Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020
-
@IRJ said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Cloud services have been proven cheaper for companies for a long time.
This isn't true. In fact, many of the companies that have claimed this in the past have turned around and said that now they realize that this wasn't true.
Run the math, when you are ittsy bittsy, yeah, cloud is the only option. But even at NTG's size, it stops being. The cost of cloud is astronomic for typical real world workloads.
The cloud was never meant to be cheaper for running normal workloads, that was never how it was promoted and its not how it is designed. It's cheaper for very, very specific workloads that are extremely niche. Ninety percent of companies can't leverage cloud age all and most of those that can, don't benefit from it except possibly on speciality, isolated workloads.
-
@Jimmy9008 said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Whenever I have looked at this for our environment it just does not work out less.
We have hundreds of customers and consult for everyone from operations with just a couple people to the Fortune 10 and the number of customers that can go to cloud is literally... zero. Literally.
Either because it is insanely costly or, over half the time, simply is impossible and don't work because they can't get reliable connections to it (without building their own ISPs, of course) it just doesn't come up, at all. We have a handful of isolated workloads where VPS services are used to get certain locations or redundancy or whatnot, but zero mainline workloads can be put there.
-
@IRJ said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
The more skilled the IT professional is the more efficient they can design a cloud environment.
That goes both ways. The more skilled an IT pro is, the less "getting the skills from someone else" is valuable. Certainly the more you bring to the table overall, the more value you have. And flexibility is important. But many "cloud skills" are also ways to word "lacking platform skills" and having the ability to efficiently run traditional workloads remains valuable and the more you can do that, the more you are worth to your company, too.
So yes, the more skilled you are... the better you can make either kind of environment.
-
@Jimmy9008 said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Owning our own server over the several years we have them costs drastically less. I have to hire IT services to manage on site, or cloud, so those costs are negligible. But, a server I own compared to one I hire for 6 years in Azure, is a lot less.
Exactly. Assuming you have the relatively rare circumstance where you have a company that has the option of going completely cloud (I'd guess this is about 15% of companies that get this option in the US and around 1% globally)... it takes such a tiny amount of workloads to make cloud really expensive. Very, very few companies run totally bespoke software, or have development departments that can rewrite everything for cost effective cloud computing. Like .2% maybe. For all normal companies, you have to run software.
It's weird when you talk to software firms, they will often act surprised that companies can't just run everything on cloud, or even on extreme architectures like serverless computing.
Then you talk to real world IT and the actual struggles are getting software with a web interface or that can run over a WAN or that don't need a fat client. The disconnect between how people think cloud will work and what real companies deploy is huge. The idea of cloud is great, but the reality is so different.
I work with several industries that have so many barriers to cloud it's laughable. From their needed physical locations to their uptime requirements to their ISP options to the available software available in the industry - any of those pieces would stop them from even considering cloud, and for almost every company, it's all of them.
-
@scottalanmiller said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@Jimmy9008 said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Whenever I have looked at this for our environment it just does not work out less.
We have hundreds of customers and consult for everyone from operations with just a couple people to the Fortune 10 and the number of customers that can go to cloud is literally... zero. Literally.
Either because it is insanely costly or, over half the time, simply is impossible and don't work because they can't get reliable connections to it (without building their own ISPs, of course) it just doesn't come up, at all. We have a handful of isolated workloads where VPS services are used to get certain locations or redundancy or whatnot, but zero mainline workloads can be put there.
What F10 are you consulting?
Containerization and FaaS have made development and deployments insanely easier and less costly.
Statements like zero of your customers can "go to cloud" whatever that means are ridiculous.
-
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Statements like zero of your customers can "go to cloud" whatever that means are ridiculous.
Not in the least. Thinking that every customer can afford to go cloud just because it's the cool new term is what is crazy. Why would they pay all that money and gain nothing? It's just not valuable for real world businesses. Some, of course, but not the majority. And of the ones it makes sense for, almost none can use it for everything.
Remember, I led the move to cloud on Wall St. Built the first cloud there. It was great, but only for certain workloads. Big workloads, important workloads, and it was massively beneficial. But the majority of the workloads couldn't use it. Just didn't make sense. Whether they needed more predictable performance, faster performance, or just didn't scale in that way... even in that environment wasn't practical to pay the overhead of cloud or it couldn't meet the needs.
-
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Containerization and FaaS have made development and deployments insanely easier and less costly.
True for non-cloud installs as well, though. And only true for cloud when you have completely reliable networking. Hosted FaaS is a real challenge if your ISP drops.
-
@scottalanmiller said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Containerization and FaaS have made development and deployments insanely easier and less costly.
True for non-cloud installs as well, though. And only true for cloud when you have completely reliable networking. Hosted FaaS is a real challenge if your ISP drops.
No it's not. Because you can just go somewhere else. It's a real problem when it's locally hosted and your ISP drops and no customers can access it.
-
@scottalanmiller said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Remember, I led the move to cloud on Wall St. Built the first cloud there.
To be transparent, I 100% do not believe this is true.
-
What did you use to provide the infrastructure?
-
How did you manage scaling without the needed APIs?
-
What year was this?
-
-
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@scottalanmiller said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Containerization and FaaS have made development and deployments insanely easier and less costly.
True for non-cloud installs as well, though. And only true for cloud when you have completely reliable networking. Hosted FaaS is a real challenge if your ISP drops.
No it's not. Because you can just go somewhere else. It's a real problem when it's locally hosted and your ISP drops and no customers can access it.
That 'somewhere else' is also more expensive though... Doesnt mattwr if its AWS, Azure, Google Compute, or another... It costs a more. Local hosted doesn't mean one bad ISP will destroy service for customers. We have out infrastructure in our facilities in London, Aberdeen, Barcelona, Middle East, Singapore, Houston, Calgary, Toronto and Singapore. I can lose any site and that will have no impact to customers. Still a lot cheaper than cloud.
Last time I looked, a single one of my calculation servers came to around 80k GBP. High core count, multiple TB of RAM. In Azure, over 6 years, MS were coming back with something in the range of £450k ex VAT. Insane difference.
-
@Jimmy9008 said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Last time I looked, a single one of my calculation servers came to around 80k GBP. High core count, multiple TB of RAM. In Azure, over 6 years, MS were coming back with something in the range of £450k ex VAT. Insane difference.
...again, legacy thinking will almost always lead to increased costs.
-
@Jimmy9008 said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@scottalanmiller said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Containerization and FaaS have made development and deployments insanely easier and less costly.
True for non-cloud installs as well, though. And only true for cloud when you have completely reliable networking. Hosted FaaS is a real challenge if your ISP drops.
No it's not. Because you can just go somewhere else. It's a real problem when it's locally hosted and your ISP drops and no customers can access it.
That 'somewhere else' is also more expensive though... Doesnt mattwr if its AWS, Azure, Google Compute, or another... It costs a more. Local hosted doesn't mean one bad ISP will destroy service for customers. We have out infrastructure in our facilities in London, Aberdeen, Barcelona, Middle East, Singapore, Houston, Calgary, Toronto and Singapore. I can lose any site and that will have no impact to customers. Still a lot cheaper than cloud.
Last time I looked, a single one of my calculation servers came to around 80k GBP. High core count, multiple TB of RAM. In Azure, over 6 years, MS were coming back with something in the range of £450k ex VAT. Insane difference.
That's completely different than the scenario he's talking about.
80k seems like an insane cost for a single server considering we did CFD, thermal, acoustics, and mechanics solves on a 20 node cluster made up of servers that were around $8K a piece.
-
For instance, each box had dual 20 core CPUs and didn't need a ton of RAM. There were a few clusters but the had to be shared by all of the engineers so it slowed everyone down. With AWS you can fire up a few c5.18s and pay $3.45 an hour. So a normal job for the engineers would cost around $165. But there's no maintenance of clusters, no infiniband to maintain, no waiting for other jobs to complete in the queue, and no OS maintenance. Create your AMI and spin up patched machines when necessary.
-
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
For instance, each box had dual 20 core CPUs and didn't need a ton of RAM. There were a few clusters but the had to be shared by all of the engineers so it slowed everyone down. With AWS you can fire up a few c5.18s and pay $3.45 an hour. So a normal job for the engineers would cost around $165. But there's no maintenance of clusters, no infiniband to maintain, no waiting for other jobs to complete in the queue, and no OS maintenance. Create your AMI and spin up patched machines when necessary.
What kind of calculations are that? I thought all HPC clusters ran on GPU power. A server with a couple of Nvidia V100 GPUs are expensive. Also infinitely more powerful than regular servers.
PS. Seems like a Nvidia V100 GPU card is about $6K each.
-
@Pete-S said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
For instance, each box had dual 20 core CPUs and didn't need a ton of RAM. There were a few clusters but the had to be shared by all of the engineers so it slowed everyone down. With AWS you can fire up a few c5.18s and pay $3.45 an hour. So a normal job for the engineers would cost around $165. But there's no maintenance of clusters, no infiniband to maintain, no waiting for other jobs to complete in the queue, and no OS maintenance. Create your AMI and spin up patched machines when necessary.
What kind of calculations are that? I thought all HPC clusters ran on GPU power. A server with a couple of Nvidia V100 GPUs are expensive. Also infinitely more powerful than regular servers.
PS. Seems like a Nvidia V100 GPU card is about $6K each.
These were for nuclear pumps. Since we had all kinds of engineers using them it was a general setup for everyone.
We had a couple of boxes with 4 GPUs each but they weren't used as much. It was more for Windows work.
They used ANSYS for most of their work and then also some internally developed software.
-
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@scottalanmiller said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Remember, I led the move to cloud on Wall St. Built the first cloud there.
To be transparent, I 100% do not believe this is true.
-
What did you use to provide the infrastructure?
-
How did you manage scaling without the needed APIs?
-
What year was this?
We were published and everything. JvU talks about the demo in that article, not production which was much larger. The cloud elements weren't added until after that point. Year was 2007. We had 10,000 nodes. We used a mixture of custom code (this is WS, you get all kinds of custom stuff built all the time and huge teams of dedicated people doing this all the time) and Symphony from Toronto, with their engineering constantly involved as we were the really big proving ground at the time. Scaling was pretty fast for the time, not like stuff today. But crazy for the time. They used the term grid back then, but it was a grid based cloud. At the time, the terms weren't very well known or solidified.
While there weren't containers used, the platform operated a lot like containers do today. Not only could 10K nodes / 40K CPUs dynamically run any workloads in any numbers that were needed, but the operating systems for them would dynamically adjust as well. It was a constant scaling operation and one of the largest of its time and the first on WS.
-
-
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@scottalanmiller said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Containerization and FaaS have made development and deployments insanely easier and less costly.
True for non-cloud installs as well, though. And only true for cloud when you have completely reliable networking. Hosted FaaS is a real challenge if your ISP drops.
No it's not. Because you can just go somewhere else. It's a real problem when it's locally hosted and your ISP drops and no customers can access it.
Most companies can't dynamically move their workforce when the services are for a physical location. It's absolutely true in the real world that access to your IT infrastructure is critical where you work. Most of the world can't just send everyone home or own backup sites.
Tell a doctor's office, vet clinic, restaurant, retail store, manufacturing plant that their access to servers "doesn't matter" and that as long as they are hosted cloud they don't need Internet from where they work because they can just "relocate" anytime there is a blip. What?
-
@Pete-S said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
What kind of calculations are that? I thought all HPC clusters ran on GPU power.
Probably the majority do, now. But HPC was a bit deal long before GPU offloading was a thing (in those scenarios.) GPU use in HPC is relatively new. Really great and useful, but new and only works most of the time. There are still workloads needing HPC that don't benefit or benefit much from GPUs and those GPUs tend to be pretty expensive in absolute terms.
-
American Banker talks about it as it grew to the first product size. It's 2008 and the article misuses Cloud to mean public hosted cloud, but clearly JvU talks about CG's internal cloud being the future. They were using grid to refer to private cloud. This was 12-13 years ago and the terms were still getting popular acceptance - cloud only became a common term in later 2007 and we all remember what a cluster (pun intended) that was to figure out. Grid was a term being thrown around for this private cloud stuff in the later 2000s, but it all just got renamed cloud soon thereafter.
JvU now heads cloud infrastructure at Toronto Dominion.
-
@scottalanmiller said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@stacksofplates said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
@scottalanmiller said in Microsoft plans on retiring the MCSA,MCSD,MCSE certifications in June 30,2020:
Remember, I led the move to cloud on Wall St. Built the first cloud there.
To be transparent, I 100% do not believe this is true.
-
What did you use to provide the infrastructure?
-
How did you manage scaling without the needed APIs?
-
What year was this?
We were published and everything. JvU talks about the demo in that article, not production which was much larger. The cloud elements weren't added until after that point. Year was 2007. We had 10,000 nodes. We used a mixture of custom code (this is WS, you get all kinds of custom stuff built all the time and huge teams of dedicated people doing this all the time) and Symphony from Toronto, with their engineering constantly involved as we were the really big proving ground at the time. Scaling was pretty fast for the time, not like stuff today. But crazy for the time. They used the term grid back then, but it was a grid based cloud. At the time, the terms weren't very well known or solidified.
While there weren't containers used, the platform operated a lot like containers do today. Not only could 10K nodes / 40K CPUs dynamically run any workloads in any numbers that were needed, but the operating systems for them would dynamically adjust as well. It was a constant scaling operation and one of the largest of its time and the first on WS.
That's LSF. It's just a job scheduler like OpenFOAM, ANSYS, Slurm, Flux, etc. That's like a big Beowulf cluster but it's grid computing, not cloud computing.
-
