71–80 of 141 posts
With such slim margins AWS and the larger cloud services industry of today will become what the airlines industry has become.
Earlier quoted context omitted.
What's the basis of your argument that Google Cloud is better? My personal experience has been exactly the opposite - AWS is light years ahead of Google Cloud. Does Google Cloud provide Machine Learning service? For the longest time, it was only supporting Java and Python SDKs. And the Google's API is still very much lacking in terms of features and stacks. In fact the better support and much comprehensive services p…
> Does Google Cloud provide Machine Learning service? Yes they do. [0] > And this is common, as we've just seen people complain about Google Cloud 4 days ago. To me, this is an isolated incident. On the other hand, seemingly every AWS customer of non-trivial size had to reboot a significant part of their machines in the last year because of Xen vulnerabilities (with no way to predict if you will land on a patched mac…
in the style of Drake equation: $1.5B/quarter, supposedly 1/3 of it is billed CPU time, at $1/hour per one 1U computer, at 500 hours billed per quarter per computer we have 1M computers - at 40 boxes per rack we have 25K racks, at 50 racks/row it would be 500 rows, at 50 rows per "warehouse" it will be 10 "warehouses". The storage money and racks i'd suppose is comparable to CPU boxes in numbers. So in result it woul…
I like your approach, and have some thoughts relevant. What's a computer? The $1/hour revenue is generated by AWS from FAR less than 1U. SuperMicro sells a 2U machine with 4 full sized computers in it with 6 drives. You can put 12 cores in that, and an AWS 'ECU' core is a hyperthreaded core. So, in the Supermicro "quadnode" [1] version you get 4 * 12 * 2 = 96 cores in 2U, or 48 cores per U. An AWS computer/instance t…
i don't think AMZN or Google put 4 sockets per 1 U. While obviously it is possible, it would be pretty dense for normally powered chips. My understanding is that 2 socket/1U is more typical. At 100W/chip it would be a difference between 8KW and 16KW heat from a rack. But anyway, you're right - AMZN probably gets more than $1/1U. Their cluster 2-socket (16 real, 32 v-cores) instances go for $3.5/hour. Starting from that number we can adjust it down somewhat for pre-paid/reserved/spot pricing. Anyway, lets hope fuzziness in one number is compensated (not amplified :) by fuzziness in another ... Actually from the 2013 link you posted in the response below (that one - https://storageservers.wordpress.com/2013/07/17/facts-and-st...):
"Amazon data center- Amazon has around 450,000 servers in its data centers sited in 7 locations in the world. "
So adjusting for 1+ years of growth (of at least 10-20% as they had 49% revenue increase) my original estimate was less than 2 times wrong :) Edit: actually adding storage servers - may be up to 4 times wrong if they have all 1U servers, so the most probable parameter to adjust would be $/1U to something like $2+/1U.
Earlier quoted context omitted.
Not parent but I'd guess they are talking about OpenStack but if they are they are wrong. Running a data center is no easy or cheap task. Sure you might have SOME of the logistics cut out for you using OpenStack but the barriers to entry are still VERY high.
Openstack + spare capacity in a colo facility + a bunch of few years old servers offloaded in a facilities auction and you can build one pretty quickly. Tech isn't the hard part - barriers to get a service up and running are dropping drastically. Hard part is marketing & scaling, and as AWS expands services there are risks from pure-play vendors that do just one slice of the stack 2-3x better than Amazon does.
1. OpenStack still has plenty of bugs and issues you are going to hit.
2. You'll still need a really good colo facility.
3. High bandwidth and Multiple internet providers are expensive. If you look at Amazon they have an excellent blend of bandwidth including direct connectivity to many eyeball providers (Comcast, Charter, RCN etc.) which usually makes your data get there quickly and prevents peering bottlenecks.
4. If you need more than one facility then you just get to double your costs.
5. People costs, you'll need people that can manage your infrastructure at scale.
6. Hardware capex can severely drain companies.
7. Vendor sprawl and support. (You still may need a CDN, DNS provider, etc.)
Doing your own thing can quickly become much more expensive than Amazon or other cloud providers.
"These have since become common building blocks of Internet-based computing systems, supporting companies ranging from large enterprises like Infor US Inc. and Netflix Inc. to startups such as Instacart Inc." When I see a brand name mentioned randomly in an arricle, I wonder who paid for it. Why would Bloomberg mention instacart specifically? Is it common practice for companies to pay for these subtle pseudo-mentions…
Earlier quoted context omitted.
I like your approach, and have some thoughts relevant. What's a computer? The $1/hour revenue is generated by AWS from FAR less than 1U. SuperMicro sells a 2U machine with 4 full sized computers in it with 6 drives. You can put 12 cores in that, and an AWS 'ECU' core is a hyperthreaded core. So, in the Supermicro "quadnode" [1] version you get 4 * 12 * 2 = 96 cores in 2U, or 48 cores per U. An AWS computer/instance t…
> SuperMicro sells a 2U machine with 4 full sized computers in it with 6 drives. You can put 12 cores in that i don't think AMZN or Google put 4 sockets per 1 U. While obviously it is possible, it would be pretty dense for normally powered chips. My understanding is that 2 socket/1U is more typical. At 100W/chip it would be a difference between 8KW and 16KW heat from a rack. But anyway, you're right - AMZN probably g…
Earlier quoted context omitted.
I like your approach, and have some thoughts relevant. What's a computer? The $1/hour revenue is generated by AWS from FAR less than 1U. SuperMicro sells a 2U machine with 4 full sized computers in it with 6 drives. You can put 12 cores in that, and an AWS 'ECU' core is a hyperthreaded core. So, in the Supermicro "quadnode" [1] version you get 4 * 12 * 2 = 96 cores in 2U, or 48 cores per U. An AWS computer/instance t…
> SuperMicro sells a 2U machine with 4 full sized computers in it with 6 drives. You can put 12 cores in that i don't think AMZN or Google put 4 sockets per 1 U. While obviously it is possible, it would be pretty dense for normally powered chips. My understanding is that 2 socket/1U is more typical. At 100W/chip it would be a difference between 8KW and 16KW heat from a rack. But anyway, you're right - AMZN probably g…
http://www.intel.co.jp/content/dam/www/public/us/en/document... covers it fairly well.
"Since its inception, Amazon has refined and expanded AWS while competitors including Microsoft and Google have tried to replicate its success with similar projects. Those companies have only recently started to deliver basic services on par with Amazon’s." Is this really true? I know that AWS offers many more services than competitors currently, but for the services for which there is a competing product, is the AWS…
Compared to Azure, yes. The IaaS offering on Azure has many, many limitations. It is also very difficult to use. There is no general pool of servers. You have to cram VMs into these groupings called "Cloud Services" which are limited to 50 VMs each. No support for TLS termination on their load balancers. SSDs are currently a "preview" feature and only work on a subset of hardware types. In all fairness AWS had a 4-5…
But no TLS termination? At a company that has invested $15bn on their cloud products and has in-house expertise on the actual source code of one of the most widely-used networking stacks? That's... surprising, to say the least.
Earlier quoted context omitted.
Most analysts seem to think that AWS is way bigger than Google's public cloud. That channel lead may well be right in terms of total number of servers or MW of datacenter capacity. BUT scale is not the only thing that the cloud brings. AWS is not just S3 and EC2. I think it's pretty clear that Amazon is much further along the road of productizing their cloud. i.e. providing all the products that businesses and most n…
Published information I have dug up shows Google's cloud is larger, but only by 50% and the dates were in the 2012-2013 range. Regardless of whether the channel lead's motivations are to say one thing or another, he did say something that struck me with conviction: "We're not going anywhere. We're NOT relinquishing this market." My experience with Kubernetes and Managed VMs and as an App Engine dev since 2009 compare…
"We're not going anywhere. We're NOT relinquishing this market."
I think for a while some didn't believe how serious Google was about the whole selling cloud services thing. Now they do.
Regarding Google and Amazon's disparate profitability:
1) Cash is arguably more important than profitability from an investment financing perspective. Amazon and Google are in the same ball park regarding free cash flow (>1B, 2) Secondly, for the sake of their stock price, Google would be very hesitant to have their high margin search/ads business become dominated by a low-margin commodity cloud business. Their scope to use all of the free cash generated from search/ads for their cloud business is limited. Stock analysts don't like declining profitability whereas Amazon already has basically zero profits and is basically priced at a Price/Sales ratio like a large retailer.
With such slim margins AWS and the larger cloud services industry of today will become what the airlines industry has become.
To get into the same cloud market as AMZN/MSFT/GOOG you are going to need:
1. Several billion dollars of Capex to get to the same scale as the incumbents. 2. A large amount of investment to replicate all the internal software that Amazon and friends have. 3. Goodwill from startups and/or enterprise CIO types.
Getting into the airline business is comparatively easy:
1. You lease the planes from Boeing or Airbus. 2. Pilots and other staff are highly skilled but otherwise fungible. 3. Marketing through traditional consumer channels and price comparison engines is very well understood and if you're willing to eat the losses then growing share is doable.