This last summer everything changed when Crashplan cancelled it’s unlimited service for Home users.  It was a shock, to be sure.  The only conclusion I could draw was that Crashplan could not be profitable in competing with the likes of Backblaze.  It is this conclusion, along with one other recent event, that made me realize that Backblaze is the only backup service for me.

It’s not that Backblaze’s backup is the best.  There are other backup systems that are faster and/or more sophisticated.  And it’s not that Backblaze’s backup is the cheapest.  Though it is the cheapest unlimited backup service out there, it’s not cheapest by much.  For example, Crashplan had recommended that its users migrate to Carbonite, a service that, at $60/year, is only $10 more than Backblaze.  It’s that, as a company, Backblaze is one of the most open technology companies out there.  And it seems to be genuinely interested in ensuring, not only that its customers are properly backed up, but that they are also able to restore their data at crunch time.

So why is Backblaze such a good company?  I have already alluded to its transparency, something which is becoming very important in the online backup space.  Backblaze has always shown how it builds its backup systems using pods and vaults.  Early in its life it even published the hardware specs for its backup pod.  It has also shown how much it costs to build its backup systems, presumably in order to allay any fears that its low prices were unsustainable.  In my opinion, this strategy has worked well.  Crashplan may have closed its doors, but there are no signs that Backblaze will.  And we have many of the numbers to prove that from Backblaze itself.  This transparency has also given its customers confidence that Backblaze is reasonably competent.  It’s difficult to disseminate so much internal information about yourself without giving outsiders a glimpse into your soul.  And so far Backblaze’s soul looks pretty darn good.

Yet, despite all the good things I had to say about Backblaze, I didn’t post my change of heart this past summer after Crashplan closed its doors.  So what is it recently that made me take the time to affirm my support of Backblaze?  It was a an e-mail I received from the company this week.  I use two-factor authentication for all the services I use and Backblaze is no exception.  I received an e-mail warning me that, although I was using two-factor authentication, I had not created backup keys or enabled SMS fallbackup in case I lost my two-factor device.  I was surprised when I received the e-mail.  First, I was shocked that backup keys or SMS fallbackup settings were available.  (In my own defense, I do not believe they were available when I enabled two-factor authentication.)  But, more importantly, I was shocked that Backblaze would have had the forethought to warn its customers about the possible pitfalls of a single point of failure like two-factor authentication.  The e-mail made very clear that if I lost my two-factor device, I would be unable to recover my backed up data.  Perhaps Backblaze had sent out the e-mail because some customers had lost their backups as a result of a similar oversight.  Perhaps backup keys and SMS fallback settings were added because of those same customers.  It doesn’t matter.  Backblaze recognized a shortcoming in their customers’ configurations and decided to advise them.

Wow.  It really doesn’t get better than this.  And that’s why I now think Backblaze is the best online backup company out there bar none.  And that’s why I have expanded my backups to also use their B2 service.  And that’s why I’m sticking with them.

The best of breed

For online backups, PC Magazine currently recommends IDrive, Crashplan and SOS Online Backup.  However, when I first began using an online backup service almost six years ago, Backblaze was recommended by many of the more reputable reviewers.  So it has been Backblaze that I have been using ever since to back up my primary desktop online.

Backblaze

Backblaze‘s main selling point is simplicity, but I have always suspected that simplicity could be a shortcoming.  Simplicity can sacrifice flexibility and may even cause problems by hiding underlying behavior.  Yet I have stuck with Backblaze because it supported the key features I required: (1) security through a personal encryption key, (2) a fixed fee for unlimited storage, (3) reliability, and (4) decent customer service.

Unfortunately, I have had cause to re-examine my choice of Backblaze.  I recently noticed that an entire directory of irreplaceable data was not being backed up.  Ever.  After working through the possible causes, I finally discovered the problem.  The missing directory was exclusively composed of log files (with a .log extension) and it turned out that Backblaze skipped log files by default.  This decision was both unexpected and unfortunate.  After all, many of Backblaze‘s default settings, such as the omission of program files (eg. .exe, .dll on a Windows computer), made perfect sense.  Restoration of these programs files would normally be done through re-installation so a backup would not be required.  But log files are data files and, in my case, contained critical and irreplaceable data.

A second issue I discovered once I enabled the backup of the log files, was that these files were not being compressed during backup.  This meant that the backup of these files took more than twice as long as it should have.  I subsequently contacted Backblaze support detailing my findings and was shocked to learn that Backblaze does not compress data at all.

As an aside, I should add that I have recently begun evaluating Backblaze‘s B2 service, which is not an unlimited desktop backup service, but rather is meant as an alternative to Amazon’s S3 storage service.  It still has some bugs but, at $0.005/GB/month for storage, it is a very interesting alternative to Amazon’s S3 and its other competitors.

Despite my recent issues, Backblaze is still a good online backup solution and a good option for many people.  But, for those of us who don’t mind a little added complexity, I no longer believe it is the best choice.

CrashPlan

So what’s the alternative?  Six years ago, when I had selected Backblaze as my backup service, a close runner-up had been CrashPlan.  It currently costs 20% more than Backblaze ($59.99 vs $50.00 per year) but for the extra cost CrashPlan will keep different file versions and deleted files forever.  (Backblaze only keeps this historical data for a maximum of 30 days.)  CrashPlan also backs up all files by default and provides a slew of configuration options that Backblaze does not provide.

What’s better/worse?

Before committing to switch to CrashPlan, I wanted to identify those features I would be adding/losing from Backblaze.  Aside from the issues listed above, the main differences between the two services relate to the process of file restoration.

One Backblaze feature that I will miss is the ability to order a flash or hard drive containing the entire backup.  I have never had to use this facility (for which Backblaze does charge a fee) but I can imagine saving a lot of time in case of a catastrophic failure.  No matter what internet speed you have, a mammoth download will take time.  CrashPlan does not offer this facility.

As convenient as ordering a flash or hard drive may be, restoring data through downloads with Backblaze was somewhat painful.  The online selection of directories was slow to load and the restoration of files was only possible by downloading a zip file containing the data.  And the zip file could take a while to create.  This was a cumbersome way to restore a lot of data.  CrashPlan, by contrast, provides a more direct means of restoration through their application or their online portal.  And, unlike Backblaze, CrashPlan does compress data on upload, potentially making CrashPlan much faster, especially for the crucial initial backup period.

CrashPlan also supports more than online backups.  It supports backups to local or network disks and backups to other CrashPlan clients.  And it provides these facilities for free.  So if friends or associates living in different locations all have nice large disks, they can provide backups for one another without having to pay a penny.  It also provides support for Linux and offers a family plan for up to ten computers, neither of which is currently available with Backblaze.

In my initial testing of CrashPlan I concluded that it provided all of the features I required.  It even provided a few extra features that Backblaze did not provide, like an encryption key generator if a personal encryption key was required.

The only real unknown would be CrashPlan‘s reliability and customer service.  Over the years Backblaze never seemed to go down and its support team usually replied within a business day, so I hope that CrashPlan does the same.  My fingers are crossed.

Time to switch

So my plan is to switch to CrashPlan as I near the end of my current Backblaze contract.  I still consider both services to be good value compared to the other major online backup vendors.  But I now think that CrashPlan is the better option of the two for anyone willing to spend a little extra time on configuration.  The chances of getting exactly the backup result desired with CrashPlan seem higher and the backup speed should be faster.  By skipping some files and directories by default, Backblaze made it possible for the user to be lured into a false sense of confidence.  It might work for some simple situations but it’s probably not the best approach for many. CrashPlan, here I come.

Before I present my first impressions of Microsoft Azure, let me begin with a disclaimer:  I don’t like Microsoft.  With the exception of Excel, I think its products have generally been inferior to and provided less value than its competitors.  Although its has released better versions of its operating system of the years (XP and 7 in particular come to mind), it continues to regress on other versions (witness the fiascoes of Vista and 8).  Compared with other worthy Unix-style competitors like Linux and Apple’s OS X, Windows is a substandard product.  Moreover, Microsoft’s recent shenanigans in trying to force users to upgrade to Windows 10 are the antics of a wounded gangster.  I think the company itself is a bully, using its considerable resources to silence its critics and competitors through unfair or unethical business practices.

But today Microsoft is fighting for its life.  The desktop computer is disappearing (at least for consumers), Windows Phone has failed and Windows 10, despite being free, has not  gained its planned market share.  So why even consider Azure when Microsoft’s flagship product is a dying entity?  Because Azure is a very different beast for the company, created to compete head-to-head with the best in the online world.  Azure is Microsoft’s answer to Amazon AWS, Google Compute, IBM Softlayer and a host of smaller Cloud competitors like Digital Ocean, Vultr, and Linode.  So when I saw that Microsoft was offering a free one month trial, I figured I should give it a try.  And I was determined to do so without preconceptions.

My Points of Reference

I currently use Digital Ocean, Vultr, and Google Compute.  Although I prefer Digital Ocean and Vultr, Google Compute offers a closer comparison Azure.  (For a summary of Digital Ocean and Vultr, see my qualitative comparison on LowEndTalk.  For more information on Google Compute, see my blog review here.)

I will therefore be comparing Microsoft Azure to Google Compute.  In the near future I will also consider evaluating IBM Softlayer and Amazon AWS, both of which seem to have offerings similar to Azure.

The Azure Portal

At the time my trial began, Azure had already rolled out a new portal, though its “Classic” portal still existed.  I found the new portal to provide a very slick interface, set up with tiles à la Windows 10.  Customization of the main portal page was very easy using a drag-and-drop process.  Compared with Google Compute, the interface itself was very nice (and pretty).  Unfortunately, that was about the only thing going for it.

Everything on the portal seemed slow compared to Google.  In particular, creating instances (ie. Virtual Machines) seemed to take forever.  There were a zillion options to configure (although that was true with Google as well), but the process to create an instance seemed unnecessarily awkward.  Since I was only interested in Linux instances, perhaps some of my complaints wouldn’t have been issues with Windows machines.  Yet, if this turned out to be true, then it would contradict Azure’s claims to fully support Linux.

I typically run Centos 6.7 instances, a popular server option.  Yet even the creation of these instances wasn’t straightforward.  I couldn’t even find it on the Microsoft list of supported servers.  I then tried to find it using a search, but eventually discovered I had used the wrong search box.  I did eventually find the image but, by this point, nothing in the portal seemed easy to do.  Google Compute’s interface was also complex, but somehow I had an easier time finding what I needed there.  Yet, these sorts of issues are teething pains; eventually, once I got used to creating my instances, everything did seem more straightforward.  But it was a more difficult adaptation than at Google.

Is this Beta?  Or Alpha?  Or Development?

I had such fundamental issues with Azure that I had to conclude that it was not yet a production system.  Take the creation of a Linux instance.  The portal advises that an instance can be created in two flavors:  Classic or Resource Manager based.  Even though neither of these versions are defined during the creation process, the portal recommends creating Resource Manager based instances.  Fine.  I followed the advice of the portal.  But there’s a catch.  After my instances were up and running, I then wanted to create a backup of my newly minted servers.  So I followed the instructions on how to create a Backup Vault and how to discover my existing instances (which involved a rather convoluted process).  But, no matter how carefully I followed the instructions, I could not create a backup.  And the error messages were not at all helpful; they simply said that (1) I needed to discover my instances first and then, (2) that my instances could not be found.  After much searching deep in the bowels of the Azure documentation, I found the following statement:

Backing up Azure Resource Manager-based (aka IaaS V2) virtual machines is not supported

Nice!  So one of the key benefits of a cloud-based Virtual Machine, access to an integrated backup system, isn’t even supported using the recommended deployment method.  Furthermore, even for Classic instances, the backup functionality isn’t available at all in the new portal.  I was always switched back to the Classic Portal to deal with backups.  It was a frustrating and irritating experience.  Honestly, Azure just doesn’t seem ready for prime time.

There are also other limitations.  The Centos 6.7 image that I selected was 30GB in size but the disk on my server was 50GB.   As a result, I ended up with only a 30GB disk.  I had had a similar problem on Google Compute, but Google’s documentation clearly explained how to recover the lost space.  I could not find a similar document on Azure so I needlessly lost 20GB.  Not good.

A Handy Estimate

I was, however, very excited to see a tile called “Estimated Spend”, which would presumably have helped me keep my costs in check.  Except for one thing: it doesn’t work.  After several days of letting my instances run, these tiles still said “Loading…”

Very slow.  Very expensive.

This quick review is a capture of my first impressions of Azure and isn’t meant to be an exhaustive analysis of Azure’s offering.  In fact, it’s not meant to be a numerical analysis at all.  But, even without crunching the numbers, I can tell you that instances with similar specifications at Google Compute and at Azure run very differently.  In a nutshell, Azure is significantly slower.  And machines than ran satisfactorily at Google kept running into difficulties on Azure.

Because I was running with a $250 free trial, I decided to increase the Azure specifications to the point that my instances ran in a satisfactory way and I discovered that Azure would then cost over three times what Google Compute cost.  (If  you read my Google Compute review you will know that it is not particularly cheap compared to other alternatives.)

A few specifics might help.  I use two different kinds of Virtual Machines, a single core machine and an eight core machine, but I only was testing the single core machine on Azure.  My machine ran fine on Google Compute using a g1-small instance (1.7GB and about half a computing core).  Google’s price for this machine was about $15/month.  I figured the equivalent machine on Azure would be the Basic A1 machine (1.75GB and one core, although I don’t know what cpu percentage was guaranteed).  Azure’s price for this machine was about $27/month.  But, even at this price, the Azure machine had problems; in particular, it couldn’t keep up with the load.  So I tried what seemed to be the next tier in terms of cost, which was the Standard D1 machine (2.7GB and one core).  This machine cost about $50/month, over three times the price of the Google Compute machine.  It was a better machine than the Google Compute machine but it was overkill for what I needed and it was certainly not worth three times the price.

Now, I’m not saying at all that Microsoft’s cloud really costs three times what Google’s cloud does.  But, given the coarsely bundled nature of the instances, in order to get my particular servers to run properly I had to spend three times what I had previously been spending at Google.  Pricing aside, were Microsoft’s instances at this price better?  Yes, I believe they were.  But I didn’t need the extra performance and I certainly didn’t need the extra cost.

No Support, Just like Google

Like Google, Microsoft offers no free technical support for Azure.  If they had, I could have asked some very probing questions and perhaps gotten some answers.  Smaller competitors, like Digital Ocean and Vultr, do offer technical support, which is one reason I like them better.  Personally, I think the no-technical-support paradigm doesn’t work.  It’s certainly bad for the customer but I think it’s also bad for the provider since it destroys customer feedback.  If something doesn’t work as advertised, do Microsoft and Google not want to know about it?  And do they not want the opportunity to keep a confused and frustrated customer?

One Extra Feature

In the interest of fairness, I did discover one extra feature that Azure offered on its instances without charge that Google Compute did not: a Resource Disk.  After reboots or crashes it’s not guaranteed to preserve data, but it’s very handy for big downloads or for temporary storage.

Conclusion

In the end I got my instances to work well on Azure (except for backups), but this success came with a very steep learning curve and at a very high cost.  So, in that sense, Azure works.  It also gives a no-charge scratch disk, which is nice.  And a nice looking portal (that doesn’t really work properly).  It’s the kind of the offering I would have expected from an undergraduate-led startup.  From my experience, I can’t image anyone using Azure who doesn’t have some sort of vested interest in working with Microsoft, or with Windows (as some companies do).

So I have two separate recommendations.  If you have to use Azure or have to work with Microsoft and you don’t mind a steep learning curve and overpaying for your resources, then Azure may well work for you.  However, if you are not particularly tied to that ecosystem and, in particular, if you want to use Linux, I highly recommend you seek another provider.  There are many good providers out there.  And, from my experience, most of them are a whole lot better than Azure.  And, in many cases, much, much, cheaper.

Because of its size and all-encompassing interests, Google would appear to be the 800 pound gorilla in Google/Apple/Microsoft triumvirate.  But that’s not quite right because we’re missing someone.  Google is actually the 800 pound gorilla in a quadrumvirate that includes Apple, Microsoft and, also, Amazon.

Who’s Evil?

Years ago, Google’s motto was “Don’t be Evil“.  So far as I can tell, Microsoft was always evil.  I don’t know if Apple is evil but, with its profit margins, it can certainly afford to be.  And Amazon?  I think the jury is out there too.  Don’t get me wrong:  this quadrumvirate has given us all a lot of stuff.  “Google” is new a verb.  “Windows “is a common punch line.  “i[placeYourDeviceHere]” is a new class of noun.  And Amazon now sells more stuff than Wal-Mart.  The first three of these companies will let us merge our phone/table/laptop/desktop and virtual existences into their respective cyberworlds.  They all claim to be concerned about security and privacy and even now, benevolently, even suggest that connectivity with their counterparts is a worthy goal.  But what is the reality?

Since this blog entry is really about Google, its purpose is to highlight some of the issues I have discovered in the last several years working with Google and its products.

Google’s Products

Google is a company that had been my hands-down favorite of the four biggies just a few years ago.  But, having now tried a few of Google’s paid-for services, I now deal with Google with some trepidation.  Of course, like so many others, I have long used Google’s search engine and GMail.  But I have now also used Google’s Android, Adwords and Compute products, the last two of which are paid-for services.  My new single best word to describe Google’s corporate culture is “arrogant”.  I have never come across a company which seems to honestly believe that it can make no mistake worth addressing.  The best example of this attitude is with Google Compute.  Compute is Google’s public cloud, which can be rented on demand.  It is broadly equivalent to Amazon’s AWS and Microsoft’s Azure.  There are other, smaller companies, that also provide similar services like Digital Ocean, Vultr, Linode, Rackspace, Softlayer and many others.  (See also my review of the first of these at LowEndTalk.)  But Google does something bizarre when it sells computing resources: it offers no access to its technical support group.  If something goes wrong with the service, a standard customer cannot contact technical support.  At all.  If you need to report a problem, or complain about a charge, you can only contact billing support.  And if you really do want to contact their technical support group?  Well, you can.  For a minimum charge of $150 per month.  Even if the technical issue if Google’s fault.  I think that’s shortsighted.  I think that’s arrogant.  And, yes, I think that’s evil.

Arrogance

But what if Google is right?  What if it is perfect and nothing ever goes wrong with its system?  What if technical problems are always the fault of the customer?  Well, having tried Google Compute’s free trial for 60 days, I can assure you that they have the same problems as everyone else.  Virtual machines sometimes don’t start.  Networking sometimes doesn’t work properly.  And, most common of all, the client application called “Google Developers Console” gives delayed or erroneous information and hangs up regularly (even when using Google’s own Chrome browser).  As a customer of many cloud service providers, I accept that none of the platforms is perfect and I would happily work with Google to improve its platform.  But not if I can only contact billing support.  Google has chosen a bizarre way to deal with a real or potential customer.

The shame of this indictment is that Google’s offering is, overall, really quite good.  It’s cheaper than Amazon’s or Microsoft’s offerings and seems to provide similar service levels.  As a cloud service, it’s still not as cheap as some of the lowest cost dedicated server providers like, say, Hetzner.  But Hetzner’s dedicated server offerings don’t provide the myriad of advantages that the cloud providers offer.

Other services

What about other Google products and services?  I recently switched to an Android cell phone.  It wasn’t my first, but it was a big improvement on an Android phone I had several cell phones ago.  My previous experience was with Android 2.2.2; now I have Android 5.0.2.  At about $100, my new Motorola phone is inexpensive but provides excellent value.  What most impressed me, however, is how well it handled the transition from my Windows 8.1 phone.  In a way it was just a lucky coincidence that I had used GMail on my Windows phone.  But, because of that happy accident, transitioning my contacts, including phone numbers and e-mails, was all but trivial.

Google == Microsoft?

Google is beginning to behave more like the company it said it would never be: Microsoft.  I discovered to my chagrin that Google+ is beginning to permeate everything I do.  When installing a game on my Android phone I had to set up a Google+ account, for which I decided to use an alias.  I was dismayed to discover that my desktop browser then somehow magically changed its identity to the new alias.  There was no warning of this.  And undoing it turned out to be a nightmare, taking a full day of searching the internet for solutions.  During some of my efforts I even received a critical Google server error that said “A team of highly trained monkeys has been dispatched to deal with this situation”.  I took this cryptic error to mean that Google would have to investigate the issue and fix it manually.  Eventually I managed to undo the changes caused by selecting a Google+ alias, but I still don’t exactly know how I did it.  I don’t consider myself either stupid or technically illiterate so I can only imagine how this would affect a more typical user.

Competition!

Don’t get me wrong, I don’t hate Google.  But I’m not as enamored with the company as I used to be and I worry that it is morphing into a Microsoft-like company.  I hope not.  I hope it’s just part of its growing pains.  But I’d love to see a new, nimbler, company take on the quadrumvirate of Google, Apple, Microsoft and Amazon.  I think that’s what we need at this point.  I think it will make Google a better company and introduce the customer-focused innovation that we seem to be losing.  Let the games begin!

I recently discovered that Google was offering $300 in credits to those who want to evaluate its Cloud Platform, an offering which includes the Google Compute Engine.  Even though the credits expire after 60 days, I decided it was worth spending a little time to test drive the platform, to see if it was all it claimed to be.  I was surprised, both positively and negatively; but, on the whole, I found the Google Platform to be a disappointment.

The Project

One of my projects requires significant computing resources in order to execute simulations, some of which can be run concurrently.  For this project I have, until now, only found two decent and cost effective computing resources: (1) my own Intel i7-4770 desktop, and (2) Intel i7 or E3-based servers rented from Hetzner in Germany. I have found my desktop and the Hetzner servers to run the simulations in roughly the same amount of time, as would be expected from their similar specifications.  But since my simulation requirements are not continuous, I would prefer to have on-demand cloud computing available.  Cloud computing would also enable me to run many simulations concurrently.  Setup would also be easier since Hetzner does not provide snapshot capabilities, meaning that the setup of each server is a manual operation. In contrast, servers in a cloud environment can be quickly replicated from pre-existing snapshots, providing significant time savings.

Google’s Interface

Google provides more than one way to interact with its cloud but seems to recommend the command line interface called gcloud, which I have yet to try.  Instead, I have been using the perfectly adequate Google Developers Console. Through this console I was able to  create, manipulate and destroy servers, disks and other objects.  I found the Google Console to be reasonably easy to use and quite flexible.  For example, it was relatively easy to create a server, take a snapshot, and then create replicas of the same server through the console.  It was also possible to create a stand-alone disk and then attach it to a running instance.  I found this feature quite helpful when I was running out of disk space but wanted to continue with the same instance.

Snapshots are very easy to use and usually quite fast.  Google claims that the snapshot is frozen in a fraction of a second (see their blog) and that the any remaining wait time (in my experience of the order of a minute) is the time taken to redundantly copy the snapshot to storage to different geographic locations.

When a server (which Google calls an instance) is created, it is possible to flag it as preemptible, which greatly reduces the cost.  (Google defines a preemptible instance as one that will never run for more than 24 hours and which Google can shut down with a 30 second notification at any time.)  The idea to allow customers to use Google’s spare capacity without any commitment on Google’s part.  From what I have seen, a preemptible instance typically costs just over one quarter of what a regular instance costs, a very significant savings for someone running simulations which are not time critical.  To make real use of these instances, however, the user must be able to save the instance’s state on short notice and then resume on a different instance later.  But, again, the savings are considerable.

Although I found the Developers Console full of features and easy to use, it was not as robust as I would have expected from Google.  I regularly found my browser window freezing up, requiring me to close the window and reopen it.  (It’s hard to excuse Google and blame the browser, since I was using Google Chrome.)  This was really just a nuisance, but I did find it surprising coming from Google.

But, all in all, the interface was well designed and an easy way to interface with the Google Cloud.

The Negatives

Unfortunately, the experience with the Developers Console was not a good proxy for my overall experience with the Google Cloud or with Google.  I found some of Google’s claims to be exaggerated and I found the costing to be very high when compared to my main alternative, which has been Hetzner.  Google likes to compare its offering to Amazon but, having never used Amazon, I cannot make that comparison.  The only other comparisons I might be able to make are with respect to my preferred cloud providers, Vultr and Digital Ocean.  (For more information on these two providers, you can read my qualitative review at LowEndTalk.)  But these comparisons are not not strictly speaking fair, since Google’s offering is much broader.

Let’s begin with a two of Google’s claims which irked me.  Consider this proud claim:

We provide all of our customers with Bronze support giving you access to online documentation, community forums, and billing support.

Strictly speaking, this is true.  But what’s the point of proudly announcing that all customers get Bronze support when the key element is missing?  Notice that there is no technical support.  At all.  I only discovered this after I created an instance but could not connect to it.  I tried everything I could think of for about 20 minutes and then destroyed the instance.  I then created a new instance in exactly the same way and discovered that it worked fine.  I concluded that there was some sort of Google glitch.  I then wanted to (1) advise Google of the problem, (2) find out if it could be fixed in the future, and (3) find out how to get the charges reversed for an instance that didn’t work.  But there is no way to contact Google technical support at the Bronze support level.  None.  I’ve never heard of such a thing.  To me, it’s the height of hubris.  It’s as if Google thinks it doesn’t make mistakes.  I don’t know  how to deal with a company that operates this way.

To be fair to Google, I did contact Billing Support with my issues.  I was told that advising Google of technical issues through Billing Support was acceptable, that but I would not necessarily be advised of the outcome of any investigation into the problem.  I was also told that, since I was operating during a trial period, no accounting adjustments would be made.

The only way to get technical support (ie. the ability to open a support ticket) is to upgrade to Silver support at a cost of $150/month.  Not impressive.

And the second claim?

The probability that Compute Engine will terminate a preemptible instance for a system event is generally low, but may vary from day to day and from zone to zone depending on current conditions.

Yet when I was running multiple preemptible instances in different zones one day, I found that more than half of my instances were preempted.  This can happen with a preemptible instance, but I had similar problems on other days as well.  So I take issue with the statement that the probability of a system event is “generally low”.

Finally, my third and last negative is cost.  In order to run a Google instance equivalent to one of the Hetzner servers, I found the pricing to be between 1.5 and 2.0 times the price at Hetzner.  If this were the whole story then I would think it was a reasonable deal: spend 50% to 100% more to have an on-demand server with no commitment.  But this calculation is depends on the Google server being preemptible.  And I have already noted that I was being preempted quite often.  A sustained-use server would therefore cost between 6 and 8 times the price of the Hetzner server at standard rates.  Even with the 30% discount Google offers for full sustained usage, Google is 4.2 to 5.6 times the price of Hetzner.  It’s a premium that is very hard to justify unless the server is only running for a small fraction of the month.

Conclusion

The Google Compute platform offers many desirable features but I think it’s just too expensive.  I will consider leaving my account open after the trial, paying to keep my server image available in case I need to run a simulation once or twice a month, but I can’t see using the platform for any significant work.  It’s just so much cheaper to go to Hetzner (or on of its direct competitors) or to simply buy and run the hardware myself.

I’m also worried about the exaggerations in Google’s marketing materials and the arrogant approach Google is taking to customer support.  I wouldn’t want to depend on a company that thinks that minimal customer service is an add-on.