WekaIO pumps up the SpecSFS 2014 build volume

WekaIO has notched another benchmark win, scoring 5,700 on the SpecSFS 2014 software build test.

This benchmark tests a variety of storage application types: software builds, video streaming, databases, virtual desktops and electronic design automation.  The chart below shows Weka moving to the top spot from its previous equal fourth position (1,200 builds)  behind NetApp (4,200 and 2,200), DDN (1,500) and E8 (1,200).

Weka used a 6-node Supermicro system for the new benchmark run. Its earlier 1,200 build score used a 4-node Supermicro setup.  A bullish Weka representative told us that it will simply add more nodes and run again if a competitor tops 5,700 builds. We have a comment about this below.

The benchmark also records overall response time (ORT), and we have charted the results again, with ORT (blue bars) compared to builds (red line):

By this measure, Weka has the second lowest ORT at 0.26secs, behind a DDN Spectrum Scale system with 0.19secs. The largest ORT is also a Weka system, running on 60 instances in AWS and clocking 3.06secs.

The general trend is for ORT to decrease as the software builds number  increases.


Regular readers will know that I am no fan of SpecSFS 2014 because there is no pricing context. This means it is impossible to compare price/performance and also non-scale-out systems are effectively excluded. This weakens the utility of this benchmark for enterprise customers.

If vendors can merely add nodes to a scale-out system to establish new records, SpecSFS 2014 is effectively a scale-out storage system benchmark only.

If price-performance were added to SpecSFS 2014, Weka and rival software-defined storage products running on X86 commodity server boxes are likely to do even better.  Such a move would put Cisco, DDN, E8, Huawei, IBM and NetApp and other scale-out systems running on proprietary hardware at a disadvantage compared to Weka and its ilk.