Scale Raptor: The All-Flash Datacentre

Deep down in the dungeons of the Data Resilience LAB’s a few quite interesting activities has been taking place over the last few months. The Scale HC3 platform as shipped is already revolutionary in that it is the simplest most affordable way to set up a complete and resilient virtualization infrastructure for any enterprise big or small. Scale HC3 is based on traditional storage i.e. spinning disks, either SAS or SATA. Until now.

A Splash into Flash.

In a Myth Buster like fashion the white robes in our LAB set out to find if it is indeed possible to do a data-in-place migration from the existing 16x 500GB SATA based HC3 and transform it into a pure Flash based system without any downtime for the running VM’s in the infrastructure. Could it be done?

If this worked it truly would be an unprecedented event. No one else can do this with out adding a completely new SAN storage device doubling the capacity during the transformation, with many hours of work setting the target system up and migrate the data from A to B. With Scale, in theory, this should work without adding anything. The curiosity for this digital heart transplant was simply to hard to resist, so off they went and as a result we got a 10 times faster virtualization platform.

Meet the Raptor.

We where able do this without introducing any additional hardware.  Nor did we disturb the VM’s (32 of them all 2Core/2GB/100GB machines) humming away up in our private cloud. Needless to say please note that this is a LAB environment, You should never attempt this in a production environment without full support from Scale Corporate. This White Robe LAB is merely to prove the point of possibility, a vision of the future. We set out to prove that the All Flash Data Centre In-A-Box is possible. With Scale.

Scale GUI 32xLAMP running on four nodes

The infrastructure we use is based on a fully redundant 10GbE Optical network (2x NetGear XSM7224S 24 port 10GbE switches) for both the front and and backplane of the Scale nodes. During the LAB we have replaced the 16x 500GB SATA drives, one at a time, with Samsung EVO 840 Consumer grade SSD’s. These drives are not recommended to be used in a production environment but again works perfectly fine for the scope and purpose of our Scale Raptor LAB. Once all the drives where replaced the whole whole system turned out to be viciously fast. This ladies and gentlemen is a screamer!

Fill her up.

As we enter the realm of the SSD’s, there are a few things we need to keep an eye on. SSD or Flash drives behave differently than traditional spinning disk drives in the way they age.

They work more like a tank of gas in your car: keep an eye on the needle and fill it up (replace the SSD) before you run completely dry. We also appreciate the fact there is no notion of SSD’s in our contemporary Scale HC3 system at this point, therefor the LAB people created a SSD wearing and reallocation monitor to check up on the deterioration of the drives every ten minutes and feed the data into a mySQL database (also running as a VM on the Scale Raptor). An example output the database scheme is shown below.

DC1 Flash Wearing 28-Apr-14

Hatching The Raptor.

In the evening of 1st of March 2014 the last SATA drive was replaced and the Scale Raptor got successfully hatched. To my knowledge this was the worlds first “All-Flash Datacentre In-A-Box”. We think this is way cool. But then again we are complete nerds here at Data Resilience. You have to be a nerd to get excited about the fact that after replacing the SATA with SSD the power consumption dropped by 25% for the 4node Scale cluster including the NetGear switch infrastructure; from about 4A to 3A: meaning about 21 Watts per VM including the complete infrastructure. Flash is the future of storage.

The picture below shows the Scale Raptor with the old drives on top of the cabinet and the Flash in place in the cluster. The combination of hyperconvergence as offered by Scale HC3 and the speed of Flash just becomes absolutely impossible to resist. This is just great.

Scale Raptor Hatched

And now…

Next time we will talk about the tests planned for the All-Flash Datacentre In-A-Box.

The setups are as follows:

– 32 VM’s running WordPress, one “site” on a separate TurnKey Linux LAMP machine.

– One external physical 16 Core/48GB payload server running multiple instances of Apache Bench (ab) setup to pound the WordPress sites across the 10GbE network.

– Meanwhile we are monitoring the wearing, the throughput and the limits, e.g. identifying at what point the sites are starting to show signs of limitations from a web service vantage point.

Till next time. ROAR!!!