, should be able to just ha proxy on one of those - or two-three w heartbeat...]
Also, maybe it would be worth contacting their support about your situation, maybe some other form of identification can be used.
Also, mind you that they have a /29 assigned from RIPE, so they would have sufficient space for 134,217,728 /56s or 524.288 /48s. They choose to leave all that address space unused for nothing, forcing their customers to work around a completely artificial scarcity of address space. Exactly the opposite of why they got a /29 from RIPE (and they could easily get more if they did actually manage to use it up).
I can't tell it apart from the dozens of other "good" cloud providers (Lightsail, Digital Ocean, Vultr, Linode)
I think the much-heralded cloud elasticity for most startups only serves to please investors "we can scale if we blow up over night!".
There is a big pit in between own DC, and a lot of VM's in terms of economic scalability.
Own DC pays pack when you have 500+ to 1000+ machines in a single location.
Therefore, they got an international reputation for delivering high quality for a cheap price.
This new offer looks very promising. I like the simple price model of 'pay per hour' + a fair 'maximum price per month'.
Any current clients know how fast provisioning is on these?
"Our quick and clever Cloud Console lets you create server instances almost instantly, usually in under 10 seconds."Hetzner
CX11
€ 2,96
1 vCPU
2 GB RAM
20 GB NVMe SSD
20 TB Traffic
CX21
€ 5,83
2 vCPU
4 GB RAM
40 GB NVMe SSD
20 TB Traffic
CX31
€ 10,59
2 vCPU
8 GB RAM
80 GB NVMe SSD
20 TB Traffic
CX41
€ 18,92
4 vCPU
16 GB RAM
160 GB NVMe SSD
20 TB Traffic
CX51
€ 35,58
8 vCPU
32 GB RAM
240 GB NVMe SSD
20 TB Traffic
OVH
VPS SSD 1
1 vCore(s)
2 Go de RAM
10 Go SSD
3,99 €
VPS SSD 2
1 vCore(s)
4 Go de RAM
20 Go SSD
6,99 €
VPS SSD 3
2 vCore(s)
8 Go de RAM
40 Go SSD
12,99 €"The host systems will be updated to fix the vulnerabilities as soon as possible. The necessary reboots will be announced on Hetzner Status. You may subscribe to be notified. "
I can't think of an approach that would be more direct to their customers than seeing it on top of their customer homepage.
Anyway, the proof is in the 264 day uptime!