Skip to content
 
If you can read this, either the style sheet didn't load or you have an older browser that doesn't support style sheets. Try clearing your browser cache and refreshing the page.

(The Stack)   Microsoft Azure beats supercomputers for scientific number-crunching   (thestack.com) divider line
    More: Interesting, Cloud computing, Azure cloud computing, Microsoft Azure, Microsoft Azure A-instances, LINPACK testing suite, cloud-based HPC services, single computing core, modern traditional supercomputing  
•       •       •

2052 clicks; posted to Fandom » on 14 Feb 2017 at 12:07 AM (4 years ago)   |   Favorite    |   share:  Share on Twitter share via Email Share on Facebook



24 Comments     (+0 »)
View Voting Results: Smartest and Funniest
 
2017-02-14 12:12:51 AM  
Can you imagine a Beowulf cluster of these things?
 
2017-02-14 12:23:37 AM  

Vlad_the_Inaner: Can you imagine a Beowulf cluster of these things?


Does it scream "I AM BEOWULF!!" every 5 minutes?
 
2017-02-14 12:30:51 AM  
So they weren't using Excel 2002?
 
2017-02-14 12:38:33 AM  
Um yes.

Throw enough shiat at the wall and you'll get great performance.

The question is one of cost (??????) and convenience (Cloud wins forever).
 
2017-02-14 12:50:54 AM  

meyerkev: Um yes.

Throw enough shiat at the wall and you'll get great performance.

The question is one of cost (??????) and convenience (Cloud wins forever).


And cost comes down to need. Each of the major cloud services have their own sewer spot in use from what I've seen.
 
2017-02-14 1:14:52 AM  
Microsoft.. FOR GREAT SCIENCE!!
 
2017-02-14 2:57:38 AM  
img.fark.netView Full Size

"This is an ad."

 
2017-02-14 4:12:59 AM  
Man, I remember when Microsoft used to pay "independent benchmarking companies" to shill like this. My guess is someone got a promotion and brought this style of advertising back.

The Apple Mac doesn't need anyone to astroturf for it. Its high speed CPUs, blazing fast memory, and sizzling hot styling make it stand apart from every other PC on the market.
 
2017-02-14 4:14:51 AM  

AverageAmericanGuy: Man, I remember when Microsoft used to pay "independent benchmarking companies" to shill like this. My guess is someone got a promotion and brought this style of advertising back.

The Apple Mac doesn't need anyone to astroturf for it. Its high speed CPUs, blazing fast memory, and sizzling hot styling make it stand apart from every other PC on the market.


4 years ago, that was actually more or less true.  I'm typing on one right now, and it's wicked.

Today?  Not so much.
 
2017-02-14 4:55:06 AM  
There is no cloud.

It's just somebody else's computer.
 
2017-02-14 6:28:29 AM  

Ishkur: There is no cloud.

It's just somebody else's computer.


In some cases running on eval licenses...

Seriously, a company I used to work for decided they wanted to offer a "Cloud Load Balancing" service.  So they sold it to a customer, then got me to build it out, but didn't actually put any money into the infrastructure.  Apart from the fact it wasn't "Cloud" or "Load Balancing" as such (it was hosted intelligent DNS)

So it ran on 3 * F5 Big-IP Strongboxes (running GTM), which are vendor eval boxes with 30 day licenses.

I left the company 18 months later, the infrastructure guy still had a reminder every 28 days to relicense the F5 eval boxes so that the 10 or so clients didn't just drop offline.  At least we staggered the renewals so all 3 boxes didn't fail together.

Someone elses computer, where the service is held together with string, spit and hope sometimes.
 
2017-02-14 7:40:37 AM  
hey Cortana... Do science
 
2017-02-14 8:24:10 AM  
So yeah, this does feel like an ad.   But it's also pretty much true.   I've done some amount of HPC for 20+ years, built my own beowulf cluster, and we had our own Cray at my office for a long time.   Last weekend I spun up an Azure VM with a GPU and ran some machine learning stuff on it.    It was just a test and it worked fine.    The thing is that it was easyto do.   The ML stuff is easy to parallelize, and VMs are trivial to clone.   So if I felt like solving a problem and had a budget, I could spin up whatever I want and when I'm done  **poof** it's gone.
 
2017-02-14 8:32:05 AM  

Ishkur: There is no cloud.

It's just somebody else's computers.


FTFY. But yes, that's exactly what "the cloud" means. You're off-loading the need for maintaining physical hardware and managing the servers to a third party who do them for you. Everyone knows this - it's not like "the cloud" is a new term by any means.
 
2017-02-14 9:12:49 AM  

Ishkur: There is no cloud.

It's just somebody else's computer.


Someone else who is legally bound by a SLA to keep those computers up and running.
 
2017-02-14 9:34:39 AM  

Mad_Radhu: Ishkur: There is no cloud.

It's just somebody else's computer.

Someone else who is legally bound by a SLA to keep those computers up and running.


...and which gets violated often enough to trim those profit margins to the bone. An SLA isn't a shield. It's just a contract - one which limits the liability exposure of the provider to its customers should that provider trigger an SLA violation. "Oopsie! Your entire business went down for 2 hours? Here's an hour or so of usage credit. What? You lost millions because your customers couldn't access your data and services? Well, probably shouldn't have relied on someone else's computer for all of that, huh? Good luck!"

Cloud computing has its benefits - and it has a lot of them - but it's a tool that has to be applied judiciously. The more critical your data and services, the less likely that they should be running on someone else's computer without some form of "oh, shiat!" on-prem fallback.
 
2017-02-14 9:56:45 AM  

FormlessOne: Mad_Radhu: Ishkur: There is no cloud.

It's just somebody else's computer.

Someone else who is legally bound by a SLA to keep those computers up and running.

...and which gets violated often enough to trim those profit margins to the bone. An SLA isn't a shield. It's just a contract - one which limits the liability exposure of the provider to its customers should that provider trigger an SLA violation. "Oopsie! Your entire business went down for 2 hours? Here's an hour or so of usage credit. What? You lost millions because your customers couldn't access your data and services? Well, probably shouldn't have relied on someone else's computer for all of that, huh? Good luck!"

Cloud computing has its benefits - and it has a lot of them - but it's a tool that has to be applied judiciously. The more critical your data and services, the less likely that they should be running on someone else's computer without some form of "oh, shiat!" on-prem fallback.


Dude... Medical I.T.? Yeah, it's all at the mercy of outside servers and your doctor's cheapest possible ISP, decided upon for him by the management company s/he "partnered" with. Hope your doc isn't on legacy copper AT&T lines that "AT&T" is trying to slowly force them off of by a process of staged, random service outages followed by the promise: "Gee, this wouldn't happen if you would pay 2x for our fiber service".

Ask me how I know this. HIPAA compliant doesn't necessarily mean "readily available".
 
2017-02-14 11:01:22 AM  

meyerkev: Throw enough shiat at the wall and you'll get great performance.

The question is one of cost (??????) and convenience (Cloud wins forever).


I would think Amazon and IBM also have a lot of shiat they can throw at the wall.

Also, Azure is a cloud service so I'm not sure what your second point is.
 
2017-02-14 11:28:59 AM  
The actual paper is interesting:

https://arxiv.org/pdf/1702.02968.pdf

AWS actually outperforms Azure in terms of teraflops, but Azure wins on frequency and "speedup". I don't know the hardcore CS enough to know why you would use one measure over the other.
 
2017-02-14 12:31:25 PM  

Arkanaut: The actual paper is interesting:

https://arxiv.org/pdf/1702.02968.pdf

AWS actually outperforms Azure in terms of teraflops, but Azure wins on frequency and "speedup". I don't know the hardcore CS enough to know why you would use one measure over the other.


Speedup is the 5 Hour Energy of cloud computing.
 
2017-02-14 2:19:28 PM  

Arkanaut: meyerkev: Throw enough shiat at the wall and you'll get great performance.

The question is one of cost (??????) and convenience (Cloud wins forever).

I would think Amazon and IBM also have a lot of shiat they can throw at the wall.

Also, Azure is a cloud service so I'm not sure what your second point is.


That's my point.

You can get a special box, or you can make a GetInstances call, spin up a couple hundred EC2 nodes with SSD's on them hooked into S3, and make all the management EC2's problem.

FormlessOne: Mad_Radhu: Ishkur: There is no cloud.

It's just somebody else's computer.

Someone else who is legally bound by a SLA to keep those computers up and running.

...and which gets violated often enough to trim those profit margins to the bone. An SLA isn't a shield. It's just a contract - one which limits the liability exposure of the provider to its customers should that provider trigger an SLA violation. "Oopsie! Your entire business went down for 2 hours? Here's an hour or so of usage credit. What? You lost millions because your customers couldn't access your data and services? Well, probably shouldn't have relied on someone else's computer for all of that, huh? Good luck!"

Cloud computing has its benefits - and it has a lot of them - but it's a tool that has to be applied judiciously. The more critical your data and services, the less likely that they should be running on someone else's computer without some form of "oh, shiat!" on-prem fallback.


And this.

/Though it's funny reading a tweet about your service. "WOOHOO, hardware failure recovery in 20 minutes!" and you're going "Biatch, give me your cluster id, because it shouldn't take 20 minutes, it should take 3, and I want to know what went wrong there".
 
2017-02-14 7:14:36 PM  

thespindrifter: FormlessOne: Mad_Radhu: Ishkur: There is no cloud.

It's just somebody else's computer.

Someone else who is legally bound by a SLA to keep those computers up and running.

...and which gets violated often enough to trim those profit margins to the bone. An SLA isn't a shield. It's just a contract - one which limits the liability exposure of the provider to its customers should that provider trigger an SLA violation. "Oopsie! Your entire business went down for 2 hours? Here's an hour or so of usage credit. What? You lost millions because your customers couldn't access your data and services? Well, probably shouldn't have relied on someone else's computer for all of that, huh? Good luck!"

Cloud computing has its benefits - and it has a lot of them - but it's a tool that has to be applied judiciously. The more critical your data and services, the less likely that they should be running on someone else's computer without some form of "oh, shiat!" on-prem fallback.

Dude... Medical I.T.? Yeah, it's all at the mercy of outside servers and your doctor's cheapest possible ISP, decided upon for him by the management company s/he "partnered" with. Hope your doc isn't on legacy copper AT&T lines that "AT&T" is trying to slowly force them off of by a process of staged, random service outages followed by the promise: "Gee, this wouldn't happen if you would pay 2x for our fiber service".

Ask me how I know this. HIPAA compliant doesn't necessarily mean "readily available".


I don't know anything about running a business, but I do know there's such a thing as "business interruption insurance" - point being, people sometimes buy it, sometimes get payouts, and regardless, keep struggling along. Also, I have the impression that a lot of people just run their business until something serious goes wrong and then fail. Failure is always an option, as they say.

You can't get too hung up on the interdependence you have with society, because even if you avoid the cloud, you're still dependent in a million other ways. You can prepare for disaster, but you can't prepare for all disasters, for an indefinite amount of time.
 
2017-02-14 10:21:09 PM  

Arkanaut: The actual paper is interesting:

https://arxiv.org/pdf/1702.02968.pdf

AWS actually outperforms Azure in terms of teraflops, but Azure wins on frequency and "speedup". I don't know the hardcore CS enough to know why you would use one measure over the other.


Glancing through, it looks like Azure had the highest single-core performance, and maintained the speed as node count increased - 32 nodes delivered roughly 28x the performance, compared to, say, only a ~4x speedup with 32-nodes for the IBM SoftLayer service.
 
2017-02-14 11:38:29 PM  
The headline is totally wrong.  Azure was compared to rival distributed systems not standalone supercomputers.
 
Displayed 24 of 24 comments

View Voting Results: Smartest and Funniest

This thread is archived, and closed to new comments.

Continue Farking




On Twitter


  1. Links are submitted by members of the Fark community.

  2. When community members submit a link, they also write a custom headline for the story.

  3. Other Farkers comment on the links. This is the number of comments. Click here to read them.

  4. Click here to submit a link.