It's nice but I think you need to test on more than 2 instances per provider. Things can vary _a lot_ and I think benchmarks like this should collect a lot more data. With Ansible and such it's not too difficult to set up and run a benchmark of 100 instances per provider and it's not overly expensive either. I've even seen it recommended that when starting an instance on AWS, you spawn 100 instead and take the one that performs best.
> We are really surprised how bad OVH does in this benchmark, they claim to have local RAID 10 storage, so it could be that they are throttling, since we can't imagine that there are that many "noisy neighbors" in two locations, but we can't say that for sure.
Your numbers match the numbers on vpsbenchmarks.com [1], so it probably is as bad as your test suggests.
It might be cheaper to route your packets far away and then back.
I remember being on an independent ISP (VDSL reseller), and my neighbor down the street was on the incumbent Cable provider. When I did a traceroute, I discovered my packets were being routed across the country, south, across the country again, and back north, before getting to my neighbor.
I'm happy Vultr customer and I can only be happier. I was looking for other providers recently for my personal needs, as I don't need that much resources and there were other providers with less that $5/month prices, but now I'm not looking anywhere.
Sweetest thing is hourly pricing. Sometimes I need Paris VPN for few hours. I have some scripts, so I can launch new instance with configured OpenVPN in 2-3 minutes, use it, then dispose and I would pay only few cents. Impressive, if you ask me. That's what cloud is for.
Network speeds are quite good, vultr has lots of features (want to install OpenBSD from your own ISO via web 2.0 KVM, no applet nonsense? Or Windows XP? No problem at all), and prices are just so low now. Also IPv6 works fine. I don't know about their support, I never had to contact them yet, but otherwise it's awesome service.
Do you have to recreate the entire system each time? That's one of the downsides to Digital Ocean for me, you need to pay even when you're not actually running your droplet.
There might be better ways to do this but each new instance on vultr takes a few minutes to set up (you can actually use their web VNC console to watch their install script run，a unique experience not offered by other VPS providers I have used) and then you can restore old config from a snapshot which takes a bit more time but it is pretty fast overall.
There are options to load your own distro image and init scripts but I have not tried them since most things I need in a transient instance can be installed by following a simple bash script calling apt.
I'm recreating the entire system each time. Basically they copy predefined CentOS image, then running my post-install script which installs OpenVPN and copies some configuration files. It's quite fast in the end.
I think, you can snapshot the entire system, then delete server and recreate it from that snapshot later.
But yes, if you're just stopping server, you'll pay for it.
I'm playing WoW and sometimes something goes wrong with my ping from Astana to Frankfurt (where WoW servers are located), but Paris still works fine. So VPN solves that problem.
I occasionally start a VPS for a few hours to download stuff using OpenVPN AS... Living in France under constant phobia of HADOPI law :)
https://en.wikipedia.org/wiki/HADOPI_law
Yeah, 2 GB RAM for ~$3 is amazing, while the mentioned Vultr plan is 512MB for $2.5 and DigitalOcean offers 512MB for $5… Sure, these companies have other advantages (e.g. DigitalOcean has very fast storage) but RAM is so damn valuable if you want to run many services, run software builds, etc.
Unfortunately Scaleway is Linux-only, they use some networked block storage that's not yet supported on FreeBSD apparently. (Needs to be mounted from inside the VM I assume? Damn "cloud" stuff.) Scaleway also used to have no IPv6, I think this might have been resolved though (but one address only IIRC? no /64?)
prgmr is my current choice for personal stuff — 1.25 GB is not as good as 2 GB, but they offer normal Xen VMs with full access — install anything you want. Also the good feeling of supporting a small team of sysadmins instead of a big corporation (esp. vs Amazon/Google, not so much with the other small hosters). The only downside is that they only have US servers, nothing in Europe.
> they use some networked block storage that's not yet supported on FreeBSD apparently
This may only apply to their physical offerings, not to the VM-based ones, and you may be able to get it to run using some nasty hacks (possibly involving loading your kernel after booting theirs). Either way, there are no official FreeBSD images so you'd be on your own.
Their IPv6 support is indeed lacking (one address, and you lose it if you stop the server).
"FreeBSD on x86_64 (C2S, C2M, C2L, VC1S, VC1M, VC1L) needs an alternative way to store data or an NBD driver for FreeBSD, and more work on the boot system"
This is a great box, I use one myself but people need to know that SSD is network attached which comes with its own slowness. It's OK it's just something you need to be aware of.
DigitalOcean's payment model is great for this. What you see is what you get without small charges or other fees. I vaguely remember Lightsail's release being met with poor feedback on HN because they really missed part of what makes DO by complicating the payment model.
Currently DO does not charge for network overages. I guess they would suspend your node for going way over the prescribed limits.
I'm not affiliated with DO, but I just really like the service.
I've got a few things running on $3/mo Scaleway boxes. Bandwidth is unmetered and 200Mbit/sec. The only extra cost I pay for is additional storage volumes (50gb included, additional 1 euro per 50 GB/month) - other than that, dead reliable and all inclusive.
I run my blog on 3 distributed instances, one of them being at Vultr. (The 2 others are DO and Scaleway.)
In my experience, all 3 are very reliable: zero downtime since I started using them 8 months ago. Vultr has the nicest dashboard: it uses the noVNC HTML5 client to give you access to the console, it supports 2FA (TOTP) for authentication, etc. They give you $50 of free credit to use in the first 60 days (or did, when I signed up). Plus they accept Bitcoin payments!
Only minor complaint I have is when you create a vps they don't give you its SSH host key fingerprint. You have to log in the console to get it...
I've been running on vultr $5/mo boxes for years now. No complaints - the boxes are snappy and the web UI/dashboard is nice. I originally picked them because they supported FreeBSD out of the box, and let you upload your own ISO to boot from if you want to run something else (like OpenBSD). I've only ever had one interaction with their support staff, and the service was prompt and professional. I've been on AWS and DO as well, and am happy with the vultr experience by comparison, so can happily recommend them if you're curious to give them a shot.
They just recently started supporting OpenBSD officially as well.
I immediately switched. This way I wouldn't have to deal with the generally more messy and weaker[1] OS that is FreeBSD, nor would I have to use old pf syntax.
I used to recommend it to everybody, but don't any longer. I had two negative experiences that made me stop recommending it.
The first is frequent restarts. Very small sample size, but both me and a friend started experiencing restarts every month or so. I don't think I got any advance warning about the restarts, but I'm not positive.
The second is that networking between instances stopped working for me in the middle of the night shortly after Vultr did an upgrade. Support wasn't able to help (though they did try) and I couldn't figure out anything on my end. I had to switch everything to AWS to get the site back online before people woke up in the morning.
I was super happy with it before those two problems, though, and the people there seemed competent and professional. I was also using FreeBSD, so it may not be as well supported as Linux (which may have contributed to the networking problem).
Thanks for this comment. I was about tell my team to investigate Vultr as an option for our next website, but I guess we'll first do some more research into this.
We're still on AWS too. Currently spending about $50 / site which adds up quite quickly when you have 20+ sites. Our current breakup is something like (2 x t2.micro - one for website, one for cron created using Elastic Beanstalk, 1 x db.t2.micro for RDS + Elastic load balancer cost). I'm sure we can bring it down to $10 / site if we were to move to Vultr or Scaleway but it's a very small site to pay for the reliability AWS has provided to us all these years (apart the recent S3 outage). So guess we'll too just stick to AWS for now until.
There are many reasons but a few from the top of my head are:
1) The whole system is running on Amazon's Elastic Beanstalk (via Docker). Beanstalk provisions 3 instances for every site (1 Web, 1 Worker/cron, and 1 RDS) for every app with ELB on front.
2) It makes installing updates easy and automatic using deployment scripts which we have on every site.
3) Everything (web/cron/database) is configured to scale up and down automatically, i.e. new servers start as per demand. This may not be possible otherwise.
I encountered the same problem with restarts you did, though I only had 4 in ~8 months. They gave no advanced warning and simply notified that my instance had been restarted which was extremely unprofessional to me.
That's... weird, actually. I'm inclined to think that this is incompetence rather than malice, if only because as malice its extraordinary incompetent. I mean, what would the point even be? I would think it's a MITM, but they already control the hypervisor, so I can't imagine any sort of compromise that isn't completely trivial to implement invisibly.
I used to colocate with one of their sister companies and I've chatted with the CEO a few times on IRC. They were generally very reliable, although they didn't return my rail screws or cables oddly enough.
I've just had one small $5 VPS with them for about a year now. No downtime, no complaints. Since my needs are so basic I think I could probably get away with one of these $2.50 offerings, however the problem - and it's the only problem I've run in to with Vultr so far - is that downgrading VPS plans isn't possible. You can only go up.
It is a pretty good deal, maybe they'll raise that to 3 since most clusters require 3. I wonder if competition could otherwise spin up a bunch and make their loss leader (if it is a loss leader) into something unsustainable.
>Our most affordable way to try out cloud computing, this plan features 512MB RAM and 20GB SSD disk storage. This is perfect for test-driving your next application before upgrading to production level specifications.
However that does sound production level, easily surpassing it even, for a small business or organizational website. There's even enough ram and storage for a light database.
I used to run a Debian 5 server with full-blown LAMP stack and it was enough even for larger sites. Today, the latest version will just say "out of memory" while installing packages. :/
I'm using similar plan from Ramnode as a sandbox for 2 years already and very happy about it. Ramnode is a highly underrated hosting provider but it has the outstanding support and great features/speed/reliability. I'm not using it for production though.
I've heard somewhere that Vultr retains your stored credit card indefinitely (even after you close your account). It sounds a bit weird, that's why I avoided their services. Can someone confirm this?
Believe it or not this is actually very common. In fact a lot of people who integrate Stripe recommend this pattern. They recommend to never delete customers and cards in Stripe, instead just remove the subscription from the customer.
The idea is, if the customer wants to come back and reactivate service they don't have to provide billing details again.
I don't do this though. When customers cancel my startup, I delete the card and cancel the subscription in Stripe but leave the customer.
This pattern can be recommended so long as there is no liability for financial information disclosure. When you're hacked and a customer who closed their account 2 years ago has their CC info used for fraudulent transactions, you just trollface and move on.
This is why I am baffled as to why the CC companies eliminated the transient number technology, which was pefect for e-commerce vendors like this. At least with those, after 3-6-12 months the number expires and no damage done when the vendor is hacked. I assume it was dropped because this technology complicated data mining.
Note that in the case of Stripe, the site doesn't hold onto "financial information" per se - they hold onto tokens where Stripe is the holder of the information, and they can't access the card number or CVV after it's been sent to Stripe - so all fraudulent transactions would have to be requested through that Stripe account.
They shouldn't have access to card number and CVV either way because usually they are not PCI-DSS certified. Stripe handles that in an iframe (requirement of PCI). CVV cannot be stored at all even by Stripe.
I was setting up a test vps there a couple of days ago and made plans to try out the $5 plan with $20 credit they were giving me on a $5 charge. So I had a runway of about 5 months but when I made payment, this $2.50 plan was right there, and it wasn't there before the payment. Now I have 10 month trial period. Never been happier.
The dashboard is nice, the ram is a bit low (but the price point is killer), I wasn't able to deploy a lets encrypt cert without pooled memory. The service seems to be excellenct in the few days I've used it. It doesn't seem to have a "don't ask for 2FA on this device for n months" option but it's a small complaint.
If you use certbot-auto to install the client, part of the installation process involves compiling some crypto library (IIRC) which tends to fail on low-memory systems, hence the need for swap.
Packaged versions of certbot (from your distribution's package manager) don't suffer from that problem and have fairly low memory requirements.
Yeah right swap memory. I'm a server management newbie. This was a cent os instance and apparently it didn't have a native certbot so I had to use certbot auto.
Been a while since I've tried Vultr, but my experience with them wasn't good. I started one instance, and all went well, decided to get a second one going. The performance on the second instance was horrible, nothing like the first. I occasionally got e-mails stating they were rebooting hardware my instances were on. And after just a few months, the first instance I had refused to boot due to file system errors.
After the performance issue with my second instance, I wrote them off as not being good for anything serious. And after the file system issue, I decided they weren't useful even for anything casual. Hopefully things have changed, but don't assume that just because you've been up and running for a couple months with good performance that things won't fall apart soon.
I recently switched to Vultr from a DigitalOcean and Compose.io setup.
Compose is nice for backups and replicas but they drop the connection like all the time and then the replicas start playing who wants to be master. Which brought downtime to my application at times. Also there expensive so I said f that and manage a Mongo instance myself. Since it was a small project that had to much overhead.
I moved from DO was primarily pricing and the lack of data centers.
Vultr's Miami DC is my jam since I'm in Tampa. Atlanta is nice too.
Just migrated from the old $5 plan to the new $2.50 plan which even has more SSD space. Unfortunately, that involved copying a snapshot to a new instance, because they don't support "downgrading", though in this particular case there was no technical reason to not support it directly.
I just "downgraded" a legacy $10 instance to a $5 instance, as it now has better specs than what I was using, so they do allow it. It seems the $2.50 plan is a special class (probably loss-leading) and doesn't fully integrate with the rest of their products.
I was running one of my projects on Vultr for a while and didn't appreciate the random restarts with no warning. Ended up using DO for my next project. I hope that DO reduces their prices, right now they're getting destroyed by Vultr and Linode.
OVH is currently building a datacenter in AU. I guess (hope) I some point they will offer their vps/cloud there as well. But so far their pricing is higher there then EU/CA.
Uhm not cool, I've just registered an account following a promotion that claimed they'd match the credit I added, I've added $10 and boom, no promotional credit whatsoever. Just opened a support ticket to see what's going on...
I signed up using that offer, $25 matched.
It was obvious at every step of the sign-up what was happening with regard to the matching. Either you had the page cached or they screwed up, but worth getting in touch they are very responsive which is refreshing for a cheapo host.
It's nice to see all this competition ($5 Linode, now $2.5 Vultr) but don't they risk overallocaton?
Let's say they have physical machines with 96 GB mem that hold 200 1 GB VMs each, in the hope that no one uses more than half their ram. Same for CPU or disk I/O or whatever resource.
Now that the entry level prices are dropping but the offered resources remains the same, people will spin up more VMs. Doesn't it become dog slow for everyone?
KVM has KSM, so there is definitely the possibility of having say, 150x 1GB VMs on a 128GB physical RAM server; possibly much more than that depending how much KVM/KSM has improved since I last looked at it.
I don't think this is true for DigitalOcean. I believe you have to destroy your VM ("droplet") in order to stop being charged. If it's just powered off, you're still billed for it. You can take a snapshot, destroy the droplet, and restore the snapshot later, but there is a small monthly charge for storing snapshot images as well.
"Am I charged while my Droplet is in a powered-off state?
Yes. Your diskspace, CPU, RAM, and IP address are all reserved while your Droplet is powered off. If you want to save a Droplet for future use take a snapshot of it and then destroy it. You'll be able to create a new Droplet from the snapshot image anytime to bring it back online."
Well, they should've planned for it. In my experience their VMs were quite good, I never felt much contention. But it's VPS, of course, you can never be sure.
https://www.webstack.de/blog/e/cloud-hosting-provider-compar...
Edit: I've submitted it as well: https://news.ycombinator.com/item?id=13798023