A recent conversation with an Oracle ERP customer led to some interesting use-cases for EMC XtremIO.
When running Oracle databases on XtremIO it is typical to see 2:1 database compression, so, for example, a ten TB database sitting on XtremIO will only require five TB of storage. Slightly higher and slightly lower ratios have been reported.
There is about a 25% storage overhead with XtremIO, used for XtremIO’s XDP RAID (8%), metadata destaging and vault space giving 75% usable capacity. Certainly much lower than the 67% overhead on Exadata three-way RAID mirroring giving 33% usable capacity.
Plus a snapshot of a source database taken on the same XtremIO cluster is initially simply a set of in-memory metadata pointers requiring no storage space, only generating storage usage as deltas are generated against the read/write snapshot. (OK, the metadata has to be stored somewhere and is part of the 25% overhead mentioned above.)
So this customer is facing two challenges with their 15TB Oracle R12 ERP environment.
They are in the middle of an upgrade to the system integrating new business units after some M&A activities.
They are looking for a more robust DR strategy.
The upgrade is generating the multiple environments for dev/test/QA/patch/training typical of ERP upgrades and implementations. But here note with a 15 TB database source database that is three-way mirrored on Exadata each environment is consuming 45 TB of storage on Exadata. Not pretty.
What is proposed is to deploy a few appropriately sized commodity servers for their Oracle RAC hosts and run a Dataguard physical standby residing on XtremIO from their prod R12 ERP residing on Exadata. Immediately they reduce the footprint from 15 TB (45 TB) to about 7.5 TB with XtremIO’s compression had they used another Exadata appliance for DR. But here is the kicker: once on XtremIO snapshots can be taken off the Dataguard standby to provision the dev/test/QA/patch/training environments requested by the implementation team. And they are “free” from a storage footprint perspective except for the deltas generated on the read/write snapshots and some additional metadata. But even the deltas and metadata are compressed. Pretty.
And should they need to failover to DR, with an IOPs sizing and design exercise based off their prod implementation, they are assured the XtremIO storage subsystem will support the IOPs requested with production usage. Pretty Polly.