Compression Ratios on VMDK's

[ron_fisher]

Greetings,

We have two cs300's and a cs500 all at 2.2.5.0-197583. These store only VMware VMDK and related files. I know compression can vary widely bases on what data types are in given files. In our case we have mostly Windows 2k3 through 2k12r2 running file services, SQL, oracle on linux and various other apps, fairly typical mix of "stuff".

cs300-1: 21% compression with 28% of 15T used, 2% snapshot space

cs300-2: 27% with 60% of 15T used, 5% snapshot space

cs500: 39% with 30% of 51T used, 5% snapshot space

Are other people seeing better compression rates in this kind of environment?

Thanks

Ron

[ron_fisher]

Wow am I the only one running VMware and VMDK's on Nimble? It would be interesting to know if anybody is getting the "advertised" 50-60% compression we heard from the sales guys. We had this early on but that deteriorates daily.

[Nick_Dyer · Accepted solution]

Hi Ron,

With any form of data reduction, any savings you see are solely dependent on the data that you have, and hope susceptible it is to compression/pattern matching. Data being stored within VMDKs has no bearing on the impact to it's susceptibility to data reduction, it's what data is inside it.

Nimble's average compression across a variety of working sets is roughly 1.8-2x from looking at Infosight, with database workloads being at the top end (2-4x), and file/email workloads being at the lower end (1-1.5x). VM OS drives tend to be around 1.5-1.8x (30-40% reduction). How heavily weighted your data sets are towards VM images/file email will impact your overall space savings vs being heavily weighted on database data sets. Also your compression figures would only deteriorate if you're placing more, non-compressible data on the array (ie moving across 10TB of MS Office 2013 documents). If you were to load on 10TB of Oracle data I can guarantee you that would provide you a nice increase in your compression figures.

Hope this helps.

[ron_fisher]

Thanks for the reply Nick. With a hundreds or thousands of VM's scenario's that customers can get into, categorizing data types becomes nye impossible. We will be adding a good bit of Oracle to one of our arrays soon so maybe we'll see an uptick. I guess I should just look at it as a frosting on the cake. The array based snaps have already saved our bacon a few times.