For about the last 6 months I’ve been working on a long term evaluation of Scale Computing’s HC3 platform.  I was fortunate to be given a cluster of three nodes to use as I choose and today this system forms the main part of my lab environment.  For those not familiar, HC3 is a scale-out hyper-converged solution based on open source KVM and intellectual property developed by Scale, including their SCRIBE distributed storage layer and a state engine to manage all of the hardware components.

HC3 is targeted at the SMB market and so could appear to be light on features compared to the VMware ecosystem, however don’t let this fool you – HC3 is a robust solution that offers remarkably simple ease of use for environments where the IT staff are usually generalists rather than specialists in virtualisation.

Architecture

As mentioned, the HC3 architecture is a scale-out node-based solution that today comes in three family models; HC1000, HC2000 and HC4000.  Scale recently announced the availability of flash in two new models – the HC2150 and HC4150.  For more information on product specifications check out the products page on the Scale website (link).  The introduction of flash into the HC3 family is an interesting one, as up to this point all of the storage capacity has been delivered using traditional spinning media (both SATA and SAS, depending on the model type).  As persistent storage is distributed across all HC3 nodes, the overhead of leaving the server, writing to and reading from external storage is reduced, one of the classic benefits of the hyper-converged architecture.  However there’s only so far HDDs can go in I/O performance and as the baseline storage nodes scale up in processor performance and memory capacity, so storage performance can become a bottleneck.  Now flash in the form of SSDs has been added to HC3, creating a hybrid model.  The flash is used to create a dynamic storage tier rather than as a cache, which means the flash capacity adds to the storage capacity of the overall system.  In the initial release, SLC SSDs are being used, but over time this is expected to change with cheaper MLC drives being introduced.

Accelerating Performance

ScaleFlashObviously flash provides a much better I/O density (IOPS per TB of storage) than standard hard drives.  The key though is in using flash effectively.  There are two aspects to this; first is the ability to target flash effectively to the right level of active data.  Second is the ability to allow the end user to set policies on identifying VMs that should be able to use flash.  The first step is achieved through SCRIBE, the underlying distributed block data management engine.  This sees all I/O activity and is able to build up a heat map of activity that allows blocks (at 1MB granularity) to be dynamically moved between flash and disk.

The second step is how end users can assign prioritisation to their workloads.  This is done through a new setting on each of the virtual disks assigned to VMs and uses a numeric scale to prioritise the amount of data for a disk that should be assigned to flash.  The scale ranges from 0 to 11 (o means no flash allocation, 11 means all data in flash) and each of the gradations on the scale is logarithmic (based on power of 2).  Obviously VMs and disks with higher prioritisation will have more of their data placed in the flash tier.

The idea of a 0-11 scale might seem a little simplistic for some enterprises, but we should remember the market that Scale is targeting and that’s the SMB/SME customer with small teams of IT staff.  The introduction of flash provides the ability to improve workload performance but do it in such a way that doesn’t require lots of detailed analysis or effort.  Simply turn up the dial on the VMs that need to go “faster”.

Gaps

This release is the first step into flash for Scale Computing and naturally there are some gaps in the implementation.  For example, only SLC drives are being used initially for resiliency as the algorithms for managing workload profiles are further developed.  Over time, I’m sure we will see MLC devices introduced as the workload profiles are more well understood and optimised.  There is as yet no all-flash model.  It’s perhaps a little premature to expect this, bearing in mind the target market, however with TLC capacities increasing, all-flash systems will become cost effective fairly quickly.  Bear in mind also that the way in which SCRIBE is implemented allows for many tiers of storage to be used across multiple node types.  This makes it very easy to take advantage of a multi-flash tiered environment as drives become available (drives can also be swapped out over time, so it’s possible to migrate/upgrade storage in place too).

The Architect’s View

I’ve been following Scale Computing for some time and have been enthusiastic about their HC3 platform since the first presentation I saw at Storage Field Day 5 in April 2014.  In terms of market presence, the company hasn’t taken the glamour route of Nutanix or Simplivity (with lots of marketing), but has focused on building a solution with no vendor dependencies (such as VMware vSphere).

What is starting to be interesting is where Scale are headed in terms of their future market presence.  The performance and capacity of nodes continues to increase, pushing HC3 further into mid-market territory.  However there are some features which aren’t yet available (like CLI and REST-API management) that could help to accelerate that adoption.  Part of the “problem” here (if it can be described as that) is the desire for Scale to be able to provide all of their customers the full level of support they currently experience today.  In this part of the market, support can be a significant drain on resources and needs to be handled correctly.

In classic “Innovator’s Dilemma” mode, I can see Scale continuing to increase the attractiveness of their solution to a wider audience, including Dev/Test environments and private cloud deployments.  Today Scale claims 1500 customers worldwide with 5500 systems deployed.  That’s a significant install base with which to learn and move up to larger customers.

Look out for more posts on the Scale platform as I delve into some of the feature specifics over the coming months.

Have you looked at HC3?  Do you already run HC3 in production?  Let us know your thoughts on how the hyper-converged market is developing.

Further Reading

You can see more details on Scale Computing’s HC3 platform, including details on SCRIBE by watching the Tech Field Day presentations listed here.

 

Comments are always welcome; please read our Comments Policy first.  If you have any related links of interest, please feel free to add them as a comment for consideration.  

Copyright (c) 2009-2016 – Chris M Evans, first published on http://blog.architecting.it, do not reproduce without permission.

 

 

We share because we care!Share on Facebook0Share on Google+0Tweet about this on TwitterShare on LinkedIn0Buffer this pageEmail this to someoneShare on Reddit0Share on StumbleUpon0

Written by Chris Evans

  • John

    Where are they finding SLC drives? Generally I see people use 10DWPD eMLC (like Intel S37xx series etc).