Looking to Buy a SAN
-
@scottalanmiller said in Looking to Buy a SAN:
@thecreaitvone91 said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
@scottalanmiller said in Looking to Buy a SAN:
Keep in mind that serverless itself doesn't imply cloud or hosted or third party. You can run serverless on your own server. It requires containers to run in, just like any other workload. If you feel that serverless is a critical part of your design, but don't want a dependency on third parties, you can always run your own.
I presume that this allows you to control latency issues common with serverless by enforcing what stays hot, although I've not tested that theory.
I assume you're talking about OpenFaaS. You can control cold starts but the complexity of setting that up along with maintaining it is light years above deploying to a provider.
As with anything, people not leveraging public cloud offerings (specifically serverless in this case and not just the big 3) is because of FUD. There are very few real cases where it can't be leveraged. As you said in another thread, don't avoid the best because it fails to be perfect.
Doesn't have the same cost benefits running on your own hardware either.
If it doesn't, then it's cloud vs. premises that is being compared, and serverless is a distraction.
What? Serverless isn't billed or tracked at all unless your function is running. It's not a distraction, it's very much dependent on that architecture for his point.
-
@scottalanmiller said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
@thecreaitvone91 said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
@scottalanmiller said in Looking to Buy a SAN:
Keep in mind that serverless itself doesn't imply cloud or hosted or third party. You can run serverless on your own server. It requires containers to run in, just like any other workload. If you feel that serverless is a critical part of your design, but don't want a dependency on third parties, you can always run your own.
I presume that this allows you to control latency issues common with serverless by enforcing what stays hot, although I've not tested that theory.
I assume you're talking about OpenFaaS. You can control cold starts but the complexity of setting that up along with maintaining it is light years above deploying to a provider.
As with anything, people not leveraging public cloud offerings (specifically serverless in this case and not just the big 3) is because of FUD. There are very few real cases where it can't be leveraged. As you said in another thread, don't avoid the best because it fails to be perfect.
Doesn't have the same cost benefits running on your own hardware either.
Def not, especially if you are within the free tier every month. The cost in electricity alone is higher let alone maintenance, patching (server and k8s and OpenFaaS), troubleshooting, etc.
That assumes that you can do everything that you need in serverless. I know literally zero companies that can do that. You have to include the cost of moving to bespoke software built purely around this with a team of devs that understand it. The cost, in the real world, is staggering.
No it doesn't. Maybe for electricity, but again if you are within the free tier per month, the cost is infinitely smaller because there is no cost to run it on a provider. Local hosted will always cost more because you have to maintain it, again if you stay within the free tier.
-
@scottalanmiller said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
I assume you're talking about OpenFaaS. You can control cold starts but the complexity of setting that up along with maintaining it is light years above deploying to a provider.
Except a problem with the providers is cold starts. So if that's needed, it's needed either way.
Depends on the language and who you use. If you use Cloudflare workers, it's an average of under 200 ms. Again, completely depends on language and provider.
-
@scottalanmiller said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
As with anything, people not leveraging public cloud offerings (specifically serverless in this case and not just the big 3) is because of FUD. There are very few real cases where it can't be leveraged. As you said in another thread, don't avoid the best because it fails to be perfect.
That's actually the opposite. I see loads and loads of people going TO cloud because of FUD. Then getting there and finding out that they can't afford it - either because it costs more (generally a LOT more) or because they can't afford what it costs holistically to make that piece of it cost less.
It sounds good, but we're talking real world here. Outside of a few unique enterprises with crazy deep pockets, or isolated workloads, even big companies can't move to 100% bespoke workloads with all of their investment being driven not by total cost, but by a desire to just say that they are 100% in on cloud.
It doesn't even begin to cover issues like companies that can't have that latency or can't be online. There is zero excuse for pushing cloud as the single answer, it isn't. Period. No ifs, ands, or buts. And calling "basic business decision making" around cost "FUD" is not fair, it's anything but FUD. It's cold, hard math.
You keep saying this, but you're referencing companies that just lift and shift. We have said over and over again obviously that's more expensive. Applications need to be redesigned to leverage public cloud efficiently.
It doesn't even begin to cover issues like companies that can't have that latency or can't be online.
What does this mean? Latency from what? services in the same AZ and VPC are averaging the same as a LAN. You can even run HPC workloads with Infiniband speeds.
Outside of a few unique enterprises with crazy deep pockets, or isolated workloads,
That's demonstrably false. I mean it's amazing you can say those words and people on this site will read it and believe it when you can just clearly look anywhere and see tons of companies relying either solely or very heavily on cloud providers.
-
@scottalanmiller said in Looking to Buy a SAN:
There is zero excuse for pushing cloud as the single answer, it isn't.
and no one here has done that. We were talking about serverless and different providers and you morphed it into how companies can't use cloud because they use it the wrong way which makes it expensive.
-
@stacksofplates said in Looking to Buy a SAN:
What does this mean? Latency from what? services in the same AZ and VPC are averaging the same as a LAN. You can even run HPC workloads with Infiniband speeds.
He's referring to SMBs that are running on Dial-up or otherwise shit internet connections that can't operate with anything that isn't on-prem. So because of that, he's saying cloud isn't an option in the real world.
-
Seriously, this would be great to listen to over a scotch and cigar. One of y'all can bring slides. Voices would get raised. Bottles would get broken. In the distance, crows flee.In the meantime, the @op has left the building without really taking anyone's advice.
-
@Grey said in Looking to Buy a SAN:
Seriously, this would be great to listen to over a scotch and cigar. One of y'all can bring slides. Voices would get raised. Bottles would get broken. In the distance, crows flee.In the meantime, the @op has left the building without really taking anyone's advice.
I'm still here. We just decided not to listen to illogical advice.
-
@JaredBusch said in Looking to Buy a SAN:
The reason cloud is so expensive for most is because they simply try to move existing workloads to it without redevelopment of the workloads to a server less design.
This is what it really the simple one sentence answer. People think its so expensive because they aren't taking the time to learn the capabilities of cloud environments. It's also not something you can pick up in a single afternoon of reading. You need to do your own IaC deployments and utilize PaaS and SaaS services in addition to IaaS.
The less IaaS you use the better, but if you do need to leverage IaaS you should design it so its scalable and data isnt stored on instances. Spot instances are way cheaper than your typical on demand instances. Up to 90% in certain cases.
That's the reason cloud centered roles are paying $$$ because the value is in the engineering not the operations on the cloud. Operations is automated and OS administration is very, very limited.
After all the pain of the initial design, you get a much more reliable infrastructure that requires MUCH less maintenance.
-
@IRJ said in Looking to Buy a SAN:
This is what it really the simple one sentence answer. People think its so expensive because they aren't taking the time to learn the capabilities of cloud environments.
Likewise, people think cloud is cheap because they don't consider all of the cost of support, retraining, and rewriting all of their software - which for most shops is not even software that they own and most shops don't even have developers.
So normally you are talking about companies hiring a new development staff and starting from scratch for years of attempting to build software at enormous cost with no benefits to be realized for years - in a field with failure rates (as in being able to complete projects) hits around 85%.
-
@stacksofplates said in Looking to Buy a SAN:
@scottalanmiller said in Looking to Buy a SAN:
@thecreaitvone91 said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
@scottalanmiller said in Looking to Buy a SAN:
Keep in mind that serverless itself doesn't imply cloud or hosted or third party. You can run serverless on your own server. It requires containers to run in, just like any other workload. If you feel that serverless is a critical part of your design, but don't want a dependency on third parties, you can always run your own.
I presume that this allows you to control latency issues common with serverless by enforcing what stays hot, although I've not tested that theory.
I assume you're talking about OpenFaaS. You can control cold starts but the complexity of setting that up along with maintaining it is light years above deploying to a provider.
As with anything, people not leveraging public cloud offerings (specifically serverless in this case and not just the big 3) is because of FUD. There are very few real cases where it can't be leveraged. As you said in another thread, don't avoid the best because it fails to be perfect.
Doesn't have the same cost benefits running on your own hardware either.
If it doesn't, then it's cloud vs. premises that is being compared, and serverless is a distraction.
What? Serverless isn't billed or tracked at all unless your function is running. It's not a distraction, it's very much dependent on that architecture for his point.
It's a distraction from the discussion. People were pointing out that serverless doesn't save you money if you run it yourself, ergo serverless wasn't the thing doing the savings but rather cloud.
It's a red herring, a distraction, from what was proposed as the money saving component of the recommendation.
-
@stacksofplates said in Looking to Buy a SAN:
@scottalanmiller said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
@thecreaitvone91 said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
@scottalanmiller said in Looking to Buy a SAN:
Keep in mind that serverless itself doesn't imply cloud or hosted or third party. You can run serverless on your own server. It requires containers to run in, just like any other workload. If you feel that serverless is a critical part of your design, but don't want a dependency on third parties, you can always run your own.
I presume that this allows you to control latency issues common with serverless by enforcing what stays hot, although I've not tested that theory.
I assume you're talking about OpenFaaS. You can control cold starts but the complexity of setting that up along with maintaining it is light years above deploying to a provider.
As with anything, people not leveraging public cloud offerings (specifically serverless in this case and not just the big 3) is because of FUD. There are very few real cases where it can't be leveraged. As you said in another thread, don't avoid the best because it fails to be perfect.
Doesn't have the same cost benefits running on your own hardware either.
Def not, especially if you are within the free tier every month. The cost in electricity alone is higher let alone maintenance, patching (server and k8s and OpenFaaS), troubleshooting, etc.
That assumes that you can do everything that you need in serverless. I know literally zero companies that can do that. You have to include the cost of moving to bespoke software built purely around this with a team of devs that understand it. The cost, in the real world, is staggering.
No it doesn't. Maybe for electricity, but again if you are within the free tier per month, the cost is infinitely smaller because there is no cost to run it on a provider. Local hosted will always cost more because you have to maintain it, again if you stay within the free tier.
Sure. But saying "if its free" is pretty silly. Who can run their enterprise off of a free tier of anything? No one. That's why they offer a free tier, so that pointlessly small workloads can sample how things work or learn how to do it before paying for the real services that they need.
But even the free tier would require you to spend a fortune.
Take a normal SMB. They need things like QuickBooks, ERP, a SQL Server database, proprietary server software for some hardware that they have to use.... how do these things with all (and it has to be ALL) of their code and storage and backups and monitoring get put into serverless computing at zero cost? or even at low cost? Or even... at all?
-
@stacksofplates said in Looking to Buy a SAN:
@scottalanmiller said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
I assume you're talking about OpenFaaS. You can control cold starts but the complexity of setting that up along with maintaining it is light years above deploying to a provider.
Except a problem with the providers is cold starts. So if that's needed, it's needed either way.
Depends on the language and who you use. If you use Cloudflare workers, it's an average of under 200 ms. Again, completely depends on language and provider.
Sure, using like Go can do a lot to keep it down. But 200ms is high, so high that common workloads like VoIP would be impacted to the point that the couldn't do it. There are definitely workloads, like email, where 200ms would even be noticed. But workloads like web where you'd kinda notice and workloads like VoIP where it would be a huge problem and ones like finance where it's a show stopper.
-
It was at that moment that he learned there are certain parts of the body that you should never Nair.
-
@Obsolesce said in Looking to Buy a SAN:
@scottalanmiller said in Looking to Buy a SAN:
@Dashrender said in Looking to Buy a SAN:
@thecreaitvone91 said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
@bnrstnr said in Looking to Buy a SAN:
@Obsolesce said in Looking to Buy a SAN:
@stacksofplates said in Looking to Buy a SAN:
@scottalanmiller said in Looking to Buy a SAN:
@Dashrender said in Looking to Buy a SAN:
@coliver said in Looking to Buy a SAN:
@flaxking said in Looking to Buy a SAN:
@ScottyBoy said in Looking to Buy a SAN:
@flaxking said in Looking to Buy a SAN:
I've recognized an IPOD and witnessed it play out.
In the end the business decided it made more financial sense to put 200 VMs in Azure.
This is for a TV station cloud simply isn't an option to run this stuff unfortunately.
My point is that putting a bunch of VMs in Azure is a pretty expensive solution, but dealing with an IPOD ends up costing the business enough that the cost is acceptable.
The other solution is to not design an IPOD.
Exactly. Buy a correctly sized Scale box - no IPOD... sure, huge upfront cost, but who knows over the long term compared to Azure. etc etc etc.. We don't have any of the other needed information to know if going to Azure was the right move or not... but it's done, so we move on.
Literally everything is cheap compared to Azure. LOL. Even with all their specialty serverless whatever, never seen it cost close to what running your own would do. The cost is just so absurd per workload.
Their serverless offering is on par with the rest. It's a million requests per month and 400,000 seconds of compute for free. After that it's only $0.20 per million executions and $0.000016 per second. That's not really expensive at all.
Exactly. I'm using in a lot of places in production with ~10k users and twice as many devices that is using the serveless functions in many areas... basically for free. And, that's just the start (one example) of it... Having a VM with enough power to process that as frequently as it's getting done now along with all the other benefits around it, there's truly no comparison. Scaling it down to how a typical SMB would use it, well that's a no-brainer, as it'd be totally free and 100% beneficial. I don't think one's ignorance of a technology justifies it's disqualification of use in the real world.
This should probably be it's own topic, but here we are... I'm totally ignorant to Azure and serverless concepts in general. What types of real world services/processes are SMBs using (or could/should be using) serverless Azure for?
There's a few different scenarios. Anything reactionary essentially. Send a message/email based on an event, do some kind of work based on messages in a message queue, transform or modify data, etc. You can even use it to build and define APIs. I have an API running in Vercel (not Azure but another serverless offering) and I don't have to run the service in a VM full time.
Invoicing and Accounts Payable is a big use of it
I don't understand how those are serverless? There is software running - right? where is that software running? This is something I completely don't understand - and I'm guessing @bnrstnr likely doesn't either - but he'll correct me if I'm wrong and he does.
It's not the best terminology. But it's the standard now. It's like "API only" processing, getting as light as you reasonably can.
API processing is just one of many.
If you want to bother to learn more about serverless tech and use cases, give this PDF a read:
https://docs.microsoft.com/en-us/dotnet/architecture/serverlessI think if you read it without bias, you'll better understand and grasp it.
I think you are missing the whole point. When hosted/cloud isn't an option because it can't be done, going to serverless doesn't fix anything.
You are acting like a niche technology, a great one, but niche, is a panacea for everything. Name me one company, any company, anywhere in the world, that has gone completely serverless and no longer needs anything but end user workpoints with zero workload on them? Any company, anywhere.
-
@stacksofplates said in Looking to Buy a SAN:
I assume you're talking about OpenFaaS. You can control cold starts but the complexity of setting that up along with maintaining it is light years above deploying to a provider.
OpenWhisk is one, too. That's one I had been looking at.
-
@stacksofplates said in Looking to Buy a SAN:
You keep saying this, but you're referencing companies that just lift and shift. We have said over and over again obviously that's more expensive. Applications need to be redesigned to leverage public cloud efficiently.
Right, but you never talk about the cost of doing that. This is generally a cost that is massively larger (orders of magnitude larger) than the potential cost savings of any infrastructure.
Take QuickBooks as a really, really common example. It costs a few hundred dollars per year to license. It requires extremely little infrastructure... Windows Server, RDS... a few things. But nowhere near $5K a year.
Now tell me what it will cost to recreate QuickBooks (on any platform.) There is a reason that when they went to cloud options that even Intuit couldn't recreate it, they had to drop loads of functionality. People who need QB, which is a lot of people because of certain tax things that it is relatively unique in, have very few options out there and essentially all are very antiquated. Remaking it is absolutely possible, but there is a reason why no software vendor out there is considering that that will be of significant cost savings enough to be willing to invest in it.
We are talking tens of millions of dollars. Even if we are ridiculous and claim we can do it for one million dollars... that's 200 years of running your own server for this, easily. And you only save that much if the resulting product never needs code updates and the serverless system is 100% free.
If you are an Intuit kind of company and can make this investment and spread the benefits around thousands or tens of thousands of customers, sure, IF it has real benefit, you can consider doing it.
But then you have to address all of the places that either for legal reasons or technology ones cannot use hosted services. No matter how obnoxious people are, in the real world, businesses commonly struggle to get fast and reliable Internet services. They do right here in the Dallas Metro, they are really screwed in the Houston Metro, we have customers all over the country that can't rely on their Internet and no providers short of "build out your own fiber" exist.
Check Tyler, TX, that's been one of our hardest. We cannot find any option, short of laying our own fiber, to beat 18/1 AT&T DSL. We've talked to everyone, we've brought in consultants.
-
@stacksofplates said in Looking to Buy a SAN:
@scottalanmiller said in Looking to Buy a SAN:
There is zero excuse for pushing cloud as the single answer, it isn't.
and no one here has done that. We were talking about serverless and different providers and you morphed it into how companies can't use cloud because they use it the wrong way which makes it expensive.
That's exactly what's being done here. Everyone keeps acting like all IT can stop managing servers, stop considering options, because (in multiple threads) the answer is just "go serverless" because it "saves money".
Serverless is well worth considering, as a component of an infrastructure, when it makes sense. Like all things in IT. Anytime it is promoted as "always cheaper" and the only way that people should do things, it's wrong.
-
We recently looked at moving our workload to Azure. Lead by the Dev team. They looked at ways they could leverage various Azure services and technology (no idea which, it's their job) and the most economy option they came back with was $23,500 CAD per month. I presume that's based on them leveraging the most economical services in the best way. No idea though as I'm not involved on that side.
Our new server infrastructure which Dev use comes to just below $350k CAD which is set to last 5 years until the next replacement.
That's just below $6,000 CAD per month. Far less than the $23,500 CAD that Azure would cost. Even if we take IT staff wages in to this were still at a much lower number for dedicated systems with no contention.
For email, CMS etc, sure. For what our Dev team use, nope. And that's them leading that costing, not IT.
-
@JaredBusch said in Looking to Buy a SAN:
The reason cloud is so expensive for most is because they simply try to move existing workloads to it without redevelopment of the workloads to a server less design.
But how do people cost effectively build their own software to replace what they already have to leverage the difference?
What you've stated is just a long winded way of stating what was already said: "cloud is too expensive for most workloads". You are explaining why that is true, and that's great, but holistically the same answer normally applies.
Serverless is only broadly an option when you have situations where bespoke software is already being created anyway, and you can redirect that bespoke process. That applies to a small number of workloads, at a small number of companies.
