When we think of scaling our infrastructure within our data centers a couple different models come to mind – we can scale out, which is essentially adding more servers or nodes into our applications cluster or infrastructure – or we can scale up which involves adding more resources into our already existing servers to support our applications running within them.  But whats with this “Scale In” business?

First, let’s look at the pros and cons of “Scale Out” and “Scale Up”

Scaling-up tends to be a little easier on the licensing side of things and helps with the cooling/power bills for our data centers, however it does impose a bit of a greater risk of hardware failure and tends to come at a cost once we start to hit maximums that a server can hold.

Scaling out, while providing nice upgrade paths and generally giving us an “unlimited” amount of availability leaves a much bigger footprint within our data centers, resulting in higher cooling/power bills and possibly more dollar signs when it comes to license and maintenance renewals.

For our average everyday workloads whether we scale out or up may not have that great of effect on our bottom line – but what about this whole new trend of machine learning and big data analytics.  These types of processes require an extremely high number of resources to process data.  Scaling out and up in these situations certainly has a huge effect on our data center bills – and often, still doesn’t provide us with the data locality and performance we need as it requires us to rely too much on our networks, which in turn, eventually need to scale up to support more data flow – so how do we over come this?

X-IO Technologies may have the answer!

At SFD13 in Denver this June X-IO Technologies invited us into their offices to see what they have been up to for the past little while.  In fact, it had been three years since X-IO last participated in a tech field day event and a lot has changed since then!   At SFD5 X-IO talked about their flagship ISE technology – a general purpose storage array targeted at the mid-market enterprise with the normal features such as performance, availability etc.  Fast forward to today and their story has completely changed – While still providing support for their older product lines, ISE and iglu, X-IO has shifted R&D resources and pivoted into the big-data market with their Axellio Edge solution – a converged storage and server appliance leveraging a lot of power and a ton of NVMe storage in the back end – their own “Scale In” solution!

Hello Axellio

Before delving into exactly how this Axellio Scale In solution performs let’s first take a look at what everyone is interested in – the hardware specs!  Axellio is a converged appliance – meaning it takes both compute, memory, and storage and combines them into one 2U rack mounted appliance.

As far as the compute and memory goes, Acellio contains 2 nodes, each node supporting up to 44 CPU cores and 1TB of RAM – so yeah, do the math – we basically have 88 cores and 2TB of memory to work with here.

That said the biggest benefit to the Axellio in my opinion is the storage back end – Axellio’s back-plane supports up to 72 dual ported NVMe SSD drives.  Currently that brings Axellio’s capacity maximum to 460TB with 6.4GB NVMe drives – in the future, with larger drives, we are looking at a whopping 1 Petabyte of storage – all on bus speed with NVMe performance – think greater that 12 million IOPs with 35 microseconds of latency!

So….back to scaling in.

To help explain their scaling in concept let’s take a look at an actual logical diagram of the Axellio platform.  As we can see Axellio doesn’t appear to function the same as our traditional converged and hyperconverged appliances we see today – the storage, in essence, isn’t distributed – meaning the server nodes do not have their own local storage that they pool together to present globally to a cluster – nor are they addressable by any sort of global namespace.  Although the FabricXpress functionality does allow for inter-node communication to support things like memory mapping back and forth from the nodes – they are essentially, two distinct server nodes.


What we have here, is basically two separate and distinct compute nodes, connecting to the same FabricXpress back-plane and basically both accessing the shared NVMe storage!  As you can start to imagine this is where we see the “scale in” concept come into play – we have the scale out advantages of having two nodes, while also combining the scale up benefits of having a lot of cores and memory – all backed by the blazing speeds of NVMe on the back end!

But the magic is in the software right?

Of course – software rules the world today – but Axellio isn’t providing you with any!  X-IO play with Axellio isn’t to sell you something to run your VMs on, or something in which you can simply pipe your data into some X-IO built analytics engine – this isn’t a general purpose server!  Axellio is basically an OEM box – a box that is targeted at companies and enterprises that need a mass amount of computing and storage performance requirements in order to solve specific problems.  Think things like streaming analytics or in memory big data applications.  In the end, it’s the customer that is left with the choice of how they want to leverage the Axellio platform – meaning they put the OS on the compute nodes, they determine if they want RAID or any other forms of availability with the storage, they decide whether to use each server node independently or to setup some form of HA between the two – the customer is in full control!

One interesting use case they had was an analytics engine where one server node takes on the role of writing the streaming data to the drives, while the other server node provides the compute and access to any real-time analytics that may need to be accessed!  Now – while this use case can be handled many different ways – Axellio does it at very high speeds, very low latency – oh, yeah – and within 2U of rack space!

So in the end I think X-IO technologies is on to something with Axellio – and honestly, it appears to me like they are still “learning” about how they plan to bring this to market!  Currently, they are focusing on providing a hardware platform to a somewhat niche group of players and looking to solve very specific use-cases and problems – a big change from directing all their efforts into the storage array market which is flooded with general purpose vendors.   And rightly so – they need to explore this area and get more and more data and use cases before going down any other roads with Axellio.  Where those roads may lead them is yet to be determined but in my opinion I can see one or two things happening with Axellio – it moves towards a reference architecture model – meaning we get in-depth documentation on how to do things like Hadoop or large scale Splunk deployments with Axellio – or maybe, just maybe X-IO technologies have something in the works in terms of their own software development that they can layer on top of Axellio!

If you want to learn more about X-IO Technologies and Axellio certainly check out their website here.  You can also find their SFD13 recorded presentations here – If you want to get really nerdy I’d suggest watching Richard Lary talk about dedup and math!  And of course don’t forget to check out the posts from fellow delegates Brandon Graves, Dan Frith, and Ray Lucchesi as well!  Thanks for reading!