# C[omp]ute

Welcome to my blog, which was once a mailing list of the same name and is still generated by mail. Please reply via the "comment" links.

Always interested in offers/projects/new ideas. Eclectic experience in fields like: numerical computing; Python web; Java enterprise; functional languages; GPGPU; SQL databases; etc. Based in Santiago, Chile; telecommute worldwide. CV; email.

© 2006-2015 Andrew Cooke (site) / post authors (content).

## Cloud Computing

From: andrew@...

Date: Fri, 9 Jul 2010 18:13:57 -0400

This is a short summary of Cloud Computing and various related ("trendy")
technologies.

The driving force for much of these ideas seems to have come from Amazon and
Google as they have dealt with scaling their businesses (it's strange how much
Amazon is involved in this, rather than Microsoft, for example). The simple
idea behind Cloud Computing is that computers, and the basic functionality
they provide, are becoming "comoditized". The usual comparison is with an
Electric Company, that supplies electricity. Just as you don't care exactly
what generator is used, so you shouldn't need to care what hardware is used to

There are two main approaches to this, one taken by Amazon, and the other by

Amazon's approach is "Infrastructure as a Service" (Amazon Web Services,
AWS). By that, they mean that they provide basic infrastructure as the
"commodity". So you can configure a "machine" and then run N instances, where
N varies depending on the current load. Similarly, you can access "databases"
and "permanent stores". So this is very like "normal" computing, except that
you have services that are "elastic" - they scale as required, typically by

Google's approach is "Platform as a Service". By that, they mean that they
offer a set of APIs that you can use to write a webservice, and then they will
run that service for you, scaling as necessary. So you work at a higher level
of abstraction: you write code that handles a single thread; Google will make
sure that as many threads as necessary run. For data storage, you save data to
a kind-of database (more like a hash table) and you don't care about the lower
level details.

So Google's "App Engine" approach is easier to use (although you need to learn
a whole new API in either Python or Java), but more limiting. In very simple
terms, Amazon's approach is like owning an expandable set of virtual machines
while Google's approach is more like having an infinitely powerful single
machine. One cool detail about Google's approach is that, for small services,
it's free.

And of course there are many more providers, generally doing things like
Amazon (partly because no-one apart from Google is big enough that it's worth
learning a new API just to use their service).

Now, tangentially related to that, are the spin-off technologies. Two are
particularly important, and are related: MapReduce and NoSQL.

To understand both, and how they are related, it's probably easiest to start
across many machines. And to run a search they need to run tasks on each
machine and then collect the results. This process is called MapReduce: "map"
is related to dividing the work out, doing the same thing on various machines;
"reduce" is the process of collecting the results together.

Now MapReduce doesn't make much sense without data. So the flip side of
MapReduce is the distributed data. And that implies some kind of distributed
database. It turns out that databases can do 2 out of 3 things well, where the
3 things are: availability; consistency; partitions. SQL databases are "AC",
but aren't so good at P. Spreading data across various machines requires
partitioning (P), so new technologies have appeared that are either "AP"
(don't always give consistent results) or "CP" (are not guaranteed
available). These tend not to use SQL and are collectively called "NoSQL".

Curiously, Google, who invented MapReduce, don't actually let the public use
it. However, Amazon do support it as a service, and there are many 3rd party
solutions that you can also run yourself (most famously, CouchDB, MongoDB,
HBase and Cassandra). All of these support MapReduce implementations. And
Google's App Engine (its PaaS - see above) does have support for something
similar, but it's not the same MapReduce that Google use internally.

Andrew

### More on CAP

From: andrew cooke <andrew@...>

Date: Mon, 12 Jul 2010 05:23:39 -0400

This clarifies CAP -
http://dbmsmusings.blogspot.com/2010/04/problems-with-cap-and-yahoos-little.html

Andrew