Application availability depends on the availability of other elements in a system, for example, network, server, operating system and so on, which support the application. Concentrating solely on the availability of any one block will not produce optimum availability of the application for the end user.
In the following diagram, a "linear" or "non-redundant" configuration of elements supporting the user application is shown. In vendor engineering publications, these elements are referred to as "blocks," although other publications may refer to them as "components."
It is evident that if any block in the series configuration fails, the user loses the use of the application. The application is as available as the weakest link in the chain, or so it would appear.
The following figure is a schematic showing a linear chain of blocks, henceforth known as non-redundant blocks or blocks in series. It is easy to see that the failure of any block in the chain will cause a failure of the service to the end user.
There are, of course, other blocks in the chain, such as operating system, middleware and so on (not shown here) but the principle is the same. A single component whose failure causes overall failure is called a single point of failure or SpoF.
The equations above demonstrate that a series of blocks is weaker than its weakest component simply because of the multiplication of several factors, all of which are less than or equal to 1.
Lusser's Law is a prediction of reliability named after Robert Lusser (He worked on Wernher von Braun's US rocketry program post-WW2) . It states that the reliability of a series system (of our "blocks") is equal to the product of the reliability of its component subsystems, if their failure modes are known to be statistically independent. This is what we see in the above diagram. The law can be stated as follows:
This lays to rest the theory that a chain is as strong as its weakest link, the thinking at the time. Lusser's Law deals with the reality of this situation.
Next in this document we will discuss using components in parallel and how to make the assessment of availability more IT-specific and not just deal with anonymous blocks or components which might represent anything in a reliability context - valves, pipes etc.
Part of availability management is the examination of the service failure points in the configuration between application and user, assess their impact on service availability and design round them. Obviously there are cost implications to going over the top in design, especially in the cases below.
Effect of Redundant Blocks on Availability
The discussion so far has dealt with linear chains of blocks (blocks in series, to use an electrical analogy), where the whole chain is weaker than its weakest link – Lusser's Law.
To carry this analogy further, it is possible to use blocks in parallel to increase the availability of the chain, assuming that one block can take over from a failed block, assuming that the blocks fail independently. The following diagram illustrates this for two blocks:
These blocks might be NICs, disks, server or other parts of a working system. The general case for 'n' parallel blocks is shown below, together with the equations for availability and non-availability probabilities.
Parallel (Redundant) Components
The next figure illustrates components configured in parallel as opposed to in series as we have seen already. The mathematics of these configurations is similar to Lusser's mathematics except we deal with 'unreliability' instead of 'reliability' entities in the math.
The basic premise in these calculations is that if the probability of being available is P, then the probability of not being available is N where;
N = (1-P) and P = (1-N)
since the total availability of being available or not available is 1. In reality, a system will consist of several sets of redundant components, for example disks, servers, network card, lines and so on. These will feed into each other, possibly mixed with single components.
The figure below shows the general case of "n" blocks in a parallel configuration. This might represent one set of components for a subsystem such as a RAID configuration or set of network interface cards (NICs). Such a configuration can be difficult to handle mathematically so a 'reduction' technique is usually employed.
Two Parallel Blocks: Example
Picture two components in parallel, one with availability probability Pa and the other Pb. The probability of both blocks being unavailable, that is, the chain is broken, is:
This assumes the blocks have different availability characteristics, Pa and Pb..If they were the same, say Pa = Pb = P, then the probability that both are not available is given by the relationship:
which is essentially a variation of Lusser's Law using the non-availability probabilities as multipliers instead of availability probabilities. The probability that 'n' redundant blocks are unavailable is (1 - P)n and the probability that they are all available is given by the relationship [1- (1 - P)n ]
As an example, consider two parallel blocks, each with an availability of 99.5 %. The probability that both are unavailable is:
N = (1 - 0.995) x (1 - 0.9995) = 0.000025
Hence its availability (compared with the availability of a single non-redundant case of 99.5%) is:
A(%)=(1-N) x 100=(1-0.000025) x 100%=99.999975%
The knowledge of each value of "P" and some mathematical skills would be needed to solve the problem of service availability for a combination of serial and parallel service blocks, which is often the case in real life. The book High Availability IT Services covers this latter case.
Business leaders are in the unique position of having immediate access to huge amounts of data in today's smartphone and laptop-dominated world. They are also under pressure to make data-driven decisions and mobile business intelligence can one of the most valuable decision making tools in their arsenal ...
Unlike some AI initiatives, AIOps doesn't always necessitate the use of a data scientist, so don't let hiring expenses put your AIOps initiatives on hold. It is always nice to have IT team members with AI skills, but this becomes less critical as more intelligent solutions come into prominence that offer AIOps features out of the box, a readily deployable option for IT ...
AIOps is rapidly becoming a de-facto option for enterprises' IT strategies, with nearly immeasurable benefits to be provided. However, AIOps is still a relatively new discipline and misconceptions surrounding the technology's capabilities and uses have caused bottlenecks and roadblocks in its widespread adoption. So, what should organizations expect from AIOps? How can organizations that want to digitally transform their IT pursue AIOps for maximum benefit? ...
In response to the global pandemic, companies have given their workforce the tools they need to work remote. And research shows it has increased their engagement and productivity. But these gains are on the brink of being wiped out. According to a new study from Citrix Systems, Inc., employees feel they've been given too many tools and not enough efficient ways to execute. And it's hindering their ability to get things done ...
The third installment of Aptum's four-part Cloud Impact Study, A Bright Forecast on Cloud, presents data showing the benefits organizations gain from cloud computing, as well as mistakes to avoid during migration. As organizations migrate workloads to different cloud platforms, they often run into unexpected challenges due to a lack of proactive planning. Here are a few key findings from Part 3 of the Cloud Impact Study ...
Currently, (and most likely well into the future) the overwhelming majority of organizations still need to monitor and maintain enterprise applications. Moreover, where these are complex systems developed, debugged and refined over years, often decades, around a business's core processes, there can also be very strong practical arguments for viewing them as classics. They can offer a valuable legacy, one best left where it is, doing what it does, how it always has done ...
Anti-patterns involve realizing a problem and implementing a non-optimal solution that is broadly embraced as the go-to method for solving that problem. This solution sounds good in theory, but for one reason or another it is not the best means of solving the problem. Anti-patterns are common across IT as well, especially around application monitoring and observability. One that is particularly prevalent is in response to the increasing complexity of cloud-native infrastructure and applications ...
The introduction of the latest technology — such as AI and machine learning — can be seen as a way for organizations to accelerate growth, increase efficiency, and improve customer service. However, the truth is that the technology alone will do little to deliver on these business outcomes. AI for IT operations (AIOps) is one area where the application of technology, if not matched with organizational maturity readiness, will fail to deliver all the promised benefits ...
SREs that fail to deliver customer value run the risk of being stuck in an operational toil rut. Conversely, businesses failing to recognize the bi-modal nature and importance of SRE activities run the risk of losing talented employees and their competitive edge ...
As part of digital transformation initiatives, IT teams are quickly adopting AIOps solutions to accommodate a new multifaceted infrastructure. However, there are still several roadblocks IT leaders must overcome when adopting AIOps — namely, understanding how to showcase ROI and changing their team's cultural mindset around adopting a new strategy ...