title: Developing Cloud Software
subtitle: Algorithms, Applications, and Tools
authors:
- Ivan Porres
- Tommi Mikkonen
- Adnan Ashraf
publisher: TUCS General Publication
date: October 2013I found a pdf of the book free online. Due to time constraints, I cannot take detailed notes for the website at this time.
p. 1
Clouds are a large pool of easily usable and accessible virtualized resources, like hardware, which can be dynamically reconfigured. Companies hosting cloud resources benefit from economics of scale for operational costs.
This chapter looks at:
We all know Amazon, but another player is Cloud Computing, Security, Content Delivery (CDN) | Akamai…
Software layer providing virtualization is called a virtual machine monitor or hypervisor.
Services can be divided into:
There are 2 main virtualization techniques:
2 main hypervisor types:
Wikipedia has a comprehensive list of different virtualization solutions, both open source and commercially licensed. The book talks about Home - Xen Project (opensource), VMWare, and KVM. The books says that XEN is marketed by Explore Citrix Products for Enterprise and Medium Business - Citrix, but it may have moved since 2013.
I also want to shout out VirtualBox - Wikipedia, because I have used it before. However, XEN is a type 1 and VirtualBox is a Type 2.
Sandboxing separates programs and resources from each other. Good for using experimental software in same infrastructure as the production software.
Goal of storage virtualization is to abstract the physical location of the data from the users and developers. Role of Virtualization storage is to provide a mapping from the perceived data to the actual physical location. Some benefits of virtualized storage systems:
OpenFlow, Product Certification - Open Networking Foundation, is the first Software-Defined Networking (SDN) and specification, and a vital element of an open software-defined network architecture. It allows developers to run experimental protocols in production networks. It is a feature added to switches, routers, access points, and basestations, allowing these datapath devices to be controlled through an external, standardized API.
Software-defined networking is expected to be one of the new emerging research topics in computer networking.
NOX is an open source network control platform that can control all connectivity on the network.
Welcome to Nox — Nox 2023.4.22 documentation (thea.codes) This Nox looks like Python automation testing…
SDN Series Part Three: NOX, the Original OpenFlow Controller - The New Stack goes to NOX Repo | Github… There’s also GitHub - noxworld-dev/opennox: OpenNox main repository., an extension of NOX.
The key driver behind cloud computing is scalability. Applications, especially internet ones, have variable demand at different times. Horizontal scaling is running applications across many machines instead of one powerful one. It can have potential cost savings on hardware and energy consumption.
We now apply that to data and create a scalable datastore. There are so many available:
Guarantees given my most traditional database systems are denoted ACID:
The NoSQL movement insists on much more limited datastore functionality called BASE:
Implementing BASE favours availability (and often low write latency) over consistency.
An ecommerce site may deploy BASE on the front-end for speed and ACID on the backend for consistency. Think the difference between adding to your shopping cart (BASE) vs billing information (ACID).
So, BASE vs. ACID. It is impossible to build a database system that is CAP:
You must discard one attribute.
A key/value store is a system built on top of a massive distributed hash-table(s). A technique called consistent hashing allows for nodes to enter and leave a peer-to-peer content lookup network with minimal overhead.
Google uses technology called the Chubby System, which is a highly fault tolerant database. The contents of the database are replicated over several servers. So if the majority of the servers are running, the database can serve requests. The design is based on the classic Paxos Algorithm, which is expensive.
Scalable file storage is also important, like for Amazon S3 system. Google has near identical competing product called Google Storage. Cost saving techniques include:
Asynchronous batch processing of data needs to horizontally scale as well. Contenders are Google MapReduce and Apache Hadoop, the opensource alternative. Basically, it works like:
Probably not the best explanation of the process.
Hadoop is written in Java. Popular extension is called Hive, providing an SQL-style like query language for the Hadoop engine. HBase is another Hadoop like database by Apache.
Then there is the Nexus Framework for running multiple other frameworks in the same cluster.
Scalable computing infrastructure on the cloud must be fault tolerant and self-healing. Key techniques to achieve this are:
Latency is fixed by the speed of light. 3 classic solutions to latency are:
Infrastructure as a Service is about providing a hosted computing infrastructure. A typical way to implement is platform virtualization. Amazon Elastic Computer Cloud, or EC2, is prime example. There’s also the Eucalyptus, an open source way to make your won clouds, and Ubuntu Enterprise Cloud.
Amazon Elastic Block Storage (EBS) offers persistent storage for Amazon EC2 instances. The also offer the Virtual Private Cloud (VPC).
Eucalyptus sounds cool because you can build your own cloud with a cluster of computers.
Platform as a Service provides a computing platform that contains a complete solution stack, hosted as a service. It’s not as flexible as IaaS. Examples include Google AppEngine, Microsoft Azure, and Heroku. Azure is a platform because you cannot choose the underlying OS or runtime.
Heroku was mainly a platform for the Ruby programming language, but I hosted a Django website there back when they had a free tier.
Software as a Service is a complete cloud hosted software system ready for use and offered to users over the internet.
Usually the user pays the provider a subscription fee for the service. This is different from the traditional way of purchasing a license. Examples include Salesforce and Facebook.
p. 43
To be continued…