SAS recently tested SAS 9.4 on MS Azure instances using Sycomp Storage Fueled by IBM Spectrum Scale (GPFS) file system (https://azuremarketplace.microsoft.com/en-us/marketplace/apps/sycompatechnologycompanyinc15881921038... ). The results of the testing were very favorable.
The test infrastructure consisted of the Medium Storage Cluster offering from Sycomp (11 MS Azure instances) defined like this:
- Five E64-32ds_v4 instances for the SAS clients – each with six P30 premium drives striped together for local storage.
- Six D64s_v3 instances for the Spectrum Scale storage cluster – each with five P60 premium drives.
During the test, we modified some of the IBM Spectrum Scale settings to improve IO throughput:
mmchconfig pagepool=200g,prefetchPct=40,maxFilesToCache=50000,maxMBpS=4000,seqDiscardThreshold=1073741824,workerThreads=768
rhel_iotest results:
The 1-node test gave us 237 MB per second per physical core for WRITEs and 267 MB per second per physical core for READs.
The 2-node test gave us 225 MB per second per physical core for WRITEs and 218 MB per second per physical core for READs.
The 3-node test gave us 167 MB per second per physical core for WRITEs and 150 MB per second per physical core for READs.
The 4-node test gave us 153 MB per second per physical core for WRITEs and 112 MB per second per physical core for READs.
The 5-node test gave us 94 MB per second per physical core for WRITEs and 86 MB per second per physical core for READs.
Please note that SAS recommends a minimum of 100 MB per second per physical for WRITEs and READs.
As you can see, the Medium Storage Cluster can support the minimum IO throughput needs of a 64-core SAS Grid application. Other testing with Sycomp has shown the ability to have even more cores achieve the minimum IO throughput needs of SAS with over 100 cores. Please reach out to Sycomp for more details.
