Software-Defined-Storage architectures are pervasive in the datacenter, and they fully leverage the improvements in server-side microprocessor architectures. As with most software-defined-storage solutions, Datrium DVX gains performance not only with faster servers and processors, but also with every new software release.
Datrium is on par with any enterprise-grade Tier 1 storage solution, providing industrial-strength data resiliency, data reduction and scalability – up to 128 servers, 18 Million IOPS (4kB read), and 256 GB/s Random Write throughput – unmatched in the industry.
I recently posted an extensive benchmark of PostgreSQL running on Datrium DVX platform (here) with only 1GiB RAM to enable the database to stress the storage layer instead of caching data in memory. If you want to understand how DAVG and Application latency differs from SAN Controller latency read my previous article here.
With the Datrium DVX 4.0 about to hit GA I decided to run the exact same PGBench workload with a pre-GA release of the 4.0 software and see if, beyond all the new features highlighted here, we would also see a performance increase for this workload.
I made sure that the infrastructure, virtual machine and benchmark options were all exactly the same. To be more precise, I have not modified this lab environment since I ran the previous benchmark. Therefore the only change is the software upgrade from 3.x to 4.x.
– Transactions per Second (TPS) increased by 15.52%
– Average Read latency remained the same at 0.3ms
– Average Application Write Latency decreased by 14.04%
Here is the screenshot of pgbench, and here you can find the screenshots for the previous run.
This performance increase is workload dependent. However, Datrium DVX 4.0 has been further optimized for large datasets with high throughput and high I/O count. Some of the application datasets we have been testing and pushing in our Solutions lab are between 15 and 20TiB with extremely demanding I/O patterns. More on that later!
That’s just another important value of software-defined-architectures — as microprocessors gain in performance and vendors keep improving their software we will just continue accelerating workloads, providing lower latencies and providing users with better experiences.
This article was first published by Andre Leibovici (@andreleibovici) at myvirtualcloud.net