The Stackable Efficiency of an ARM-Based Object Storage Building Block
Enrico Signoretti
Enthusiast, curious, forward thinker, always looking at new ideas to improve with my team and personally.
A couple of weeks ago, in a moment of excitement, somebody tweeted about a small project we are working on in our lab. This piqued some interest, and many people asked me about it.
Here is the deal: a new appliance will be soon available through OpenIO’s and Ambedded's channels. It is an object storage building block (50TB usable) with incredibly low $/GB, high efficiency, and it is very easy to use.
ARM-Based Storage
You probably already know that OpenIO SDS runs on ARM. Not just the Raspberry Pi (which is nice, but is only a proof of concept, and as a good way to start playing with object storage at home), but also in the datacenter.
More than a year ago, we launched the SLS4U-96. This is a chassis capable of hosting up to 96 nano-nodes. (1152TB in 4U, before data protection). It's aimed at special projects, where large capacity, rack density, and power consumption are key. We do not sell it directly; our partner 2CRSI does all the system integration and supports the hardware.
We have had some interesting successes with it, but it is a very large building block. In fact, to make a 96-slot chassis cost effective you have to sell it full of disks. And a building block that is around 1PB is not for everybody.
Ambedded Mars 200
Our partner Ambedded Technology, based in Taiwan, has been developing a similar device for years now; but it is smaller, much smaller!
Mars 200 is a 1RU chassis which incorporates 8 ARM nodes (very similar to our nano-node: 2 ARM cores, 2GB of RAM, 2*2.5Gb/s Eth, and 32GB SSD M2). These nodes, called micro-servers, are connected to two separate 10Gbit/s switches with 10Gbit/s uplinks. The Mars 200 is less dense than an SLS (32 HDDs in 4U total), but it is also shorter, and is suitable for most standard depth racks. All these characteristics make it perfect for small installations, enterprise customers, and local ISPs.
A Cost-Effective Solution
We tested the hardware, which works perfectly with OpenIO SDS, and we are now working on the integration part making it easy to install. Some customers are already testing it and they are excited about its prospects.
Let's take a quick look at the economics. The cost will be around 0.008 €/GB per month with 50TB of usable capacity (including a 3-year subscription to OpenIO SDS and a hardware warranty). I think this is pretty impressive, especially when you compare this with other object storage solutions on the market.
Yes, this won't include rack and power consumption, but it is a fixed price, and a very low entry point!
The Benefits of the Nano-Node (or Micro-Server)
It is always good to have low TCA (Total Cost of Acquisition) but let's take a look at the TCO and the potential ROI of this solution.
- Failure domain. Even with a single Mars 200, which has no SPOF, the failure domain is limited to one node (one disk).
- Power consumption. ARM nodes are very low power. Mars 200, including disks, is less than 110 watts.
- Scalability. This solution is for everybody. You can start as small as 50TB and grow up to PBs just by stacking more Mars 200 chassis.
- OpenIO SDS. All of this is all possible because of SDS. It is true that you could do it with other software, but OpenIO SDS was designed from the beginning to be lightweight and highly efficient. Conscience technology for dynamic load balancing, no rebalance needed when expanding the cluster, storage pools, and dynamic load balancing all contribute to make this solution highly sustainable over time.
Takeaways
One of the biggest hurdles to starting out with object storage is the initial cost. Not because it doesn't work, but just because you need at least 3 nodes to start a cluster, and with 3 x86 nodes you can't have erasure coding to improve data footprint efficiency. And this doesn’t include the cost of cluster expansions. The ARM-based disk makes it possible to start with erasure coding in a single rack unit without sacrificing any characteristic of object storage, including the low $/GB.
OpenIO SDS on ARM is good for everybody, and it extends the benefits of object storage to a larger number of customers. For example, local ISPs can start small and grow with the business while enterprises can easily begin porting a single application and consolidate others when necessary. The small building block is perfect to dilute the investment over time, with a small impact on overall IT budget, and a cost that is very low from the beginning.
Are you interested in learning more about OpenIO SDS on Ambedded Mars 200? We will be hosting a webinar on May 24 (2 sessions 11AM CET and 11AM ET). Join us to learn more.