Richard Parker's blog

I decided to write this post to provide a little technical information aimed at non-programmers (Project Managers, Department Heads and other Busy People) who want to know more about the platform; how it works and what it offers. My goal is that, after reading this article, you’ll have a basic – yet thorough – understanding of how Azure is structured so that you can make informed contributions to discussions regarding the platform. This is a work in progress.

Some of the analogies used in the following article are designed to facilitate understanding on a functional level, and may therefore be technically ‘inaccurate’. If you’ve picked that up, you’re probably more technical than this author had in mind as the intended audience!

As always, we’re all learning – if you have ideas or suggestions for improving this article, please feel free to leave a comment. Thanks!

Table of Contents

1. Introduction
2. Web Roles and Worker Roles
3. Resources
4. Storage
   • Blocks and Pages
   • Addressing blobs
5. Databases

An introduction to Windows Azure (for Busy People)

In the Azure world, you can have databases and applications all running in the cloud environment. By now, most of us know that a ‘cloud environment’ in its most basic form describes an environment in which you don’t ever see or touch the physical hardware or infrastructure as these are determined, managed and provided for you by the cloud service provider.

Developing and deploying applications onto the Azure platform requires a different approach to traditional application development, but developers can still continue to use all their existing tools (such as Visual Studio 2010) and don’t require any new software to get started. In fact, it’s actually possible to write applications for the Azure platform using the free Expression edition products provided by Microsoft.

Physically coding your applications, however, does require developers to change the way in which they build their applications, if only a little. That’s really a topic best left for someone else, or another post, to address.

On Azure, applications are referred to as ‘roles’, and there are two types of role: a “web role” or a “worker role”.

Think of a web role as a web site¹, and a worker role as some repetitive computational task that takes place behind-the-scenes without any user interface at all (a good example would be processing statistical data, or – to use examples from other blogs – a thumbnail generator for images).

Roles

Web Roles are similar to web servers, in that they allow public computers to connect to your application over standard HTTP and HTTPS ports. Typical Azure deployments consist of one – maybe two – web roles, and a number of worker roles. Worker roles are also publically accessible; that is they can talk to each other and the outside world, and other Azure services.

It is important to note, however, that one web role is not actually a web server in and of itself. It is simply an instance of your software running on a web server that is publically accessible.

Azure would not be complete without two other key service offerings: storage (some place to store all your data) and SQL Azure (a variation of SQL Server, which provides relational database capabilities to your cloud applications deployed on the Azure platform).

To recap then, Azure is a platform that provides:

• Some place to run your applications from (via web and worker roles)
• Some place to store all your application files
• SQL Azure – a relational database like SQL Server

Each of these functional areas are referred to as ‘hosted services’, and as you might expect there are limitations imposed by Microsoft as to the amount of resources available to each service.

Resources 

Though theoretically unlimited, in order to ensure all customers have resources available when required, Azure packages and limits what resources are available to specific deployments. Databases, storage and application instances are artificially capped according to the current limits (published online ², updated regularly and these are commonly expected to grow over time).

Web and worker roles come in four varieties: small, medium, large and extra-large. That’s because they are actually virtual machines (VM’s – software ‘simulations’ of physical servers, many copies of which can run on a single physical server). Each represents an increase in pricing and has a different set of specifications that govern how much RAM, local storage space and CPU cores are available to the role as described below:

Each VM is provisioned when required. The ‘magic’ of Windows Azure is that when you provision a VM, the Azure platform actually provisions a further two identically configured VMs. One acts as a recovery image, the other as a failover. If Azure detects a fault condition, it takes appropriate steps to automatically recover the damaged VM.

This is one of the most useful features of Azure, and you get it for ‘free’ – i.e., you don’t need to do anything particularly special to get this to happen, it’s simply a by-product of deploying your applications on to Azure.

Getting to Azure

To utilise Azure, you need an Azure services account (one per customer). Each account has the following overall limitations:

• Maximum 20 hosted service projects (projects contain instances)
• Maximum 5 storage accounts
• Limitation of 5 roles per hosted service project (i.e. 3 different web roles and two different worker roles, or any such combination)
• 20 CPU cores across all of the hosted service projects

Configurations of the Azure platform represent significant architectural decisions as deployments not only require the correct determination of ‘size’ but also the appropriate number of ‘instances’ of that deployment which will concurrently run. It is possible, therefore, to have two instances of a ‘small’ worker role running the same application. This would consume two of your maximum 20 cores. It is worth mentioning at this point that one can, at any time, reconfigure a deployed instance to utilise a larger VM or have a higher instance count, but that some (relatively minor) downtime would be incurred.

Storage

Storage in the cloud doesn’t work like any traditional disk-based system. That is, you’ll never have a “C:\” drive or a “D:\” drive³ (local storage is a topic I’m not going to cover here). The Azure platform makes disk space available as three distinct entities: Blobs (block and page), Tables and Queues. These three entities essentially abstract space on physical disks away into different logical units, within which programmers will never be able to ‘see’ the underlying disks or access them directly. This looks a little something like this:



Blobs are stored within containers and you can have as many containers as you can fit within your storage account quota. They’re a bit like folders, but only if you consider that you get to name them once they are created, and they cannot contain subfolders (or sub-containers, for that matter). Azure tables aren’t like tables in relational databases such as SQL Server or Microsoft Access, while queues provide a mechanism through which web and worker roles can talk to each other (instance A sends a message to instance B, which might – but doesn’t have to – process the message right away, hence why it is called a queue).

Block blobs and Page blobs

Block blobs are optimised for streaming, while Page blobs are optimised for random read/write operations. Block blobs are targeted towards streaming operations specifically because writing them is a two step process: first, you upload all of the individual blocks that will comprise the blob. Next, you must commit the blocks via a block list. During the commit phase, you can add/change or remove blocks from the blob. Page blobs, on the other hand, are updated immediately – no commit phase is required.Page blobs consist of an array of pages, where each page is 512 bytes and the blob size must be a multiple of 512 bytes.

Both block and Page blobs can be read from any byte offset in the blob, meaning it’s possible to read only a specific ‘chunk’ of either blob when it is on Azure Storage.

Page blobs: primary characteristics

Page blobs are fast and range-based, which means you can read from and write to specific ranges of a blob at a time. Page blobs are initialised with a Maximum Size, but if only half the blob contains data, you’re only charged for what you actually store in the blob. Page blobs also support leasing, which means it is possible for your application to ‘lock’ a specific range of the page blob while it is updating it, then release the lock.

The Windows Azure Storage blog has this to say about Page Blobs:

Another use of Page Blobs is to use them for custom logging for their applications.  For example, for a given role instance, when the role starts up a Page Blob can be created for some MaxSize, which is the max amount of log space the role wants to use for a day.   The given role instance can then write its logs using up to 4MB range-based writes, where a header provides metadata for the size of the log entry, timestamp, etc.   When the Page Blob is filled up, then treat the Page Blob as a circular buffer and start writing from the beginning of the Page Blob, or create a new page blob, depending upon how the application wants to manage the log files (blobs).   With this type of approach you can have a different Page Blob for each role instance so that there is just a single writer to each page blob for logging.  Then to know where to start writing the logs on role failover the application can just create a new Page Blob if a role restarts, and GC the older Page Blobs after a given number of hours or days.  Since you are not charged for pages that are empty, it doesn’t matter if you don’t fill the page blob up.

Block blobs: characteristics

Block blobs consist of, well, blocks! I’d say, in my experience, most people would want to be using block blobs over page blobs because they’re a little more flexible in terms of their sizing. For instance, a block blob does not have to declare its size when you create it: you just keep adding blocks to the blob until you’re done. There’s another benefit, too. With block blobs, you can send blocks in any sequence, then arrange them later on when you call your commit function. This makes them ideally suited to transferring large files, where your client is able to use a few threads to send the file in chunks.

Understanding the limitations of block and page blobs

Storage, like the other Azure services, is also subject to some limitations (and its own pricing structure), and the current limits are described in Table 3 below:

You can mix and match block and page blobs within your account, but a block blob cannot suddenly ‘become’ a page blob, or vice versa. Once a blob is created as one particular type, it will always remain that type. A block blob cannot contain pages, and a page blob cannot contain blocks for instance.

Addressing blobs

Blobs aren’t accessed or written to like traditional file systems, with a nice path-to-folder-and-filename approach (e.g. “C:\My Documents\My File.jpg”). Blobs use URIs to organise their data, e.g.:

blobname/which/can/have/slashes/but/dont/represent/folders/file.jpg.

It is precisely because this system is URI-based that it can be so vast and resilient to failure, as there are many copies of each individual physical drive. Therefore, it’s safe to say that when you upload a file to Azure and store it in blob storage, it’s pretty safe!

Earlier, I explained that a blob should be thought of as a container for files. This is not strictly true, but the analogy is very similar. In actuality, blobs are containers for blocks (chunks of a single file) and pages (more on those below), and blobs are actually organised into containers themselves. One file may be one block (if it is under 4MB in size; the maximum size limit for a block), or it may be several thousand. If the file is over 64MB in size, it must be split into blocks. Azure, perhaps confusingly has two varieties of blob storage: block and page.

Let it suffice to say that block blobs can be no larger than 200GB, and page blobs can be no larger than 1TB. Any combination of the latter must not exceed 1 TB. You can therefore see that the storage system in Azure is much more complex than the traditional system we are used to, but that it offers significant advantages over the former.

Databases: SQL Azure

Microsoft has redesigned some of their core applications (such as SQL Server) to work specifically on the Azure platform, and as such, they have some very appealing advantages over the versions of the products that you can buy commercially.⁴

In typical server-based implementations of SQL Server, it is common to find one server acting as the master while the other one monitors it to take over should it fail (the slave). This means the database is subject to the limitations of that server (storage space, processing power and bandwidth). It also means that although you have two servers powered on and dedicated to the task of serving a database, only one is ever actually working at any one time, which represents half the total available computing power and is a good example of why paying for hardware through a traditional hosting company is actually a less appealing concept.

On Azure, SQL Server has become SQL Azure – and now, the concept of master/slaves has gone and you have multiple servers all serving the same database, resulting in massively higher processing power and greater throughput capacity. What this ultimately means is that one can work with that database much more quickly than one can with SQL Server.

Now, there are some fundamental differences between SQL Azure and SQL Server. For example, one cannot do everything one can with SQL Server within SQL Azure. Bear that in mind when your developers explain this to you, as the two products are not exactly the same.

Databases require somewhere to store their data. SQL Azure has the following database packages available:

In addition, data transfer charges apply to the standing monthly charge:

SQL Azure offers the opportunity to pay only for what one actually uses. The standing monthly charges are amortised over the month and you only pay for the days on which you actually have the databases in each specific tier. This makes it a very cost-effective way to purchase database space in the cloud.

Also, being based on the Azure platforms means that there are a number of additional advantages:

• Data stored in an automatic high-availability environment
• Fault tolerance included
• 99.9% “Monthly Availability” SLA ⁵

This concludes our basic high-level introduction to the Windows Azure platform and I hope you have enjoyed reading it. If you have questions, feel free to post them in the comments below and I’ll do my best to answer them.

Foot notes:

A web role does not have to be a web site – it could be a web service, such as an API. A web role is publically accessible via the World Wide Web.

Available at http://msdn.microsoft.com/en-us/library/ee814754.aspx

Service quotas are expected to grow over time and automatically become available to hosted services.

“Local storage” excepted; in this document I am discussing globally available storage.

Azure is a proprietary technology and no company can install their own private instance of it. Microsoft software written purely for Azure is not available to any third party to install and host on their own infrastructure.

See http://www.microsoft.com/windowsazure/sla/ for all the Azure platform SLAs