Save $7 million on cloud by spending $600k on servers, says Basecamp's David Heinemeier Hansson
- Reference: 1677045490
- News link: https://www.theregister.co.uk/2023/02/22/cloud_repatration_savings_calculated_basecamp/
- Source link:
If that name rings a bell it's because in October 2022 Hansson [1]revealed plans to pull 37 Signals out of the cloud and in January 2023 he [2]detailed the company's $3.2 million annual AWS bill.
On Tuesday Hansson again hit the keyboard to [3]share his analysis that quitting the cloud will save the organization he serves $7 million over five years.
[4]
He explained that $2.3 million of the cloud bill went towards "app servers, cache servers, database servers, search servers, the works."
[5]
[6]
Hansson intends to replace cloud servers with eight servers "running dual 64-core CPUs (for a total of 256 vCPUs per box!) in each datacenter or 14 machines running single-core CPUs at a higher clock frequency."
"We need to add about 2,000 vCPUs per datacenter, and we run in two datacenters, so 4,000 vCPUs for performance and redundancy," he added – though he cautioned that these are "rough numbers."
[7]Cloud customers are wasting money by overprovisioning resources
[8]What's driving multicloud? War, regulation, plague, says Acronis CEO
[9]Hybrid multi-cloud is a mess to clean up, not an innovation to excite
[10]Lessons to be learned from Google and Oracle's datacenter heatstroke
All of the above will cost about $600,000, but those costs will be amortized over five years.
But wait, we hear you say: what about power, cooling, connectivity, and all the other stuff that's needed to keep servers alive?
[11]
Hansson has that sorted, thanks to hosting outfit Deft. When the CTO factored in those charges, his back-of-envelope sums yielded annual costs of $840,000 – compared to annual cloud compute costs of $2.3 million.
He did his sums and decided: "In round numbers, let's call it saving a million and a half dollars per year." Next, he tossed an extra half a million dollars into his budget to cover unforeseen expenses over the period, and still found a projected seven million dollars of savings over five years.
And he reckons even larger savings might be possible, because "we have lots of boxes still running at seven years."
[12]
His takeaway from his planning exercise is "Any mid-sized SaaS business and above with stable workloads that does not benchmark their rental bill for servers in the cloud against buying their own boxes is committing financial malpractice at this point."
Mid-sized SaaS vendors with stable workloads are not an enormous constituency. But there is [13]plenty of evidence that other cloud customers are worried about their bills, hybrid multicloud is increasingly being advocated as the new normal as organizations use it to avoid going all-in on public clouds, and "cloud repatriation" has become something of a buzzword.
Hansson might be onto something here. ®
Get our [14]Tech Resources
[1] https://www.theregister.com/2022/10/20/basecamp_decamps_from_cloud/
[2] https://www.theregister.com/2023/01/16/basecamp_37signals_cloud_bill/
[3] https://world.hey.com/dhh/we-stand-to-save-7m-over-five-years-from-our-cloud-exit-53996caa
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/paasiaas&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Y-X12IU@3Rcez5vgkt009QAAAFA&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/paasiaas&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Y-X12IU@3Rcez5vgkt009QAAAFA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/paasiaas&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Y-X12IU@3Rcez5vgkt009QAAAFA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2022/12/07/cloud_customers_overprovision/
[8] https://www.theregister.com/2023/01/19/hyperlocalization_of_it_infrastructure_is/
[9] https://www.theregister.com/2022/12/23/hybrid_multicloud_platforms/
[10] https://www.theregister.com/2022/08/22/google_oracle_heatstroke/
[11] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/paasiaas&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Y-X12IU@3Rcez5vgkt009QAAAFA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/paasiaas&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Y-X12IU@3Rcez5vgkt009QAAAFA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[13] https://www.theregister.com/2022/11/02/cloud_magic_era_ends/
[14] https://whitepapers.theregister.com/
Re: AWS is looking old
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.
Data Repatriation
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!
Re: Data Repatriation
Just wait until you see the egress bill from your cloud provider. It will surely rain be pouring down. Your data is the hostage in the repatriation.
Re: Data Repatriation
Please don't tell me that it's an actual thing for businesses to not keep local backups of the data (often times their own valuable ip) they fling onto other people's servers?
Re: Data Repatriation
But, but... the cloud offered a complete deal including backups... So we went for it.
/s (I wish, I know many who handed their entire business over to the cloud provider)
Re: Data Repatriation
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.
You mean?
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.
Re: You mean?
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.
Clouds clearing
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.
AWS is looking old
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.