Had an excellent conversation with a customer today about Oracle Exadata and the organization challenges it poses. Keep in mind that this conversation applies to many other converged appliances and not just Oracle Exadata. What the customer expressed, contrary to Exadata marketing, is this big appliance must be managed my multiple groups and not just the Oracle DBAs. For example, The Infiniband switches and their management (patching, hardening, etc.) was the ownership of the network group. As this was explained to me it really struck home as the network engineers are the experts and should own the infiniband switches and the management switch.
Okay but that’s all that might have to be delegated outside the DBA group, right? Not really, as the KVM, rack and all physical connections were managed by a physical datacenter group. Understandable, as power, cooling, weight, airflow and access do require other expertise including security.
Lastly, this customer had a tools group responsible for managing all the all software tools including Oracle Enterprise Manager (OEM), Oracle Grid Control (GC) and many others. The same concerns do apply to tools: security, management and standardization.
So who really owns Oracle Exadata? The customer I talked to expressed that the "who managed and owned what" within Exadata took two weeks of meetings to nail down. Wow! The real answer might be that it’s different from company to company but considering this recent customer conversation I believe most DBAs really shouldn’t plan on owning everything Exadata.
Fascinating! I ran into some of the same issues back in the late nineties and early 2000s over NFS for Oracle databases. While the manageability and cost advantages of NFS for Oracle were obvious and compelling, many organizations could not implement it because of exactly this type of political issue.
Oracle databases were owned by the DBA group. SAN storage was owned by the storage networking group. Who owned the IP network used for NFS (typically a flat, isolated network for only Oracle data NFS traffic)?
IP networks were typically owned by the network administration group. The DBA group worked well with the SAN group, and they did not want to give up that connection.
In one case, there were even collective bargaining agreements in place covering all of this!
Of course, what Oracle is really attempting to accomplish is a return to the days of Big Blue, where eveything was under the control of one group, and there was really only one vendor in the data center: IBM. (For this reason, I have frequently heard the Oracle / Sun / Exa**** strategy referred to as "Big Red".
The rub on that idea is of course what do you do with your Exchange, SharePoint, SQL Server, Sybase, DB2, etc., etc.? Unless and until Oracle comes up with an interesting and compelling story for managing all of the workloads in a data center (which admitted IBM did successfully accomplish up until the early 90s), this strategy seems to be fatally flawed to me.
In addition, if there are hardware failures in exadata storage systems( consider simple disk failure), this task will be compelled to taken out of the DBA group (same scenario as "network engineers are experts"); again it may vary a little site to site