I Stand Corrected – WekaIO is #1 on the IO-500 10 Node Challenge Ahead of IBM Spectrum Scale
Barbara Murphy. February 3, 2019
Let me be up front, this blog post is a retraction of my 2018 post titled “IBM Spectrum Scale – 12x the memory, 3x the NVMe drives, 5% faster”.
I was commenting on our performance on the initial posting of the IO-500 10 node challenge where WekaIO Matrix™ placed within 5% of IBM Spectrum Scale, on the world’s largest supercomputer.
Since that time, the IO-500 has identified an error in the way it calculates mdtest, a benchmark test that measures file system metadata performance. As a result, WekaIO’s prior submission has moved from #2 to #1 position ahead of IBM Spectrum Scale.
In the latest ten node challenge — a benchmark that measures how well the storage system can satisfy a small cluster of 10 compute clients — WekaIO Matrix is now 31% faster than the world’s largest supercomputer at the Oak Ridge National Laboratory (ORNL) utilizing IBM Spectrum Scale. Why is this test so important? Well, because the biggest challenge that high-performance compute (HPC) clusters suffer from is poor I/O performance leading to I/O starvation to the applications. Anyone still using NFS to try and do HPC workloads knows this.
These results validate our claim that Matrix™ is the world’s fastest file system, making the best file system choice for artificial intelligence (AI), analytics, and HPC workloads. It’s well-known, that if you throw expensive hardware at any storage problem it will eventually improve performance. But what is truly compelling about Matrix is that this benchmark was run on less than half a rack of equipment. Meanwhile, the IBM ESS Storage system takes up 40 racks of space at ORNL. If there was a metric for storage efficiency, we would have blown the socks off the benchmark.
If you look at the infrastructure used to drive performance, it is clear that WekaIO Matrix is the performance leader—offering the best performance per U of anything available on the market today. The IBM system is comprised of 154 storage nodes compared to WekaIO’s 32. The IBM system was populated with 1 Terabyte of memory per storage server compared to just 192GBytes of memory for WekaIO. In addition, each IBM storage node had 8 NVMe drives to boost small file I/O performance, while WekaIO had only 6 NVMe drives per node.
In summary, it took 40 racks of hardware, 12x more memory and 3.2x more NVMe for IBM Spectrum Scale to deliver 31% less performance than Matrix accomplished in less than half a rack.
If you look at the performance score per storage node, Matrix is more than double the performance of Spectrum Scale.
|IBM Spectrum Scale||WekaIO Matrix||Difference (%)|
|Number of Nodes||
Total Memory (GB)
|Number of NVMe drives||
|Rack Space (RU)||
To see the specifications of the IBM Summit Storage System, click on the link below.
AI AND HIGH VELOCITY ANALYTICS NEED A NEW FILE SYSTEM
Lectus arcu bibendum at varius vel pharetra vel. In cursus turpis massa tincidunt.