I have written a book about Optimizing Oracle on VMware, plus have published papers and given presentations regarding the same. These basically all espouse essentially the same fundamental belief – that database virtualization is here to stay, and is rapidly moving towards mainstream (if not there already).
But a couple legitimate questions or concerns are always brought up (and they are good questions). I’m going to address the three that I hear most often, because I believe having the answers to just these will knock down 80+% of all the roadblocks to virtualizing your databases.
Oracle does not support VMware. Not true. Roughly speaking – a little over a year ago their position was not to support any VM technology. Then about a year ago, once they got their own VM solution (OVM) - they supported that, but really not others. Then shortly after that, they opened up and now support non-OVM solutions such as VMware. But there is one very simple and reasonable caveat - Oracle support retains the right to ask the customer during troubleshooting efforts to at some point to re-host to non-VM in order to eliminate that possible point of contention. That’s more than fair. Plus with the Oracle acquisitions of Sun (who already had several VM technology offerings, including Virtual Box and Solaris Containers) and Virtual Iron – Oracle has truly become a virtual powerhouse in the VM arena (pun intended). So it’s in there best interest to support and propagate virtualization of everything – including databases.
VM overhead is too debilitating for databases. Not true. Servers and memory are dirt cheap these days. And while storage costs have come down, the larger sizes of databases these days has grown to often eliminate savings. So today’s database servers have plenty of bandwidth to host more than one thing. And for those DBA’s overly concerned about the maybe roughly 10-15% overhead, CPU’s and memory are so cheap that factoring this cost into your cumulative or aggregate server needs is really not that big an issue – really. Often the direct power savings as well as the indirect power savings (e.g. lower cooling needs) can more than compensate for the minor increase in sizing calculations.
Databases are too special or critical to virtualize. OK – BUT, DBA’s freely accepted virtualized storage for databases almost a decade ago. Gone are the days of going into the server room and pointing to your database’s disks. And since IO is the “Achilles Heal” or weakest link in terms of performance for any database, thus we’ve already virtualized the most important performance aspect. Thus it seems a little odd to debate or argue about virtualizing the remaining aspects, which generally impact database performance much less than the IO. So we’re just taking the next logical step in a progression we started a long time ago.
With these two issues put to rest, hopefully you and your management will more quickly embrace what’s inevitable – virtualization of database servers. Because none of us like swimming upstream. [;)]