I'm currently experimenting with clustered columnstore indexes on a test system, and I have found compression to be high for my data warehouse - a previously 320GB database has now shrunk to 22GB.I have also been looking at the disk activity, and while the larger database commonly hits the cap for disk I/O (250MB/s, I am using an azure DS13 VM) when I check resource monitor, I have found that for a clustered columnstore index, then the I/O is commonly only reaching about somewhere around the range of 20MB/s to 50MB/s. I'm not sure what this means in terms of what kind of disks I need for my production server, however. If I am not hitting my network bandwidth cap for I/O, is there going to be any return on using SSD's for faster I/O and raising the bandwidth cap? I don't seem to be hitting the maximum utilization for my CPU's, so I am not sure how I would be able to force data through more quickly when using columnstore indexes, so I am closer to my I/O cap? I think the delay may be due to decompressing the compressed columnstore data.
↧