Why Review Linode, Digital Ocean and Vultr?

For many years I have depended on LowEndTalk and LowEndBox for information about service providers. I have acted on many of the reviews, comments and recommendations I have read; for example, I did discover and use ToggleBox and ChicagoVPS among others through this site. But, over the last few months, my research has led me to believe that, of the inexpensive providers, Linode, DigitalOcean and Vultr were the best ones to consider.

I will also consider writing a review of ToggleBox. I won't be able to review ChicagoVPS because I haven't used them in about three years.

Please note that some links in this review are referral links from which I may benefit so, if you find this review helpful, please consider using one of them. Also, from this point forward I will refer to DigitalOcean as DO. This seems to have become a relatively standard acronym.

Quantitative vs Quantitative Comparisons

A simple Google search will show that there have been a number of quantitative comparisons of the above three providers; eg. Qing Wang's comparison.

But numbers are not the only reason to choose a provider. There is a flavour to each provider, a feeling you get about the features, competence, and customer-focus of each company. That often helps me decide more than anything else. So, after more than a month of experimentation with the three (though I gave up on Linode very quickly), I would like to share my personal experience with these companies without significantly addressing the actual numbers.

This posting is a little long, so I have broken it up into sections in case you want to skip some aspects. I hope it proves helpful.

Technical Support

First, let me begin by saying that technical support agents from all three services were all very polite and genuinely seemed to make an effort to help.

DO and Vultr had similar support levels. They would both typically reply to a request in less than an hour, often in much less time. Linode support was of a lower calibre. I rarely received replies in less than an hour and, in one instance, it took about four hours to receive a reply regarding an ongoing case. After I cancelled my Linode service I was told to send an email to support@linode.com but, after four days, I still had not received a reply. Only after I opened a new account (just to create a proper ticket), did I finally receive a response.

Notwithstanding the above comments, I did have one disappointing interaction with DO. I discovered an issue with their backup system, which caused my most recent backup to be removed. DO's initial response was that I misunderstood how their backups worked. But I didn't give up and pushed the issue. Eventually I dealt with the customer support manager who, after some additional correspondence, admitted that the issue was DO's. She also confirmed that, because of my persistence, the issue had now been fixed. She apologized profusely for the problem and thanked me for my efforts. To my surprise, I she also added a $25 credit to my account for my help. Now $25 is not going to make anyone rich, but I was impressed with DO's genuine concern for the customer and their effort to somewhat compensate me for my time.

DO also went once step further on the customer satisfaction front. They have set up a separate site to handle customer suggestions and complaints at digitalocean.uservoice.com. The site hosts discussions concerning issues and suggestions. In some threads DO even provides a service roadmaps for those wondering where the service is headed. Customers can even vote on which improvements they would like to see. Pretty nice.

Features

With respect to features, DO and Vultr have similar offerings. Both offer reasonably quick setup, usage graphs, automated backups, snapshots, reverse DNS, and a limited API. DO only had a couple deficiencies in this respect. The first was its inability to provide total network traffic usage, although that's easy to overcome by installing vnstat on the server itself. The second was the way it handled kernel upgrades. Kernels are loaded by the hypervisor so a change of kernel is a bit finicky and must occur through the portal, not on the server itself. (This is a correction to the original posting that erroneously attributed this issue to Vultr. Thanks to those who pointed this out.)

But, when it comes to features, there is no comparison to Linode. The Linode feature set can only be described as outstanding. The interface is a little old-style but I cannot imagine a feature that they have not included.

Ease of Use

I found DO and Vultr both very easy to use, but this is probably because their feature sets were limited when compared to Linode. Linode took a little to get used to, but I didn't find the portal too difficult to navigate once I got the hang of it.

Competence

I did have issues with both DO and Vultr. DO had a network issue in NY which meant I could not reach my server for a period of time. DO also had the backup issue I discussed above in Technical Support. Vultr had a kernel failure on the hypervisor managing my server which meant my server was down for a period of time. In both cases, technical support was helpful when I requested an update. In the case of Vultr, I'm not convinced they knew about the problem until I advised them. But they did indicate it was a one-off issue and offered to move my server to a different hypervisor should it recur.

Linode was a different story. In the vast majority of reviews I have read, Linode always came out top. But, during my brief experiment, Linode was the least competent of the three providers. The first server I deployed didn't work. I could connect to the server but could not network out. It took some time, but I eventually discovered the problem was related to Linode's own DNS servers, which the server had been configured to use by default. Their own DNS servers were not accessible from my server. This is bad. But, worse, Linode technical support did not believe the issue until I had provided significant evidence. And, even after they were convinced, they had no solution other than what I had myself had done by switching to Google's public DNS servers. To be fair, they did say they would eventually sort out the issue. But they were not concerned but the problem and, after many hours of back-and-forth discussions with me, they had not resolved the issue.

But that was not the only Linode issue. I initially wanted to set up a 32-bit Centos server but I could not find it on the Linode list. So I set up 64-bit Centos and then opened a ticket to ask how to access the 32-bit version. The answer I received was wrong; I was told how to change the version on the control panel, but that was not possible. I did finally receive an offer to migrate my server to a 32-bit version but by this point I had wasted a lot of time on nonsense. Linode support simply wasn't as good as their competitors.

I advised Linode of my disappointment and frustration and my intent to abort my trial but, despite corresponding with several different support agents, I never received a satisfactory resolution to my complaints. If Linode were truly customer-focused, their support team should have done what DO did: offer me a credit to offset my wasted time, assure me that the problems would be addressed, and then fix the problem.

Please note that, given my frustration with Linode, I only used their server for a single day, even though they offered a 7 day money-back guarantee. Yet despite my disappointment in them, I still believe that Linode has the potential to be one of the best providers out there. It's a shame they're dropping the ball.

Processor

My review of processing power is only qualitative so I'm not going to provide benchmarks; others have already done that before. But I wanted to give a feel for the raw processing power provided by the three companies. And I think I'm well placed to render an opinion since I ran a very high cpu load using the Java VM.

It quickly became clear that Vultr had the best cpu performance but that Linode was very close. DO was way, way behind. In rough terms, Vultr seemed to be 10-20% faster than Linode but Vultr was about 2x faster than DO. Now, for many applications processor power will not be the determining factor ... but it's always nice to know that there's room to grow.

Please take my processor comparisons with a grain of salt. This is based on a very qualitative experience where I looked at the cpu %used as I ran various processes. For the most part I depended on the relatively heavy cpu load provided by a Java VM for my evaluation. This is by no means meant to be a scientific comparison. There are other, more objective, analyses of processor performance (along with disk, network performance and other benchmarks) elsewhere on the Internet. And none of those comparisons shows as extreme a difference as I found during my admittedly limited experiment. I can only conclude either the nodes I received from one or more providers were atypical or that my use case was somewhat atypical.

Disk

All three providers now use SSDs. Both DO and Linode state that they use battery backed-up RAID for disk redundancy. Vultr doesn't publish information about redundancy.

When I asked Vultr the question directly, I was told that they don't go into specifics but that "uptime and data integrity are prime concerns of ours". It's hard to believe that they don't have some form of redundancy in place but, without a clear statement from them, I have to consider the possibility that they don't. While this is a concern to me, it is not a deal-breaker. RAID controllers can fail too so there is always a need for a good disaster recovery plan. The real question is whether Vultr storage fails more than storage at DO or Linode. Unfortunately, that's a question that is almost impossible to answer. It's certainly a negative consideration that should be balanced against the other aspects of the service.

Backups

I am providing a special section on backups because I consider them so important. All three providers offer automated backups for an additional fee. I did not stick with Linode long enough to try their backup system so I can't review it here.

DO's approach to backup is disappointing. They keep the last three backups but they only seem to perform a backup every four days or so. This is simply not good enough. I have read that they have a new backup system, which has apparently been deployed in their Singapore and San Francisco datacentres. For now, in other centres (I'm using New York), a backup every fourth day seems a poor offering.

Vultr's approach to backup is better but still weak. Vultr keeps that last two backups and performs a backup daily. But there is no backup older than one or two days.

Both providers need to provide a proper backup service. A minimum of three backups should be available and the most recent backup needs to be no more than one day old. The age of the backups should be varied; for example keeping backups that are one day old, one week old and one month old. That seems to be the approach used by OnApp and seems more sensible than either DO's or Vultr's approach.

Snapshots

Snapshots are different from backups in that they can be performed manually at any time. I did not use Linode long enough to test their snapshots but I have used snapshots extensively at both DO and Vultr.

Vultr's snapshots work very well. And, at the time of writing this review, they are free because they are still considered in Beta test. I have created snapshots and then spun up new servers from them. Despite being in Beta, they just work. They also do not require a server to be shut down to use them. I have found them to be a great tool, using them to replicate servers and, sometimes, just to create a snapshot of my server at a certain point in time in case I need to revert back to a previous configuration.

DO's snapshots work but have limited use. The server must be shut down before the snapshot can be taken. Because snapshots can take a while (my experience has been between 10m and 1h) the server is down for a significant time. This deficiency can be partially mitigated by converting an existing backup to a snapshot but, because backups happen so infrequently, it's not all that useful.

Pricing

DO and Vultr have very similar offerings. For example, at the low end, for $5/month, Vultr offers 768MB RAM, more than DO's 512MB. But DO offers 20GB of disk, more than Vultr's 15GB. As has already been discussed, Vultr seems to offer significantly more processing power than DO, but for many customers processor power may not be the most significant consideration.

Linode doesn't offer a $5/month plan but it's $10/month plan is quite equivalent to the $10 plans at Vultr or DO. In other words, pricing is not the differentiator any more.

Conclusion

It should be clear at this point that I cannot recommend Linode. And this is a huge disappointment to me because, from all of the reviews I had read, they provide the most mature platform. Given their history, they should also be the most reliable. But, from my experience, they are too frustrating to deal with. That doesn't mean they should not be considered. I only gave them one day to prove themselves, and I hope they improve. But, for now, they're not good enough for me.

Between DO and Vultr, I would have to say that I trust DO more. Their approach to their own failings and their genuine concern for the customer have earned my trust. They state openly that they provide redundancy on their disk drives, a clear advantage over Vultr (which won't comment on their approach). But DO's processors make them somewhat uncompetitive so they are not my first choice, even though I will continue to use their servers for low-cpu applications.

For now, Vultr marginally gets my vote. Aside from their refusal to confirm disk drive redundancy, their offering is simply the best technical solution for me and their support so far has been quite acceptable. If Vultr would openly state their means of providing data integrity then I would not have to hedge my recommendation for their service.

Of course these companies are in a fluid, fast-moving environment. What is true today may easily change tomorrow.

I wrote this review because of all the discussions I have benefited from at LowEndTalk in the past. I hope you found it helpful but please let me know if there is any way it can be improved. I am particularly interested in learning of any errors so that I can address them.

B Regan Asher