Implementing Converged Infrastructure

The first thing to do when examining the idea of using converged infrastructure in your datacenter is to ask, “Why would I do this?” Having a firm idea of what you want to achieve is key, because there are quite disparate alternatives available under the “converged” label.

Look at the benefits. First, buying a converged stack simplifies vendor interfacing by effectively creating a one-stop support system, at least for hardware. Next, past frustration with integration issues should go away, since pre-integration of all the component classes is a fundamental of the converged approach.

So far, so good, but this is where some divergence in approaches appear, depending on vendors. Software is sometimes part of the bundle, handling automation of the new cluster. This may reach to the point of delivering a complete private cloud solution, such as converged clusters with a version of Azure bundled.

In time, the software issue may be significant for many shops. Getting a pre-digested OpenStack cloud with everything guaranteed to run out of the box might allow many admins to sleep better at night. The same is true for SAP or Oracle clusters, for example.

Software is a critical question for converged systems. I’ll be frank, we are moving to an era totally dominated by COTS systems. The integration issues of the past are largely behind us and most COTS elements fit together like Legos. Though the system vendors point to integration as critical, reality is that the software that manages your new cluster is what will determine success or failure.

Most IT operations are today thinking beyond virtual clusters to a hybrid cloud setup, where system management such as adding or removing systems is largely automated and the allocation of work to virtual instance is orchestrated completely. That migration has to be in your planning as you select a vendor. You have to choose a supplier that can take you along the whole path to the cloud, because the last thing you want to do is have to change automation and all the processes and training already invested.

One problem with converged systems in general is that they lock you in to a vendor’s solution. This tends to become expensive over time, since you are effectively captive. This is especially true in software. Perhaps the best analogy from recent history is the transition from UNIX to Linux, where vendors had their own, proprietary, tweaks to UNIX, making it difficult to go elsewhere. This can be seen today in OpenStack, where the infrastructure management tools from many vendors smack of the same proprietary nature.

It is quite easy to build equivalent systems from low-cost gear such as so-called whiteboxes (which today are coming from the same top-notch ODM vendors that deliver millions of units to AWS and Google). This of course avoids the premiums charged for converged systems by traditional vendors, but it moves the responsibility for choosing gear back in-house and complicates support , at least until the ODMs establish a stronger support base in the US.

A good plan might be to start with a converged solution from a traditional vendor as a way to get started more quickly, but look hard at buying servers, storage and networks independently going forward. This avoids the issue that’s bound to arise when you transition from the strongly competitive environment of buying that first cluster to the single-sourced environment when lock-in holds sway.

This hybrid approach will work if there is an open automation suite. Such a toolset will allow mix and match of a variety of vendors’ products, expanding your ability to respond to real growth urges (much more storage, faster networking etc.) as independent paths for improvements in productivity, avoiding the one-size-fits-all restrictions that converged and especially hyper-converged solutions can bring.

One further consideration is the rapid expansion of software-defined infrastructure. This aims to virtualize control, removing it from devices, commoditizing them and so lowering their cost. Well within the life of any new installation, SDI will be a major way to improve agility from central IT all the way to the user.

The best hybrid plan thus needs a good, agnostic software-defined infrastructure solution such as StrataCloud’s that makes hardware platform mix and match easy.  This type of automation solution has adapters to interface new hardware families as they arrive and supports installation and integration, while adding a single-pane monitor for the cluster.

Converged and hyper-converged solutions may in the end be a blip in the road to inexpensive commodity-class COTS solutions that always fit together like Legos, with SDI acting as glue. For all of these good agnostic management is crucial to success.