AWS Catastrophic Data Loss
-
@Pete-S said in AWS Catastrophic Data Loss:
@IRJ said in AWS Catastrophic Data Loss:
@dbeato said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
@dbeato said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
@Dashrender said in AWS Catastrophic Data Loss:
@BRRABill said in AWS Catastrophic Data Loss:
because the chances that MS's DC is going to blow up is extremely small
And yet, it is what this thread is about ... exactly that happening.
Except that it's Amazon, not MS.
MS was US Central this year or late last.
MS was the world when their authentication mechanism went down I think it was a year or so ago.
MS was Europe offline with VMs hosed and a recovery needed. Weeks.
MS has had plenty of trials by fire.
Not one of the hyper-scale folks are trouble free.
Most of our clients have had 100% up-time across solution sets for years and in some cases we're coming up on decades. Cloud can't touch that. Period.
And no updates correct right? to have 100 % Up-time you must never do updates.
In a cluster setting, not too difficult. In this case, 100% up-time is defined as nary a user impacted by any service or app being offline when needed.
So, point of clarification conceded.
Yes, I know you could do a cluster and that's how Cloud Providers give you that 99.9% up-time or SLA. Right now it is hard to believe no one has any issues, if cloud providers in a large scale have issues then smaller companies do have them as well. That said, no cloud provider provides any backups for anyone unless you set them up either through their offering or your company.
Yeah and you can only fault yourself, if you are one AZ that fails. Most serious deployments are in different regions as well.
Well, except that:
@Pete-S said in AWS Catastrophic Data Loss:
As we have further investigated this event with our customers, we have discovered a few isolated cases where customers' applications running across multiple Availability Zones saw unexpected impact
At some point, you have to be willing to accept some risks by by not using a different region, generally the risk is VERY, VERY low which is why many customers use AZs.
You have to do risk anaylsis, and see how often these events occur and how likely you would be to be one of the "few" that were impacted.
You can dig in the weeds all you want, but across multiple regions this wouldnt have happened. Which is true HA
-
@IRJ said in AWS Catastrophic Data Loss:
@Pete-S said in AWS Catastrophic Data Loss:
@IRJ said in AWS Catastrophic Data Loss:
@dbeato said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
@dbeato said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
@Dashrender said in AWS Catastrophic Data Loss:
@BRRABill said in AWS Catastrophic Data Loss:
because the chances that MS's DC is going to blow up is extremely small
And yet, it is what this thread is about ... exactly that happening.
Except that it's Amazon, not MS.
MS was US Central this year or late last.
MS was the world when their authentication mechanism went down I think it was a year or so ago.
MS was Europe offline with VMs hosed and a recovery needed. Weeks.
MS has had plenty of trials by fire.
Not one of the hyper-scale folks are trouble free.
Most of our clients have had 100% up-time across solution sets for years and in some cases we're coming up on decades. Cloud can't touch that. Period.
And no updates correct right? to have 100 % Up-time you must never do updates.
In a cluster setting, not too difficult. In this case, 100% up-time is defined as nary a user impacted by any service or app being offline when needed.
So, point of clarification conceded.
Yes, I know you could do a cluster and that's how Cloud Providers give you that 99.9% up-time or SLA. Right now it is hard to believe no one has any issues, if cloud providers in a large scale have issues then smaller companies do have them as well. That said, no cloud provider provides any backups for anyone unless you set them up either through their offering or your company.
Yeah and you can only fault yourself, if you are one AZ that fails. Most serious deployments are in different regions as well.
Well, except that:
@Pete-S said in AWS Catastrophic Data Loss:
As we have further investigated this event with our customers, we have discovered a few isolated cases where customers' applications running across multiple Availability Zones saw unexpected impact
At some point, you have to be willing to accept some risks by by not using a different region, generally the risk is VERY, VERY low which is why many customers use AZs.
You have to do risk anaylsis, and see how often these events occur and how likely you would be to be one of the "few" that were impacted.
You can dig in the weeds all you want, but across multiple regions this wouldnt have happened. Which is true HA
Well, different regions wouldn't be enough for true HA. You'd need different cloud providers as well.
Otherwise you have something called common mode failure. Which is for instance that they are running on the same architecture, maybe even the same hardware and as such could be susceptible to a single problem that will affect the entire cloud.
-
@IRJ said in AWS Catastrophic Data Loss:
@Pete-S said in AWS Catastrophic Data Loss:
@IRJ said in AWS Catastrophic Data Loss:
Adding this graphic again...
The data is on the customer!
Where is that table from?
I'm just wondering, not disputing it
I learned this while doing my Cloud Security Cert. As you can see all major cloud providers follow this model as this was set by CSA (Cloud Security Alliance) as proper customer responsibility.
https://pen-testing.sans.org/blog/2012/07/05/pen-testing-in-the-cloud
https://aws.amazon.com/compliance/shared-responsibility-model/
https://gallery.technet.microsoft.com/Shared-Responsibilities-81d0ff91
Awesome, thanks!
-
@Dashrender said in AWS Catastrophic Data Loss:
@DustinB3403 said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
@DustinB3403 said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
@DustinB3403 said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
Most of our clients have had 100% up-time across solution sets for years and in some cases we're coming up on decades.
Really, decades of uptime. Not a single bad ram module, raid failure, CPU, PSU or MB issue. No site issues (fire, earthquake, tornado etc) in all that time.
@PhlipElder said in AWS Catastrophic Data Loss:
Cloud can't touch that. Period.
You're full of it.
I'm quite proud of our record. It's a testament to the amount of time and money put in to research, proof, and thrash the solution sets we've sold over the years. We don't sell anything we first don't proof.
So you're using technology that is at least a decade old for every one of your customers, because by your own word you can't possibly have had the time to test anything from this year and sold it to a customer!
Not sure how that conclusion came about but far from it.
We've had plenty of NDAs over the years to proof with upcoming tech so that we're on the right page and current.
You've said you've tested everything that you sell. How could this possibly be true to make claims of decades worth of up-time. Power supplies fail, switches die, disks die, MB's die, sites lose power (which people still have jobs to do - just because the lights are out. . .)
So you're still full of it. Not to mention performing any update will eventually require a restart. Windows updates, file server migrations etc. All require some downtime.
all of those things can fail - as long as they have an HA solution that accounts for those failures.
As he said earlier - the customer has NEVER been impacted - that's the point of measurement.
Thank you sir.
This one is relatively recent:
http://blog.mpecsinc.ca/2018/06/our-calgary-oil-gas-show-booth-slide.htmlThis is one of our PoC sets: http://blog.mpecsinc.ca/2018/01/storage-spaces-direct-s2d-sizing-east.html
Systems we built on the current generation
before now(had me wires crossed):
http://blog.mpecsinc.ca/2017/11/intel-server-system-r2224wftzs.htmlA half Petabyte setup: https://www.youtube.com/watch?v=OKnRzEgHHKA
At our peak working with these we had three of them here in the shop: https://www.youtube.com/watch?v=26U6pDsdz5M&t=321s
Drove the neighbours crazy with the jet engine sounds coming out of here. Plenty of Ear Defenders to be had.
That help?
EDIT: Any guesses on the cost for the four node S2D setup with Mellanox 40GbE RDMA dual switch fabric?
-
@PhlipElder said in AWS Catastrophic Data Loss:
@Dashrender said in AWS Catastrophic Data Loss:
@DustinB3403 said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
@DustinB3403 said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
@DustinB3403 said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
Most of our clients have had 100% up-time across solution sets for years and in some cases we're coming up on decades.
Really, decades of uptime. Not a single bad ram module, raid failure, CPU, PSU or MB issue. No site issues (fire, earthquake, tornado etc) in all that time.
@PhlipElder said in AWS Catastrophic Data Loss:
Cloud can't touch that. Period.
You're full of it.
I'm quite proud of our record. It's a testament to the amount of time and money put in to research, proof, and thrash the solution sets we've sold over the years. We don't sell anything we first don't proof.
So you're using technology that is at least a decade old for every one of your customers, because by your own word you can't possibly have had the time to test anything from this year and sold it to a customer!
Not sure how that conclusion came about but far from it.
We've had plenty of NDAs over the years to proof with upcoming tech so that we're on the right page and current.
You've said you've tested everything that you sell. How could this possibly be true to make claims of decades worth of up-time. Power supplies fail, switches die, disks die, MB's die, sites lose power (which people still have jobs to do - just because the lights are out. . .)
So you're still full of it. Not to mention performing any update will eventually require a restart. Windows updates, file server migrations etc. All require some downtime.
all of those things can fail - as long as they have an HA solution that accounts for those failures.
As he said earlier - the customer has NEVER been impacted - that's the point of measurement.
Thank you sir.
This one is relatively recent:
http://blog.mpecsinc.ca/2018/06/our-calgary-oil-gas-show-booth-slide.htmlThis is one of our PoC sets: http://blog.mpecsinc.ca/2018/01/storage-spaces-direct-s2d-sizing-east.html
Systems we built on the current generation
before now(had me wires crossed):
http://blog.mpecsinc.ca/2017/11/intel-server-system-r2224wftzs.htmlA half Petabyte setup: https://www.youtube.com/watch?v=OKnRzEgHHKA
At our peak working with these we had three of them here in the shop: https://www.youtube.com/watch?v=26U6pDsdz5M&t=321s
Drove the neighbours crazy with the jet engine sounds coming out of here. Plenty of Ear Defenders to be had.
That help?
EDIT: Any guesses on the cost for the four node S2D setup with Mellanox 40GbE RDMA dual switch fabric?
This post is a bit dated. But it states clearly, and concisely, exactly where we're at as far as investing in our folks here:
http://blog.mpecsinc.ca/2016/11/whats-in-lab-profit.html
My attitude is simple: If we ain't learning we're effing sh#t up.
-
@Pete-S said in AWS Catastrophic Data Loss:
Well, different regions wouldn't be enough for true HA. You'd need different cloud providers as well.
HA has to do with the uptime, not the amount of redundancy. Different regions from AWS is definitely way more than enough for HA by any standard. You can do HA with a single datacenter, just not an AWS datacenter. But lots of single datacenters provide HA at a facility level.
But your app has to support the multiple datacenter model. That's what is really hard for most people.
-
@scottalanmiller are you purposely avoiding the "major cloud provider didn't have its own backups" discussion?
-
YES YES YES SCREW AWS, they have this big marketing scheme for CEOs that force us to work for those CEOs that believe everything is better in AWS, and the server wont work properly unless its AWS, then when the bill comes we have to explain to them that we can never calculate the cost accurately cause it is Amazon AWS, and they charge for IOPS, and there is no way I can calculate that shit, its meant to be bill sinkhole for to pay bezos divorce settlement .
-
@BRRABill said in AWS Catastrophic Data Loss:
@scottalanmiller are you purposely avoiding the "major cloud provider didn't have its own backups" discussion?
They aren't supposed to have backups. That's an additional feature that people opt out of.
-
@Pete-S said in AWS Catastrophic Data Loss:
@IRJ said in AWS Catastrophic Data Loss:
@Pete-S said in AWS Catastrophic Data Loss:
@IRJ said in AWS Catastrophic Data Loss:
@dbeato said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
@dbeato said in AWS Catastrophic Data Loss:
@PhlipElder said in AWS Catastrophic Data Loss:
@Dashrender said in AWS Catastrophic Data Loss:
@BRRABill said in AWS Catastrophic Data Loss:
because the chances that MS's DC is going to blow up is extremely small
And yet, it is what this thread is about ... exactly that happening.
Except that it's Amazon, not MS.
MS was US Central this year or late last.
MS was the world when their authentication mechanism went down I think it was a year or so ago.
MS was Europe offline with VMs hosed and a recovery needed. Weeks.
MS has had plenty of trials by fire.
Not one of the hyper-scale folks are trouble free.
Most of our clients have had 100% up-time across solution sets for years and in some cases we're coming up on decades. Cloud can't touch that. Period.
And no updates correct right? to have 100 % Up-time you must never do updates.
In a cluster setting, not too difficult. In this case, 100% up-time is defined as nary a user impacted by any service or app being offline when needed.
So, point of clarification conceded.
Yes, I know you could do a cluster and that's how Cloud Providers give you that 99.9% up-time or SLA. Right now it is hard to believe no one has any issues, if cloud providers in a large scale have issues then smaller companies do have them as well. That said, no cloud provider provides any backups for anyone unless you set them up either through their offering or your company.
Yeah and you can only fault yourself, if you are one AZ that fails. Most serious deployments are in different regions as well.
Well, except that:
@Pete-S said in AWS Catastrophic Data Loss:
As we have further investigated this event with our customers, we have discovered a few isolated cases where customers' applications running across multiple Availability Zones saw unexpected impact
At some point, you have to be willing to accept some risks by by not using a different region, generally the risk is VERY, VERY low which is why many customers use AZs.
You have to do risk anaylsis, and see how often these events occur and how likely you would be to be one of the "few" that were impacted.
You can dig in the weeds all you want, but across multiple regions this wouldnt have happened. Which is true HA
Well, different regions wouldn't be enough for true HA. You'd need different cloud providers as well.
Otherwise you have something called common mode failure. Which is for instance that they are running on the same architecture, maybe even the same hardware and as such could be susceptible to a single problem that will affect the entire cloud.
I toyed with this idea, but it is a bit unlieky, however that said when you are vendor agnostic and you have Centos box in DigitalOcean and another one using Vultr.
-
@Emad-R said in AWS Catastrophic Data Loss:
YES YES YES SCREW AWS, they have this big marketing scheme for CEOs that force us to work for those CEOs that believe everything is better in AWS, and the server wont work properly unless its AWS, then when the bill comes we have to explain to them that we can never calculate the cost accurately cause it is Amazon AWS, and they charge for IOPS, and there is no way I can calculate that shit, its meant to be bill sinkhole for to pay bezos divorce settlement .
The Great Firewall of Cloud Marketing has done a great job of suppressing the billing shock that cloud brings with it. It's also been great at suppressing the movement back on-premises where costs are fairly well established.
We have a client we work with that has a handsome cloud credit every month well into five figures. They did some testing for their application work in-cloud to see how it would work. They burned through that five figure credit in a matter of a few days much to their surprise. They put their workload into that cloud, get it up and running, and then the following year that credit disappears. So, they get a billing spike on top of the six figure count it would cost them to run entirely all-in. We have a high performance all-flash hyper-converged solution set just for them.
