Save $7 million on cloud by spending $600k on servers, says 37Signals' David Heinemeier Hansson Spending $600,000 on servers, and more to have them hosted, will save SaaS project management outfit 37Signals over $7 million, according to CTO David Heinemeier Hansson. If that name rings a bell it's because in October 2022 Hansson revealed plans to pull 37 Signals out of the cloud and in January 2023 he detailed the company …
AWS gives you a choice between slow or incredibly expensive. "EBS-Only" is worthless unless your workloads could run fine on a single 20 year old hard drive. The SSD options are limited and expensive. You can maybe get EBS running faster if you pay for huge IOPS upgrades, tune the kernel, disable synchronous writes, and set aside 10+ GB for filesystem caching.
m5d.8xlarge : 36 CPU, 128GB RAM, 2x600 GB NVMe, 10 Gbe networking costs $1.808 an hour. That's ~ $79000 over 5 years for specs that are within the range of a $6000 build-to-order.
So you want more modern specs? How about a i4i.32xlarge? That's ~ $481000 for 5 years. There's going to be some downtime so better get 4 of them for about $2 million for 5 years, just to be safe. Don't forget to reserve in advance to get a little discount and make sure that Amazon doesn't discontinue your hardware after a batch of hardware failures.
Don't even get me started on Amazon's APIs that are throttled to rates one might expect 30 years ago.
Obviously, this will very much depend on your organisation, but here in our org we have bought 3 refurbished servers recently, with 3 year advance warranties with NBD delivery of parts etc. This provides us 84 cores, 576GB RAM, and 36TB of (brand new, not refurb) SSD storage, plus 40GBe networking for less than £10k. Set up in a hyper-converged manner, we will get about 56 cores, 384GB and 24TB of actual use out of them. Doing the same in AWS or Azure? Would bankrupt us.
Sure, costs are higher and profits are high for some clouds. A lot of cost is due to investment in software for keeping things running at scale - some of this you won't need but some you likely will. There's enough competition in the market that running on a cloud is not some 10x more expensive thing for the same product - your workload might be one that's much cheaper self hosted but it might not. Clearly AWS is not the cheapest option out there even if you do decide on cloud but it's unlikely to be the worst value for money either.
There's enough competition in the market that running on a cloud is not some 10x more expensive thing for the same product
That's the problem though. If you run on AWS, you write for AWS. If you run on Google Cloud, you write for Google Cloud. If you run on Azure, you write for Azure.
Sure, Azure may be the easiest for a Windows show like us, but already there are things that will run locally that don't stand a chance on Azure. And to retool all that to run on a different cloud would probably cost more than a year's fees anyway.
Vendor lock in is real, and they're milking it as hard as they can.
Agreed that for just hosting virtual machines, the price-per-performance on AWS (or Azure) just hasn't kept up. 15 years ago having 8-10 VMs with 2-4 vCPUs each was an amazing thing and we all were excited at how easy it was to launch EC2s.
Today that's just slightly above what I could do on my laptop, easily done on a very modest server, but the cost on AWS to do it hasn't scaled down accordingly. Leaving VMs running with the hourly charges accruing just isn't efficient.
Where AWS still shines (and they push customers towards) is serverless computing. Step functions, lambda, and App Runner are where the bursty workloads are designed to be.
But all of this, the maths being discussed here, just makes sense. Renting from someone must always factor in their overheat + profit desires, and therefore will *always* cost more based on any rational argument.
We've been telling World + dog that "cloud" only moves the hardware to someone else's server room. And, that server room is never going to come for free. People should have considered "cloud" as the equivalent of a short-term lease; in the majority of instances, if you settle down into a 'location' (business paradigm in this case, to switch metaphors) it is usually better own rather than continuously pay the landlord.
The server vendor probably has a financing subsidiary that will supply the same gear on OPEX basis. In the case of HPE, there's a packaged offering that does this (HPE Greenlake).
Disclosure: My company is the marketing solutions provider for a Top 10 HPE Greenlake partner.
Buy your own servers? Yes please! So that data sovereignty returns, albeit for financial reasons. I'd prefer control of data to be the driving motivation, but that at least would be a side effect - and an imperative to my mind.
I think "cloud repatriation" is a misnomer[1], "data repatriation" would say what's in the tin.
[1] If returning to one's home country we just say repatriation, it's obvious as nationality is involved. Here we need more, but naming where we're repatriating from makes little sense, other than to use that word! I think I could forgive the buzziness of "cloud" just to get the data back!
Okay brym, we won't tell you that! [Edit: Oopsie! b0llchit just said it!!]
b0llchit: I was misleading when I said "I think I could forgive the buzziness of "cloud" just to get the data back!". I was speaking generally, on behalf of the industry as a whole (over which I have no authority, ofc). No data of mine or any of my clients is CoLo, other than backups for off-site resilience. I've happily walked away from clients that have insisted on using "the cloud" for live business critical data (quote brym "valuable ip") against my strong advice.
And that is the point. Ignorants willingly go to the cloud as zombies and buy into the empty promises of cheap and secure.
At least we are seeing some people pulling out because they realise they have been buying into a dream and castles in the air.
Good computing and infrastructure is hard. You need experts and experience. The cloud promises to solve it all. However, it is only a partial solution and it costs too much when you finally realize what is actually needed. The cloud hype is over. Reality is kicking in.
Outsourcing everything isn't cheaper than running your own kit? Shocked. Shocked!
You know, I'd hazard a guess that having their own inhouse expertise rather than outsourcing the hosting to Deft as well... Having that in-house expertise is an intangible benefit as well (though, the bean counters rarely seem to take that into account).
Honestly, at this point I'm really surprised that any large organisation uses public cloud offerings. For SME's with relatively small numbers of employees, it makes total sense - you don't necessarily have the scale to build out everything you need in house. But when you are spending millions on SaaS/IaaS, I'd happily bet money that you could save a chunk of cash by building out your own setup instead.
Honestly, at this point I'm really surprised that any large organisation uses public cloud offerings.
Hansson did make a qualification, "for a stable workload". If you have a workload that can benefit significantly from scaling up and down it can be cheaper. But it needs to be significant, only needing 50% capacity at night isn't enough.
But none of this is unexpected, it's pretty similar to other rent/buy scenarios.
Stable works are far more expensive as pointed out putting it in the 'cloud'. We had for a short period, some instances in aws (48 core, 256GB RAM), the cost of that was the same as buying the physical servers after 3 months (64 core, 256GB RAM).
We were using it because the person that was doing a project decided they needed it now and just started using it, because the 'cloud' is cheaper and quick to get up. When the cost of it was seen, it was asked, why, you have spent more than it would have been to buy it and install it in our datacenter.
Owning your own server are good for stable workloads and unstable workloads that you don't have the ability to scale yet.
You should to get the most out of your money if you have the ability to auto scale your work load, find out your average load (work out how much it costs), buy the hardware for that and they scale out to the 'cloud' when you exceed your own hardware.
This use-case is the perfect reason to use The Cloud (i.e., "Somebody else's computer"). If the project is only going to take 3 months, it makes sense to rent a computer for 3 months rather than fork over the money up front and then wonder how to re-purpose the computer after the 90 days is done.
Startups, without a lot of capital up front, who want to deploy something quickly are the best use case. Many can only hope they will be around in a 2-5 years and have the luxury of being able to forecast and amortize computer costs over the next 5-7.
The move to the cloud brought lots of short term benefits - particularly the ability to spin up and down instances for testing all kinds of things. I still love that. I also love the fact that I can use multiple cloud instances spread globally to build resilience without the need for the due diligence involved with hunting down co-lo in disparate regions or, even worse, building DC capacity in multiple regions. This is great, it rather democratises computing and delivery power, which again I love.
However, the idea that cloud is the backbone to an organizations operations is a dead-duck I think. The creep out of that is underway and I suspect it will gain momentum over the coming years.
Where does that leave us - with bit of both, leaning heavilly on the in-house, I presume.
Cloud isn't dead, its just finding it's niche.
"The move to the cloud brought lots of short term benefits - particularly the ability to spin up and down instances for testing all kinds of things."
Really? I've always found it easier to use a VM on my local machine, and more recently Docker containers. No need to be connected to the cloud, no unexpected costs and no f*cking IAM complexity.
I work for a small company that hosts our web app on AWS - we spend a couple of hundred bucks a month on lambda, SQL, S3. we don't need a server room or servers; we don't need to worry about downtime or scheduled maintenance, and we know we could scale to 10x overnight if we needed to
Cloudy stuff wasn't always super expensive but with more and more people ( especially our beloved PHBs! ) all buying into the BS that is pure-cloud the providers have you over a barrel once your code and apps are safely locked into their systems, next thing they start pushing up the prices and you will play ball. If not then you can put together yet another multi-million dollar transformation project plan and see if the board of your company says yes yet again to yet another tedious IT black-hole spend-a-thon with no tanigible benefits to the company!
Bezos didn't get to play "rocketman" by giving stuff away!
Caveat as there are several forms of "cloud". IaaS cloud, which is what basecamp is mainly referring to has been super expensive for at least the past 13-14 years probably longer. It's not a big deal if you are very small, i.e. if your bill is say under $1k/mo, big whoop for a typical company. But it's easy to get to $100k/mo for even a small company on IaaS. I saw it myself for the first time back in 2010, with all cloud ROIs being between 5-10 months for bringing(or keeping) stuff "on prem" (aka co-location), and that was giving cloud every benefit that things would operate AT LEAST as good as on prem, which everyone knows would not be the case I think given the availability model of the big IaaS clouds (I have called it "built to fail" since 2011, indicating you need to do more work to cover for infrastructure failures in public cloud by design).
(before anyone tries to imply data centers fail often and you need to protect against that "on prem") -- I've experienced 2 data center failures in the past 20 years of professional co-location, both at the same data center, both in 2007 I believe it was, and I was already planning to move my org out of that facility before the failures hit(I inherited the facility after starting at a new company), it was a badly designed facility. By moving out I avoided a much longer outage at that facility that hit a couple of years later when they had a fire in the power room. So data center failures are extremely rare in my experience. That said, I would probably exclude automatically a good 75% of the world's data centers from my list of suitable locations off the bat just for inadequate design(that includes 100% of the big IaaS player's data centers for the same reason). So choose wisely. I don't consider on site "server rooms" to be data centers, data centers are purpose built facilities(preferably standalone) with dedicated staff, backup generators with fuel contracts, and routine testing of systems.
SaaS can be effective, as the billing is totally different, usually per account or some other metric unrelated to infrastructure. Much easier to justify/not in that case, especially with the value add of the software itself. That is assuming of course the SaaS provider operates the system properly with security, backups, availability etc(and in almost all cases you can never be sure they are doing it properly, just hope they are and they have a SLA that meets your needs).
Unsurprising.
I've posted this before but cloud is very good for lowering Capex. Opex? not so much. For known demand it's little more than expensive hosting.
At a previous job we had a customer who moved their SQL Server production databases in to AWS. Successfully. This is quite an achievement. They then discovered they were Amazon's biggest customer in the country (many million $ per year) So moved nearly all of them back on prem.
I remember organisations (40 years ago) getting a small mainframe, then using Bureau services, then bringing it in house, for control, availability, security and cost. Then "cloudifying it, then bringing it back in house.
Another example was "the main frame is expensive - we can do it cheaper with departmental machines". Then realising this simplistic view does not work - you need multiple machines for availability, you need to put fixes on it!, you need extra staff to support it - you need to do security audits, and backups, you need to manage (and pay lots of money for) software licenses. You need... a whole departmental IT sub-department. You then find your box sits there most of the time doing very little.
You find this is very expensive, so this then swings to having a central site manage all of these aspects and making much better use of people, and resources. As a result to use the centralised services you have to put requests in which get prioritised and there is a few weeks delay, so you get your own machine... or put it in the cloud - till you find the cloud is too expensive,
"you need multiple machines for availability, you need to put fixes on it!, you need extra staff to support it - you need to do security audits, and backups, you need to manage (and pay lots of money for) software licenses"
You do need VM redundancy for safety. The rest, you do not.
You can hire 2 experts on virtualization, run 2 clusters of Proxmox, automate virtually everything by writing scripts. All redundant.
Earlier in my career, the cool thing for , very senior, technically illiterate (and proud of it), managers was to get everything on a Microsoft platform.
The next generation, although not so loudly boastful of their lack of understanding if IT, really liked the idea of getting it "into the cloud".
The current generation, seemingly driven by simple arithmetic, now seeks to reverse that last trend.
Can we hope that the next generation will make the MS platform less universal? That will make nice nested sets of brackets. { { } }
I'm just waiting for about 5 or 6 years when all the cloud providers triple their prices - just after a significant swag of big business and government services have moved everything to cloud and can be held hostage because it's too hard to move back on prem.
Then I'll be able to laugh at all the CIOs and CFOs who thought 'cloud' was better than self-owned hardware.
> but there's enough competition in the market
Competition? What competition? It's basically a duopoly - maybe 3 players at a stretch. AWS 34%, Azure 21%, Google 11%, Alibaba 5% and it just get's smaller from there. The 3 biggest players between them control 2/3rds of the market. 2 or 3 players is just as bad as a monopoly, as when 1 raises the prices, the others will raise theirs as well to just under the one who raised prices first, so that they are still cheaper than that early mover, but they've upped their prices nevertheless. It happens again and again in limited competition markets. Perfect example is AMD/Intel in CPUs and AMD/Nvidia in GPUs. Once AMD started to get market share they jacked up their CPU prices and GPU prices to be similar prices to the incumbent Intel and Nvidia so they can make the same margins as the incumbents. Once they were satisifed with theior market share gained with lower prices - lower margins and profits - they then jacked up prices to increasze margins and profits. I'm not picking on AMD, all corporations do this.
A new player in the market now would have to spend 10's of billions to build up enough infrastructure to seriously compete with the big 3. 'Cloud' is a huge negative income for years due to the large upfront costs to build out the capacity to be actually useful, something big business and government would be willing to shift workloads to. It was hard enough a few years ago when everyone jumped on the bandwagon to try to compete with AWS. But now these are all bigger - capacity-wise - than they were 5 years ago, so it'd be even more expensive to enter the market in any meaningful way to compete.
Think of the size of the workloads that big government departments or companies have, I'm talking social security agencies, tax agencies (HMRC, IRS, etc.), companies like Walmart or Boeing or GE. The compute resources they require in cloud infrastructure can't just be moved between providers willy-nilly. It's HUGE. Only one of the top three would have the available capacity for say a Walmart to pickup lock, stock and barrel and shift to a new cloud provider. Anyone else would take a year or 2 to build out the capacity to accept a Walmart or IRS migrating to them. It'd be even worse trying to insource it again, it'd take 5 years to build up a soverign compute capacity and and the skiiled staff to run it and do an actual migration.
No, in 5 or so years, when these big organisations have lost all inhouse enterprise IT expertise after having moved everything to the cloud, they'll be trapped, it'd either be impossible to move to another 'competitive' (i.e. smaller, limited capacity) provider, or too expensive (in corporate profit/loss cycles or government budget cycles) and time comsuming (in election cycles - that's the next governments problem!) to bring it back inhouse, the big 3 will jack up their prices knowing the big boys who they make their money from are, effectively, trapped.
It can, depending on the circumstances.
Competition has made consumer durables much better over time.
Back when I started driving, cars (and motorbikes) often just didn't work if the weather was cold, and wore out or rusted away after five or ten years. And we thought this was normal. Well, it was normal for British-made hardware. Foreign competition changed that. If you were a Midlands or Dagenham/Luton metal basher then you might not consider this to be a good thing.
Competition has made consumables cheaper. Notwithstanding recent increases, the cost of living is today much lower than it was in my youth. (If you are a farmer or small shopkeeper, you might not consider this to be a good thing.)
What competition can't fix is natural monopolies, chiefly utilities. Allowing for-profit companies to run them just invites abuse.
Natural monopolies exist where the cost of entry to new players is prohibitive. Supplying stuff via underground ducts has been there for a long time. Cloud is there right now. Online search too. Social media, probably.
I'm not advocating nationalisation. I might be persuaded that public ownership is helpful, so long as politicians are not in charge of business decisions. But what's really required is a different kind of company , one whose objectives are set out in, say, a charter, and are not mainly (or at all) about financial return.
-A.
Is it still competition when the government decides who the players are?
Want to work? Need a license
Need to build? Need a permit
Want to operate? Need a permit and a license.
Naturally this is not everything, but it does seem to be cheaper when they are not involved.
Ah look, a "gubbermint is too big and makes things expensive for me" rant.
None of these things prevent competition. What they do prevent is Billy the Useless Builder from building whatever and wherever he likes, without needing to prove any actual skills or qualifications to do so.
This is generally a good thing.
Yes that is correct, that is why I put “Naturally this is not everything”
I will give you an example:
In our state if you install network cables that any one of might be plugged into a camera later you are committing a crime.
When the security industry started using ethernet for cameras it made simple wiring guys criminals.
All installers are required to have a license and it requires 2 years working for a licensed security company before you can apply for a managers license to do the work yourself causing less competition. Hence, you must ask permission before being able to work.
Please ask for context before you judge.
> All installers are required to have a license and it requires 2 years working for a licensed security company before you can apply for a managers license to do the work yourself causing less competition.
How is this any different to electricians or plumbers having to undergo an apprenticeship and earn their 'ticket'?
Just because 100 years ago any bob, dick or harriet could connect their sewage line to the town fresh-water, or set a house on fire or take down the entire neighbourhoods electricity by self-installing electrical wiring doesn't mean that should still be the case and that demonstrated training and education in the subjet shouldn't be required to get a license to do that sort of work nowadays.
Data-wiring is in the same boat. At first anyone could lay data cable and cross connect it to an electical cable and fry themselves, or advertise they are professional data-cable installers and then run bundles of data in high-voltage electrical cabling ducts, but after a while the government decided it'd be a good idea if people display a modicum of knowledge/competence before being allowed to install it. Seems reasonable to me.
By the time the less agile organisations want to move, there won't be the up-to-date skills available to do the job. There will be a whole new generation of infra managers who have never managed on-prem and won't know the pitfalls. In theory everyone piles back into training, but were do the good instructors with the practical experience come from? Vendor book knowledge isn't the half of it.
And the really lumbering giants won't even know their own workloads in any practical terms. They'll pay exorbitantly to relearn the same old lessons the hard way. These are the sorts of organisations whose costs ultimately get paid for by everyone.
There's an interesting/chilling scenario over at:
https://www.consortiuminfo.org/general-news/avoiding-a-cloud-computing-armageddon
While this is an extreme scenario, I have always said that cloud DCs are a perfect target not just for hackers but extremists who love to blow stuff up!! Even more so if they hit several at once like this article states. It's not something people right now would expect but they economic and financial chaos it would cause would be next level.
The advantage of cloud computing is the flexibility. That flexibility comes obviously with a price tag.
If the workload is fairly constant and predictable, you'll probably be better off (at least financially) with your own servers, as long as there's the staff to keep them running.
For workloads that have very high short duration spikes, or are very unpredictable, cloud servers could be cheaper.
Cloud works if your compute demands (including staffing) are bursty. That usually includes SMEs, and a few LEs (eg. Retail at December rush, ticket companies for big events &c.) but, with the exception of signifant new programmes, I think the size of most LEs smoothes those demands sufficiently for cloud to lose its competitive edge.
In addition, I feel that a lot of hosting complexity is down to aircon. I wonder if the real "[from] cloud repatriation" will kick off with liquid cooled commodity servers.
Wait for silicon photonics inside servers - that kind of throughput and compute power close to you will smash the cloud providers initially out of the park. It will be massive driver for repatriation. Intel already know they are approaching the maxims of a copper based bus and have been working on this for years. Once this technology is released and mainstream, cloud providers will need to spend multi billions changing out their compute platforms to compete with this. For business and customers, it will become a case of spend now to get much better performance and near zero latency in our own DCs or we can wait for years for cloud provider X to upgrade all their kit for us to get that performance.
I think I know what a lot of people will do.
I have been telling public cloud evangelists in my market for years that there will be a swing back and that not all workloads will stay in the cloud.
I was constantly told I was living in the past but it's good to see the global industry finally seeing that public cloud is not the answer to everything despite the mass marketing machines of AWS and Azure.
The costs haven't been adding up for a long time unless you are prepared to use or require a dynamic platform that allows for spin up / spin down of workloads.
I built a private cloud of sorts for developers in a place I used to work 13 years ago where this was meant to be the process for them testing code, building code etc.
They just weren't interested and always wanted static servers which semi defeated the purpose of the environment.
My point being a lot of customers still want a good percentage of workloads to be static which cloud is uber expensive for - IaaS is just over priced rubbish.
Another strength of cloud are SaaS offerings and some PaaS offerings.
I also never bought into the whole "we are more secure" rubbish and cloud is just becoming more and more of a target.
I mentioned on another article I commented on, one of the big three cloud providers I reckon will get done with a multi regional, cross tenancy hack which will cause them permanent brand damage.
These environments are so vast with thousands of tenancies there are so many places to hide and launch attacks from.
Cross tenancy vulnerabilities have been found in both AWS and Azure last year and will continue to be found.
Cloud has it's place and is another tool in the box.
Everyone just needs to use it as such and consider it appropriately.
It's not going to bring forth unicorns and rainbows and be the answer to world peace! :)
Money isn't all equal. Its value to business depends on what accountants consider it to be.
The reason cloud exists has little or nothing to do with efficiency or flexibility and everything to do with shifting spending out of capital expenditure and in to operational. Accountants really like that. I'm not really sure why, but then I'm not an accountant. The thing is, most financial directors are.
-A.
Think of a mainframe as several racks of servers and comms, rigged such that you can put a bomb in any rack and the entire workload just carries on running. It's that fault tolerance that attracts the price tag.
I heard a story (which might be apocryphal) of one mainframe that got half-destroyed when the 6kV distribution transformer next door to it went bang. The users didn't even notice. No transactions lost. If it can't do that, I'd argue that it's not a mainframe.
Self-aggrandizing, and "I built Rails to solve my problems, not yours."
Don't take him seriously on anything.
Beyond that, AWS & GCP allow you to do things that are maddeningly difficult to do on your own. Rapid scaling is one. Resilience is another. If you don't need those, then sure, no need to feed their profit margin.
This post has been deleted by its author
This post has been deleted by its author
For bursty stuff - great....
for general run of the mill stuff.... not so much.
I used to work for a company that several 10s of database on SQL.
The thought was that by moving from on-prem (there was no SW Assurance paid for), that they get that for 'free', AND the DBA won't need to do much on a d2d basis, so they can get rid of him.
That rationale was based on 'Azure SQL can auto create appropriate indexes on the fly as it sees fit'. In reality, it did - tonnes of them many of those being duplicates or covering indexes that covered one of the others. Cost of storage went up, cost of backup time went up... so they employed a DBA to clean it up and manage it.
It's a license to print money, and it gives NO benefit in reality.
...and, the benefits too low.
Most people and businesses could get by with their own servers and data management these days. The hardware is good, memory is cheap and the knowledge of how to manage them securely and privately is available.
From a more strategic view, becoming a renter of cloud services means you don't own the service or even your own data or IP once it's out the window. Once your stuff is in the cloud the government or corporation can shut you down completely with a few keystrokes. Think about it. Stuff happens.
The cloud is not always more expensive than running your own kit.
We've cut our data center costs by 40% going to AWS. But if your not managing your cost proactively an manage what I ca;; "Cloud Creep" your going to wake up one day with a massive bill!
Example: When I took over SQL Servers after merging 2 of out operating companies I instituted proper SQL backups. One step was to copy SQL backup files to S3 for long term storage. Surprise! On the next quarter spend audit we saw S3 cost jump up by a big margin. We were not routing S3 traffic over our VPC so we were incurring internet ingress costs on those buckets.
If your not proactive watching where your costs are increasing, if your compute is over provisioned, are you using compute instead of server-less, you costs are going to skyrocket!
There are many services or apps that can help you manage costs. The cloud isn't the answer to every need.
But if your serious about 24/7 availability without a lot of the headaches and unforeseen costs that go along with it, a cloud provider is a far better solution.
TBH there are plenty of environments in my market who have on prem 24x7 availability and have it still cheaper than cloud so to say a cloud provider is a far better solution cost and headache-wise I reckon is a bit of a stretch.
Especially when whole continents for some cloud services have gone down.
As I said in my original post though. if you can make cloud work for you than great - use it.
Managing cost though is not just specific to cloud - you need to do that with on prem as well to ensure there is no overspend.
The economics of opex v capex are just different but hardware vendors through XaaS are now bring cloud consumption opex based costings to on prem solutions.
Hardware vednors are getting market share back with this as it's using the same cost model as cloud but we total environmental control if the customer wants it.
From a development point of view cloud is much simpler. Need a new development sandbox - run the Terraform script (other infrastructure scripting languages are available). Need a message queue, or an RDBMS, or an in-memory database. A few clicks and they are there. OK making such things production ready is more work, but it really improves developer productivity.
If you have a stable environment without much in the way of development, then that's a different scenario. But from where I am, cloud is a big enabler because of the services on offer. It's not all about the tin.
I'm not a great fan of the cloud. I can see the benefits. Easy expansion (just rent more drive space, RAM and CPU cores), and you don't pay to maintain, power and cool the hardware.
But it also has downsides. There is more that can go wrong. If the servers are locally hosted, your servers can be taken out by network problems, power failure, server hardware failure, and possibly a cooling failure. Add in the cloud and you also add your internet connection and potentially hundreds or thousands of miles of internet cabling and hardware between you and your computers. They also have the same potential problems in the data centres.
Now, in a reasonably designed cloud system, there will be redundancies, so you *should* be protected to some extent. But you may not. And what happens if something happens at the cloud provider and they lose a large part of their systems, even if for a few hours? We've had both Amazon and Microsoft lose almost entire continents. You can mitigate this somewhat by maintaining a duplicate of your systems on another cloud provider, but then you have the problems keeping the two in sync.
Still, It does appear to save money. After all, you can get rid of potentially thousands of machines that require power, maintenance (which requires staff) and cooling., and only pay for what you use. Depending on your setup, you may also be able to free up some rooms, even entire buildings, for other uses, or even selling. This looks good on the books, so it seems as though people suggest the cloud because it appears cheaper, and don't really evaluate it.
Even where I work, I was joking with one of our System admins about our move to the cloud, and I half jokingly half seriously pointed out it was single point of failure, and if it did fail, we may be unable to function. My friend looked at me in all seriousness and warned me to stop discussing this. But our head of IT was very definately in the "Cloud=good, on-prem=bad" camp.
There are a number of providers that do a good job of renting older physical servers. My 3 node VMWare development environment at ScaleWay with a basic firewall and S3 services has a month cost that is about the same as my single production EC2 instance on AWS.
It's not just the instance costs - many of the internet-based CI providers have high running costs as all the CI tasks run on AWS instances, such costs can be greatly reduced if they allow you to define instances on your own hardware.
It's a False comparison. Cloud gives global presence, so you can keep customer data within the customer's country - think GDPR. Cloud has availability zones to keep running despite natural disasters. Cloud machines are maintained for you - you're not paying for operators or server rooms, and your not maintaining Disaster Recovery sharing agreements with other companies. DRM can be rendered obsolete with High Availability that can't be achieved with racks of servers.
Kubernetes clusters and Virtual Instances allow for elastic scalability so you don't pay for a VM all month, you pay for capacity as demand increases and subsides.
You can avoid "vendor lock in" by avoiding PAAS services, and instead using licensed software for messaging, database, firewalls, appliances, etc., and take that with you if you jump from one Cloud to another, but doing so brings its own significant costs of maintaining all that software and dealing with installations, upgrades and patches. The Shared Costs of Paas maintaining that for you is part of a trade off.
Cloud offers DDOS protection that you can't acheive from your rack space. It offers continuous data backups and ability to restore in minutes. It offers ability to scale up 100-fold in minutes for 9/11 events.
Everything is a trade off, and each solution has some pros and cons. It is a mistake to look at any one solution and say it's bad because there are some cons. It is a mistake to cherry-pick cons and ignore alternate solution's cons in order to make a false argument.
Cloud is a lot more than just servers yes but then again, so is on prem these days.
Anything you can do in the cloud, you can do on premise if you want.
I'm sorry but that is true.
To a few of your points
- You are still going to need "operators" regardless of whether it's in the cloud or not. PaaS services need to be administered as do IaaS VMs in a tenancy. Depending on your SaaS instance as well, it will still need apps support to a point maintaining your service instance. Ok sure, you don't have to worry about server room or DC costs but unless the cloud provider is doing to end to end support as managed service (which costs a lot more), you still need to do a proper cost comparison.
- I agree warm DR can be rendered obselete with active-active HA but you don't need cloud for that. I have done it myself with on prem solutions and have had plenty of customers who have done the same. To see it's not possible with "racks of servers" is just bollocks.
- You most certainly can have continuous backups with on prem and also the ability to restore in minutes. I have even seen ability to restore in seconds with on prem provided you have the right kit in place. The ability to scale 100 fold quickly does depend on what kit you have available and you are right, this most likely is easier to achieve for a hyperscaler.
- yes with elasticity and having your workloads designed to work with this in cloud can make this more cost effective. But to be very frank, you can do this with on prem solutions as well with the right tech in the mix and therefore optimising your on prem costs.
While some people here are being critical of IaaS (and very rightfully so) a lot of the criticism is also more widespread than that. As I have said in other comments on this thread, everything does have it's pros and cons and that includes public cloud and on prem solutions. In my opinion there is way too much evidence globally that customers are paying way too much for public cloud for the services they are getting. Bill shock is quite common and as the CEO of 37 Signals is finding, it can be cheaper to bring it all or some of your services back home. However, there are situations where cloud can be the right option. Just treat it like another tool and do the costings for it just like everything else.
The "we are more cheaper" catch cry is just rubbish and IMO, the public cloud provider salespeople need to stop using it as it's starting to cause brand damage. As I said before as well, the "we are more secure" sales pitch also needs to stop. Plenty of public cloud breaches in recent times. Okta copped a couple last year and what about the classic LastPass hack? Last their ENTIRE global DB. We are going to here about these more and more in the next few years.
I have often wondered why some companies use the cloud. OK, if you happen to be Spotify or Netflix there are massive benefits to be had from accessing an existing global, scaleable network, but for companies like us, a small SaaS enterprise software supplier, it's too expensive.
We buy dual CPU AmD boxes, last lot were running 7H12 64 core chips, x2, with 768GB RAM and 4x 10Gig network cards. Get 8 or 10 of those (even refurbished) in a VMware cluster and you've got fantastic performance, immediately available, and fantastic resilience if you choose your colo partner well. Add a couple of SANs with replication, some juniper switches and Cisco firewalls and you have your mini AWS in a UK data centre and you have full control over all configurations and costs.
You still need a couple of people or more to manage it, same as you do with AWS/Azure/Google clouds, but it's all cost controlled, no network fees for larger customers using more bandwidth, your database traffic stays where you put it and doesn't cost you $$$$ when traffic rises.
I use AWS for off site storage, it's not very cheap but it's reliable and available and fast. A DR site in another colo DC gives me the ability to switch over if necessary. We don't need the ability to switch in 10 seconds, we want to be sure the problem is significant before switching to DR so a 4 hr window is plenty.
I know AWS has a number of bolt ons like fast emailing and SMS services which we do use, you don't need to be located in AWS to use these additional services.
Too many companies rushed into the cloud without first properly evaluating the costs and I think a lot more will be coming back to quality colo facilities.
And for this it's about damned time. It's no different than renting vs owning a house - a renter pays the costs of the house just like a homeowner does, plus a profit to the owner. It does not save money to rent vs own. In the case of data, you don't even own your data anymore, it belongs to the host. Once it's in their clutches they can use it for their own needs and the minute you're no longer profitable enough (notice I didn't say profitable, I said profitable ENOUGH) then click, and your business is gone. Cloud providers with deep pocketed investors have been subsidizing the actual cost to suck people in, and now they have your data so you will start paying more and more to keep it. The hosts will do this frog in a pot style, by slowly raising costs until the frog is done. This company going back to hosting their own is a step in the right direction and I hope others follow.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
