I think these situations have been happening for at least a decade, and no changes are apparent. Google’s calculation seems to be that whatever money they save by not having a support department is more than it costs them in lost business through bad publicity. Who knows, they might be right, but I won’t trust Google to store any critical data for my business, and I’m extremely wary of using any of their other services. Even if I don’t lose access to my account at some point, the likelihood is that the whole service gets shut down anyway!
“Apparently, we should have thought about that before our project was suspended.”
No, what the author should’ve thought about was the common sense realization that totally trusting a 3rd party with your work gives them full control over the security or availability of the work. The standard practice is to have at least two copies of something, one local and one remote. You stay keeping local copies in some simple storage like a filesystem or CD’s. That way, if you loose remote copy, you still have your work to use with another solution. Likewise for keeping remote in sync in case your local installation fails for any reason.
This disaster seems predicted using faith over reason. The practice of faith is that the cloud will be all you need and always take care of you. The practice of reason is not to put all your eggs in one basket as systems and services often fail. Best to always avoid faith in IT matters.
The practice of faith is that the cloud will be all you need and always take care of you. The practice of reason is not to put all your eggs in one basket as systems and services often fail. Best to always avoid faith in IT matters.
The promise of cloud systems isn’t based on faith - it’s outsourcing those practices to someone who can do it better & cheaper than you can. I ship massive amounts of data to S3 because it’d cost me a freaking fortune for redundant storage, run jobs in Lambda because it’s nearly free… and when I push to S3 the data durability is several 9s more than I can promise.
The risk then becomes “does Amazon accidentally nuke my account?” We mitigate this with enterprise support - I have the direct cell phone of a technical manager who answers 24x7. We’ve contacted our TAM before when we’ve had billing issues, security problems detected, etc, and you definitely get what you pay for.
I think it’s disingenuous to point to “faith” when in reality a good cost/benefit analysis can give the same result. (That’s not to say that the CBA in this case was good…)
Ok. Ill back off on the faith part. Too harsh. It’s just lastest article with pattern of: put all resources and trust in cloud service; it fails; what’s wrong with these clouds!? Old school stuff still chugging along during all these failures. Alright, maybe your workload is impossible to do without a cloud and specific, barely-tested service. Let’s briefly assess that.
What kind of HW resources do you need for processing and storage?
What’s your incoming and outgoing monthly traffic?
What’s cheapest servers plus colo that can handle that?
Can you reduce network costs by running same inputs on each node in HA setup? Can you reduce storage or networking by using differentials or just storing summaries?
If none of this applies, then 1-2 cloud providers might have been best choice for you. Amazon and 1 other given tried-and-true principle. So, what’s your situation look like in light of above criteria?
It’s just lastest article with pattern of: put all resources and trust in cloud service; it fails; what’s wrong with these clouds!? Old school stuff still chugging along during all these failures.
In reality, my AWS experience is significantly more reliable than my data center experience! Lots of people have complaints when they push out a half-baked architecture that doesn’t take into account what the cloud means - for example, you have to assume your EC2 instances go away. That’s good, IMO - that means you design state out of your application, which is almost always correct.
Alright, maybe your workload is impossible to do without a cloud and specific, barely-tested service
Cheeky. But hardly true.
My company has gone with AWS because the testing, security, and scalability Amazon gives us is far beyond what we’d be able to put in place for a similar cost. And 95% of our workload is very possible to accomplish with local hardware, don’t get me wrong - but it’d be far more expensive :)
I can assure you of one thing: this decision is under a microscope but almost every time it’s come up it’s been proven to be the right one.
Let’s briefly assess that
I really wish I could share some numbers. I’d be drawn, quartered, and fired unless I somehow got all lobste.rs users to execute an one-way NDA with $company ;)
Just remember - in that comment above I’m talking about paying for AWS enterprise support. That’s a five figure per month minimum
AWS has enabled us to drive a significant reduction in IT costs; quite a bit comes from not over provisioning your data center but significantly savings come from being able to shift huge teams from an operational support workload and retraining them so they have more functional abilities.
The funny thing is that this has been true in my experience from startups to major enterprises - we always go through these exercises and always see significant cost improvements.
It seems my comment from last night never submitted for some reason. Let’s try again.
“In reality, my AWS experience is significantly more reliable than my data center experience! Lots of people have complaints when they push out a half-baked architecture that doesn’t take into account what the cloud means”
I believe both of those claims.
“My company has gone with AWS because”
Whoa! Rewind. The article was about Google cloud that I recall. Unless Google is doing the NIDS for AWS now. That’s the “specific, barely-tested service” vs king-of-clouds, tested-to-death AWS.
“ I’d be drawn, quartered, and fired unless I somehow got all lobste.rs users to execute an one-way NDA with $company ;)”
At least you say no with style.
“And 95% of our workload is very possible to accomplish with local hardware, don’t get me wrong - but it’d be far more expensive :)”
“AWS has enabled us to drive a significant reduction in IT costs”
Ok. So, let’s assume you’re competent and that the above statement is true. I think it’s a good assumption. So, is it still much cheaper if it’s local at two sites syncd vs two clouds (one Amazon) synced at their per CPU & packet rates? Because you’d need at least two cloud services, one well-understood, to avoid the single-dependency and new provider risk.
“I’m talking about paying for AWS enterprise support. That’s a five figure per month minimum”
I imagine that’s going to be one of my future talking points. ;)
“The funny thing is that this has been true in my experience from startups to major enterprises - we always go through these exercises and always see significant cost improvements.”
I might need an update on cost-benefit analysis in these situations. As angersock said, I can imagine specific workloads better suited to clouds. I’ve seen some nice case studies on that. What you describe, though, are the kind of companies that were probably doing it in-house on systems they personally managed and so on. The companies whose IT is probably inefficient as hell. Almost sets up clouds to look good during a transition since they constantly invest in efficiency, automation, and availability. There’s alternatives in terms of IT solutions and non-cloud hosting that similarly invest in efficiency, self-managing, etc. I’d like to see an enumeration of such services that remain today and comparison of cloud options to them.
I’ll give you an example from back when I did this stuff. Long time ago. Smaller businesses commonly needed an email server, a web site/server, file storage, backups, a printer, and a few other things. You’d buy them computers & network the place up. Those other things usually ended up on one or more servers. If it was Linux, you needed an admin doing a bunch of command line magic on a bunch of software not meant to work together. Each setup would be custom, making contracting harder for when problems show up. Very manual, costly, and plenty of boxes. Like the IT shops in the cloud examples. ;)
Then, you had smart companies like Net Integrators' trying to do it better. Their Nitix appliances were like a UNIX/Linux answer to AS/400’s: all-in-one appliances with easy UI for configuring stuff, tested to work together, extra HD with auto-backups, HA setup, and UNIX to avoid Windows' problems of the time. A box cost just under $3,000 where Windows appliances cost around $2,000. Largely fire-and-forget operations whose support could be done by Indian contractors if you wanted since it was just some GUI work on software updates. Physical HA was placing an extra box in a colo somewhere for three digits a month. Solution had its rough edges like others built on then-rough UNIX components by small teams. Worked so well IBM bought it, pulled generic one off the market, integrated it with pricey Lotus BS, and sold it that way.
So, modern solutions like that for web apps, platforms, email, whatever in efficient languages (reduce HW cost) on machines in colos (esp cheap cages). Love to see some surveys of those with comparison to cloud equivalents. That would be real test of cloud’s cost-benefit as it’s more apples to apples.
No, what the author should’ve thought about was the common sense realization that totally trusting a 3rd party with your work gives them full control over the security or availability of the work.
Fred’s down in Houston–believe me, this is something I’m quite sure they considered.
There is unfortunately a category of work that is both really awkward to do remotely and locally, especially with large datasets.
I think these situations have been happening for at least a decade, and no changes are apparent. Google’s calculation seems to be that whatever money they save by not having a support department is more than it costs them in lost business through bad publicity. Who knows, they might be right, but I won’t trust Google to store any critical data for my business, and I’m extremely wary of using any of their other services. Even if I don’t lose access to my account at some point, the likelihood is that the whole service gets shut down anyway!
“Apparently, we should have thought about that before our project was suspended.”
No, what the author should’ve thought about was the common sense realization that totally trusting a 3rd party with your work gives them full control over the security or availability of the work. The standard practice is to have at least two copies of something, one local and one remote. You stay keeping local copies in some simple storage like a filesystem or CD’s. That way, if you loose remote copy, you still have your work to use with another solution. Likewise for keeping remote in sync in case your local installation fails for any reason.
This disaster seems predicted using faith over reason. The practice of faith is that the cloud will be all you need and always take care of you. The practice of reason is not to put all your eggs in one basket as systems and services often fail. Best to always avoid faith in IT matters.
The promise of cloud systems isn’t based on faith - it’s outsourcing those practices to someone who can do it better & cheaper than you can. I ship massive amounts of data to S3 because it’d cost me a freaking fortune for redundant storage, run jobs in Lambda because it’s nearly free… and when I push to S3 the data durability is several 9s more than I can promise.
The risk then becomes “does Amazon accidentally nuke my account?” We mitigate this with enterprise support - I have the direct cell phone of a technical manager who answers 24x7. We’ve contacted our TAM before when we’ve had billing issues, security problems detected, etc, and you definitely get what you pay for.
I think it’s disingenuous to point to “faith” when in reality a good cost/benefit analysis can give the same result. (That’s not to say that the CBA in this case was good…)
Ok. Ill back off on the faith part. Too harsh. It’s just lastest article with pattern of: put all resources and trust in cloud service; it fails; what’s wrong with these clouds!? Old school stuff still chugging along during all these failures. Alright, maybe your workload is impossible to do without a cloud and specific, barely-tested service. Let’s briefly assess that.
What kind of HW resources do you need for processing and storage?
What’s your incoming and outgoing monthly traffic?
What’s cheapest servers plus colo that can handle that?
Can you reduce network costs by running same inputs on each node in HA setup? Can you reduce storage or networking by using differentials or just storing summaries?
If none of this applies, then 1-2 cloud providers might have been best choice for you. Amazon and 1 other given tried-and-true principle. So, what’s your situation look like in light of above criteria?
In reality, my AWS experience is significantly more reliable than my data center experience! Lots of people have complaints when they push out a half-baked architecture that doesn’t take into account what the cloud means - for example, you have to assume your EC2 instances go away. That’s good, IMO - that means you design state out of your application, which is almost always correct.
Cheeky. But hardly true.
My company has gone with AWS because the testing, security, and scalability Amazon gives us is far beyond what we’d be able to put in place for a similar cost. And 95% of our workload is very possible to accomplish with local hardware, don’t get me wrong - but it’d be far more expensive :)
I can assure you of one thing: this decision is under a microscope but almost every time it’s come up it’s been proven to be the right one.
I really wish I could share some numbers. I’d be drawn, quartered, and fired unless I somehow got all lobste.rs users to execute an one-way NDA with $company ;)
Just remember - in that comment above I’m talking about paying for AWS enterprise support. That’s a five figure per month minimum
AWS has enabled us to drive a significant reduction in IT costs; quite a bit comes from not over provisioning your data center but significantly savings come from being able to shift huge teams from an operational support workload and retraining them so they have more functional abilities.
The funny thing is that this has been true in my experience from startups to major enterprises - we always go through these exercises and always see significant cost improvements.
It seems my comment from last night never submitted for some reason. Let’s try again.
“In reality, my AWS experience is significantly more reliable than my data center experience! Lots of people have complaints when they push out a half-baked architecture that doesn’t take into account what the cloud means”
I believe both of those claims.
“My company has gone with AWS because”
Whoa! Rewind. The article was about Google cloud that I recall. Unless Google is doing the NIDS for AWS now. That’s the “specific, barely-tested service” vs king-of-clouds, tested-to-death AWS.
“ I’d be drawn, quartered, and fired unless I somehow got all lobste.rs users to execute an one-way NDA with $company ;)”
At least you say no with style.
“And 95% of our workload is very possible to accomplish with local hardware, don’t get me wrong - but it’d be far more expensive :)”
“AWS has enabled us to drive a significant reduction in IT costs”
Ok. So, let’s assume you’re competent and that the above statement is true. I think it’s a good assumption. So, is it still much cheaper if it’s local at two sites syncd vs two clouds (one Amazon) synced at their per CPU & packet rates? Because you’d need at least two cloud services, one well-understood, to avoid the single-dependency and new provider risk.
“I’m talking about paying for AWS enterprise support. That’s a five figure per month minimum”
I imagine that’s going to be one of my future talking points. ;)
“The funny thing is that this has been true in my experience from startups to major enterprises - we always go through these exercises and always see significant cost improvements.”
I might need an update on cost-benefit analysis in these situations. As angersock said, I can imagine specific workloads better suited to clouds. I’ve seen some nice case studies on that. What you describe, though, are the kind of companies that were probably doing it in-house on systems they personally managed and so on. The companies whose IT is probably inefficient as hell. Almost sets up clouds to look good during a transition since they constantly invest in efficiency, automation, and availability. There’s alternatives in terms of IT solutions and non-cloud hosting that similarly invest in efficiency, self-managing, etc. I’d like to see an enumeration of such services that remain today and comparison of cloud options to them.
I’ll give you an example from back when I did this stuff. Long time ago. Smaller businesses commonly needed an email server, a web site/server, file storage, backups, a printer, and a few other things. You’d buy them computers & network the place up. Those other things usually ended up on one or more servers. If it was Linux, you needed an admin doing a bunch of command line magic on a bunch of software not meant to work together. Each setup would be custom, making contracting harder for when problems show up. Very manual, costly, and plenty of boxes. Like the IT shops in the cloud examples. ;)
Then, you had smart companies like Net Integrators' trying to do it better. Their Nitix appliances were like a UNIX/Linux answer to AS/400’s: all-in-one appliances with easy UI for configuring stuff, tested to work together, extra HD with auto-backups, HA setup, and UNIX to avoid Windows' problems of the time. A box cost just under $3,000 where Windows appliances cost around $2,000. Largely fire-and-forget operations whose support could be done by Indian contractors if you wanted since it was just some GUI work on software updates. Physical HA was placing an extra box in a colo somewhere for three digits a month. Solution had its rough edges like others built on then-rough UNIX components by small teams. Worked so well IBM bought it, pulled generic one off the market, integrated it with pricey Lotus BS, and sold it that way.
So, modern solutions like that for web apps, platforms, email, whatever in efficient languages (reduce HW cost) on machines in colos (esp cheap cages). Love to see some surveys of those with comparison to cloud equivalents. That would be real test of cloud’s cost-benefit as it’s more apples to apples.
Fred’s down in Houston–believe me, this is something I’m quite sure they considered.
There is unfortunately a category of work that is both really awkward to do remotely and locally, especially with large datasets.