One point is dbfs is supported on Linux but not on Solaris because of the absence of fuse support.
Apart from that I believe some of it depends on your familiarity with one OS over the other. Many of the advantages of Solaris on Sparc are due to the sparc architecture and you will not gain those here for example hugepage size is actually limited by the intel architecture on Exadata.
Many admins swear by dtrace. hat may be one reason to choose Solaris over Linux.
Then again Linux admins may be cheaper than Solaris admins.
I'll jump in and opine here. The availability of Solaris as an operating system for Exadata database servers is quite new. Prior to that, Linux was the only available option. So naturally, most of the Exadata installed base is running Linux database servers, so a lot of bugs and compatibility issues under Linux have been fixed over the years. Solaris has a much shorter track record. So I'd recommend Linux unless you have a good reason to need Solaris (security requirements, for example).
While debating the merits of Solaris vs. Linux may be the intended goal here, I typically suggest that the "best" platform to run a critical business function is probably the platform that your company is "best" at managing. If your environment consists of all Solaris machines and one little Linux server over in the corner, your team probably knows Solaris pretty well and can manage it much more efficiently and competently than you can Linux. On the other hand, if you've never seen Solaris before and have managed Linux machines for several years, I wouldn't suggest that putting your most critical application on Solaris systems makes the most sense.
"Oracle Procedural Gateway for MQSeries" is not available under solaris 11, but it's there for Oracle Enterprise Linux. Make sure all existing database products avaiable for Solaris 11 on Exadata and plus existing support staff.
There might be some performance gains, but probably minimal if any, ZFS will help you with upgrades, dtrace with troubleshooting...
Regarding stability, Oracle is widely deployed on Solaris, and Solaris used to be the preffered Oracle plaform before the rise of Linux.
Now if you look into the future, I believe sparc will become the ideal platform for exadata.
Based on the publised roadmap, Oracle will expand the instruction set of the sparc architecture to speed up the Oracle DB (Number data type, compression..)...
you definitely need good Linux knowledge to work with Cell.
I would say the opposite, it is the compute nodes which needs some (but not a lot) OS skills.
On the cells I hardly do any OS work the only exception being the patching of the new storage server software which I run from the first compute node anyway.
Maybe I read the messages file on a cell after a disk failure but that would be it.
On the database machine the most needed would be DBA skills.
Thanks for all your valuable comments. We went for Linux. After cut-over Exadata, hit few bugs.
But patches are available. I think most of the customers are using Linux and it is working fine.
So, I think we can go for Linux.
If sparc for exadata is what you are looking for then its available today but they called it sparc t4 supercluster.
it has the same storage,infiband and idp protocol and I belive also supports smartscan .it is more expensive than exadata however.
I believe that Oracle develop first on Linux and then port their products over to Solaris. Not sure if that's still true, but should show the O/S on which your product might be considered 'most mature'...