[Interest] Interest Digest, Vol 80, Issue 7

Sat May 12 15:48:36 CEST 2018

Sorry for the delayed response. Been helping with planting season back 
on the family farm.

On 05/07/2018 11:39 PM, Thiago Macieira wrote:
> On s?bado, 5 de maio de 2018 07:58:45 PDT Roland Hughes wrote:
>> Data centers have been moving to ARM (the new absolute bottom) and
>> Z-machines for quite some time now.
> And yet they represent less than 0.1% of the market share.
You make me smile man. You really do.

https://www.youtube.com/watch?v=T-NpLu2xC38

I don't even like IBM, but, that cute little video is the reality. Once 
your server farm exceeds a certain size, you need to move to big-iron to 
lower heat, power consumption and future cost.

I love the accounting. It reminds me of Janet Reno investigation of 
Microsoft where Microsoft forced motherboard vendors to record a sale of 
Microsoft Windows with every motherboard sold, despite the fact no 
Windows product went out and the fact it was the peak of Netware 
servers. North of 80% of those mobos went into a cabinet where DR DOS 
and Netware were installed.

When a firm brings in something like a Z13 it replaces somewhere north 
of 200,000 x86 servers (depending on which Z13 model/configuration). 
Under this accounting the 200,000 x86 stay on the books and the Z13 is 
counted as a single unit.

One, granted it's the biggest one, Z13 has been tested and rated to 
handle the transactional throughput of 200 Cyber Mondays at once. Not 
just for Amazon. Not just for Walmart. The measured Cyber Monday traffic 
of the entire Internet including all retailers and porn surfers.

The Z14 goes even farther.

The Big Endian default you wish to change was put there to communicate 
with these galactic throughput engines so that lesser species might 
still have a purpose in life.

>
>> The x86 has both heat and power problems which cannot be overcome.
> I'm sure my colleagues at Intel would be very interested in your detailed
> analysis that shows why the problems cannot be overcome. Please forward me
> your papers if you can.
>
> If not, you're boasting and making unfounded generalisations. You're spreading
> FUD.
Here is some light reading. I pulled some of the interesting snippets 
out and pasted them below the links. You really should read the entire 
Cloudflare article though.

https://siliconangle.com/blog/2017/03/09/microsoft-paves-way-arm-based-server-chips-data-center/

First, the company said it’s collaborating with Cavium Inc. to run 
Microsoft Azure cloud workloads on Cavium’s ARM-based server processors.
The announcements suggests Microsoft is making a significant shift 
towards ARM processors

https://www.electronicsweekly.com/news/business/china-driving-arm-based-server-market-2015-10/

PayPal found, says Williams, that using ARM instead of Intel delivered 
900% higher node density, 85% lower power consumption at a 45% lower 
acquisition cost. Total cost of ownership of an ARM-based system is 35% 
lower than an x86-based system, he said.

http://www.datacenterknowledge.com/design/cloudflare-bets-arm-servers-it-expands-its-data-center-network

To keep further expansion affordable, Cloudflare is planning a massive 
architectural change. The company will soon start deploying ARM servers 
in its data centers, Prince said, expecting that the alternative to x86 
will be cheaper to buy and to keep running due to their lower power 
requirements and to the nature of Cloudflare’s workload.

The company has already moved away from Intel SSDs after finding the 
performance wasn’t enough for its needs. (All Cloudflare servers use 
SSDs to cache data from the web sites it protects.) “I’d give 
better-than-even odds that by Q4 this year we will no longer spend any 
money with Intel,” Prince told us.

“We think we're now at a point where we can go one hundred percent to 
ARM. In our analysis, we found that even if Intel gave us the chips for 
free, it would still make sense to switch to ARM, because the power 
efficiency is so much better.”

http://www.electronicdesign.com/embedded-revolution/are-intel-processors-apple-s-mac-chopping-block

On April 2, news spread quickly of the rumor that Apple plans to replace 
Intel processors with its own internally designed Arm-based processors 
for its Mac platform.

https://www.anixter.com/content/dam/Suppliers/Upsite/Questions%20and%20Answers%20on%20Data%20Center%20Cooling%20Issues.pdf

Q. Blade servers consume unprecedented amounts of power, all
of which is converted to heat and must be managed. What impact
will this new generation of products have on my data center?
A. The power consumption and density for this generation of servers
is much higher than for previous generations. All major manufacturers
have similar products with power consumptions ranging
from 8 kW to 20 kW per rack, with 30+ kW products on the drawing
board. Many sites will have severe problems cooling even small
quantities of these computers. The volume of air delivered to the
rack from the underfloor will become even more critical. Activities
such as eliminating bypass airflow, placing the right number of
perforated tiles or grates in the cold aisle only, installing blanking
panels in rack openings and matching cooling airflow with server
cooling needs are all essential to directing the available supply of
cold air accurately.

https://venturebeat.com/2018/02/05/intels-former-president-leads-arm-based-server-chip-maker-ampere/

https://www.computerworld.com/article/3005816/servers/server-vendors-tap-arm-chips-to-give-users-alternative-to-intel.html

Five computer makers have announced servers with ARM processors that 
will challenge x86 systems in the mainstream market. The systems are 
largely for Internet and cloud workloads and have the 48-core Cavium 
ThunderX chip, which is based on 64-bit ARM architecture.

The servers from Gigabyte, Inventec, Wistron, Penguin Computing and E4 
Computer Engineering are based on designs commonly used in x86 servers, 
but have ARM processors. An interesting twist in some new servers is the 
ability to also use Nvidia's Tesla graphics processors, adding extra 
processing punch for graphics, engineering and other high-performance 
computing applications.

https://www.reuters.com/article/us-cyber-microchips-cloud/cloud-companies-consider-intel-rivals-after-security-flaws-found-idUSKBN1EZ1A4

https://software.intel.com/en-us/blogs/2014/02/19/why-has-cpu-frequency-ceased-to-grow

Linear frequency growth causes power dissipation to be increasingly 
cubed! If the frequency is raised only twice, there will be eight times 
greater heat that must be accommodated or the processor will melt or 
shutdown.

http://www.informit.com/articles/article.aspx?p=2142913

Unfortunately, while the number of transistors that can be put on a chip 
cheaply has increased, the power consumption per transistor hasn't 
dropped at a corresponding rate. The amount of power per transistor has 
dropped, but more slowly than the size of the transistor has shrunk. 
Eventually, you hit some real physical limits; this problem isn't just 
theoretical. The first mainstream microprocessor to encounter this 
serious problem was the Pentium 4, a decade ago.

The heat generation per unit area of an integrated circuit passed the 
surface of a 100-watt light bulb in the mid 1990s, and now is somewhere 
between the inside of a nuclear reactor and the surface of a star. When 
something is generating that much heat, keeping it in the temperature 
range where silicon transistors work is quite difficult.

https://www.chieftain.com/business/tech/flip-phones-are-the-new-protest-statement/article_eca7bb5a-8c97-5a17-a95a-46a2e01f5114.html

https://www.ibm.com/blogs/systems/the-next-phase-of-secured-computing-ibm-cloud-private-on-z/

https://www.zdnet.com/article/the-mainframe-lives-on-in-ibms-linuxone/

https://www.wired.com/2015/01/z13-mainframe/

https://www.wired.com/insights/2014/04/back-future-cloud-wont-replace-mainframe/

https://www.suse.com/c/mainframe-versus-server-farm-comparison/

https://www.computerworld.com/article/2872096/ibm-s-z13-and-the-case-for-the-mainframe-cloud.html

http://enterprisesystemsmedia.com/article/mainframes-are-still-at-the-heart-of-the-modern-tech-world#sr=g&m=o&cp=or&ct=-tmc&st=(opu%20qspwjefe)&ts=1525461227

http://www.datacenterknowledge.com/design/cloudflare-bets-arm-servers-it-expands-its-data-center-network

https://www.suse.com/c/mainframe-versus-server-farm-comparison/

Throughput is a distinct mainframe characteristic, as it supports large 
numbers of simultaneous transactions and massive I/O without slowing 
down. An x86/ARM server loses efficiency at >20% total load, and will 
slow down when any single component is overloaded; for example when its 
CPU is running at 100%, or memory is exhausted, or the hard disk 
thrashes. This doesn’t happen on a mainframe, which maintains peak 
performance up to 90% load.

If a hardware component fails replace it, and work around the failure 
with some kind of automatic failover that transfers its workload to a 
different component. You don’t nurture your individual pieces like pets, 
but treat them like herds of cattle, a faceless mass of disposable 
units. Docker, Apache Mesos, Marathon, ZooKeeper, Chef, Puppet, DC/OS… 
all of these new technologies enable amazingly flexible service 
development and delivery, at the cost of an extremely complex system 
that is burdened with mitigating regular failures.

>> Even HP switched to selling ARM based servers in 2014.
> Where by "switched", you mean "added to the sale roster but still sells less
> than 0.1%".
>
No I mean as in added new ARM based products and deprecated x86 based to 
sales staff or so I've heard.

-- 
Roland Hughes, President
Logikal Solutions
(630)-205-1593

http://www.theminimumyouneedtoknow.com
http://www.infiniteexposure.net
http://www.johnsmith-book.com
http://www.logikalblog.com
http://www.interestingauthors.com/blog
http://lesedi.us/
http://onedollarcontentstore.com

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.qt-project.org/pipermail/interest/attachments/20180512/c588c988/attachment.html>