If you don't need all the cloud stuff - Hetzner has also storage server i.e. SX134 with 10x16TB for 250€/month. For 50€/month you can order a 10Gbit upgrade and after 20TB free traffic it's 1€/TB egress. It's an ordner of magnitude cheaper. Of course that's because you are responsible for checking the disks, setting everything up and keeping it up to date - but at least for the problems we have where I work there is absolutely no need to use Google or any other cloud and Hetzner is more than good enough.
You need to multiply those numbers by 3 to reflect that the cloud storage service has geographic redundancy, and you need a software stack which provides the same replication and bitrot protection, etc.
Most use cases do not need triple redundancy with geographic redundancy. If you really do need this, just order this and use Wasabi as well ($6/TB, pricing below) and you will get all that along with having cross-provider redundancy.
I’m not saying that this is the only factor but most places do need redundancy and the cost of providing it is non-trivial. There’s still plenty of margin but you need a fair amount of storage to balance out the staff time you’d need to run that service on bare servers.
> you need a fair amount of storage to balance out the staff time you’d need to run that service on bare servers.
Do you have any proof of your statement? Eg. same type of org, one using Hetzner the other using one of the big three? Are the sizes of the DevOps team that different? And don’t forget to calculate the overhead on the big three from having to architect for the cost of egress.
I know there is the sales fairy of being able to scale out quickly, does that really matter in both cases? And if really required, can’t it be done for the couple of hours/days on one of the big three? Not to forget that scale-out isn’t available in some regions, as we’ve seen discussed here recently.
I'm not sure what counts as proof, but I think it's fair to point out that comparing a bare metal server to managed cloud storage is not like for like. For example, Google cloud storage offers automatic migration of cold data to tape, replication across regions allowing fast recovery in case of a failure, durability guarantees etc.
If that's not something you need then you could definitely save yourself a lot of money, but if you need to build some of that stuff out (and then monitor it, improve perf, upgrade when new drives come out, be on call for down time etc.) Then the cost comparison becomes less clear.
People say you should manage your own data. That's great until you have to demonstrate compliance with security regulations. If I'm a big cloud provider, the more regulations and requirements there are the better it is for my business.
If you run petabyte size storage clusters, engineering for compliance should still be small fraction of dollars saved compared to what you’d be paying in cloud (even with old pricing).
My guess is that it's more along the lines of GCP being expected to be profitable. They're a distant #3, possibly #4 in the market so it seems unlikely that they're trying to exploit market position when that'll just give customers another reason to switch to AWS.
https://canalys.com/newsroom/global-cloud-services-Q1-2022 has it has AWS at 33%, Azure at 21%, and GCP at 8%. Oracle and AliCloud are the two I'd expect to try for that #3 spot, especially since that could be a single acquisition if e.g. IBM is interested in selling.
The potential confound here is how you measure it, of course: Microsoft gets a big jump from all of the people using Office 365, Google has G-Suite, etc. but I don't think there's any world in which Google has enough leverage to increase pricing if they don't have to.
G-Suite is not a Microsoft 365 competitor. Only businesses with limited needs seem to use G-Suite, and theu supplement it with Microsoft software a fair bit to make using G-Suite palatable.
I say this as a wholly anti-Microsoft person. Even Nextcloud is doing better at delivering features to small and medium businesses than G-Suite.
You might not consider it a competitor but plenty of places do. My point was simply that it’s a separate class of product which is often bundled which makes market share numbers harder to compare.
I have to keep saying this because somehow this myth just will not die: Microsoft does not report O365 revenue as Azure revenue. They broke those out separately many, many years ago.
They appear to define “Intelligent Cloud” like this:
> • Server products and cloud services, including Azure and other cloud services; SQL Server, Windows Server, Visual Studio, System Center, and related Client Access Licenses (“CALs”); and Nuance and GitHub.
> • Enterprise Services, including Enterprise Support Services, Microsoft Consulting Services, and Nuance professional services
Doesn’t like Xbox subscriptions and other unrelated malarkey factor into azure too? If you took out all the income statement stuffing I’ll bet you aws is in the 80% range.
GitHub, Teams, Office 365. Compare that to Google Chat, Google Docs and nothing for code sharing, plus Microsoft has every darn Managed Service Provider and IT guy shoving their product down businesses throats.
Google has an inferior product with no sales infrastructure, its impressive they have gotten as many clients as they currently have.
in the last round of storage price increases, the biggest change for us was reading multi-region storage, like five or six figures a month increase. 'just rearchitect your application.' it seemed to really be a punishment(sin tax if you will) on a behavior they want people not to do. or align revenue with costs i guess. it really sucks working with a partner that thinks punitive actions are ok.
the whole but youre on a contract so it doesnt apply yet is such bs, it just delays the pain, and we lost track of the many places we saw price increases.
Yes, yes, and storage in cloud was always much more expensive than just "amount of hardware for that amount of redundancy".
Bandwidth price to datacenter is below $1/Mbit at bulk rate. That is either sold in form of whole link (say 10Gbit) or per 95th percentile (essentially "peak transfer rate") not "per MB sent"
Cloud pricing is insane, we are handling around 3 billion dynamic requests a month and serving 100+ TBs of static content a month for a few hundred euros, all rented/dedicated servers, with zero problems. So we would pay like 10-100x times more in the cloud for the same hardware and traffic.
We had bosses come few times to price the cloud migration and every single time it ends up being few times as expensive.
"What about extra work to keep the physical hardware running" - for 7 racks we have maybe a dying drive every month or two. Installing a rack of servers takes about a day with all preparations.
"What about setting up software" - we have automation and paid it off long time ago. Genuine concern for company that starts up but not really for one that needs at least few racks worth of servers.
Cloud pricing is often not competitive with an optimized on-prem platform that has noteworthy resource usage. It is wild to think some people see it as a cure all, when it has real risks of destroying a budget.
The resource usage is not too bad, but the fees all the cloud charges for bandwidth are insane. Charging bandwidth in gigabytes should get you laughed out of the room. I have an unmetered connection at home; if you cannot provide one in your own data center, why are you even in business?
That said, the cloud is great if you need to train AI models, convert old data, do end-of-year processing, or do something else that requires a massive amount of resources for a short time.
Unless you are Howard Huges, why would you live in a hotel? And if you wouldn't live in a hotel yourself, why would you pay to have your data live there?
The price per GB aside from obvious (earns more money) have other reason; it's just so much easier to meter when you have hundreds different services.
You just increase the counter. You don't have to worry when service sent data.
How much BW the service actually costs is hard to calculate because you are essentially buying peak capacity, so only thing that matters is that hour in a day where there is peak traffic and which services use that up. Sure there is a bit on power used flipping actual bits in the network devices but that's tiny fraction of that.
More likely the cable owners or the reseller Google is going through for most of their LATAM traffic boosted their rates up, resulting in crazy transit costs.
Transit costs should really be per ASN (eg: Sending 1TB to an ISP we peer with is free, sending that to an ISP we pay is $5), not these extremely broad abstractions of per continent pricing.
Egress is why I own a back end server for my projects rather than rent one from AWS/GCP. Unfortunately you still need at least a small instance to host a website.
Colo and even office/home hosting can be surprisingly affordable for non-critical workloads.
$0.10/kWh power, Gigabit Fiber for $65 to $90 a month and free heating (computers are just as efficient as electric space heaters) can make for very low operating costs.
Note that Cloudflare is cracking down on heavy caching, its an Article 10 violation. If they see you using terabytes of bandwidth a week (especially if your end users are in expensive regions like New Zealand or Australia) they will suspend your site: https://community.cloudflare.com/t/the-way-you-handle-bandwi...
Preferably far far away from the cloud servers that perform workloads on that storage, so you can enjoy high latency plus ingress and egress costs to access it?
None of these costs are valid in isolation. The DIY savings is probably not a savings for most use cases.
// Disclosure: 10+ years back, our company undersold S3 at 1/10th the cost and much higher performance, but only for multi hundred megabyte up to multi gigabyte files. S3 does the generalized object store shape of work extremely well, likely better than anyone, and is worth it -- even before adding IAM, local querying, and all the other S3 goodies when native AWS. While underselling them for storage we operated, we also used them for our own storage needs.
Well there you have it - don’t use cloud severs either ;) Also ingress is free and if that’s your requirement a lot of time you can colo in the same locale/building as cloud region and plug directly into their network
Latency to in-region storage is not minimal like using SSDs on the same server. Often data in the same region can be at different physical data centers.
True, but SSDs don't provide instant access to petabytes.
Also, it is not well known that for a long time you could colocate in the same facility or in the same biz park and enjoy LAN latencies to that campus' AZ of your CSP.
Curious if issues of deeply centralized (think: monopoly) storage services give airt time for the several decentralized storage options such as Sia [0], File [1], and tried and true torrenting.
Best to Checkout alternatives e.g. https://text-generator.io does text to speech over 8x cheaper which is crazy, cloud storage costs going up a lot but compute costs have always been high esp for managed services/ml so shop around and on a code level, ideally use a wrapper around your Cloud usage to make switching easier
I actually think this is the sign of GCP growing up, and trying to become a profitable business. Long gone are the days where pricing was set by a PM throwing shit against the wall, with little attention paid to cost structure or margins (who remembers free GKE clusters?)
I know people hate oracle, and its features and polish is much lower than AWS. But you should check out oracle Cloud pricing, it's really good. And its free tier is absurdly good.
When Oracle decides it's time to start making money instead of gathering market share with a loss leader, it's gonna hurt a lot more than the adjustment from Google here.
