Oracle sneaks out Solaris 10 refresh
Perhaps the most important announcement that Oracle made concerning Solaris 10 is that the OS has been synchronized with the quarterly patch updates that Oracle does for its database and middleware software. Now Solaris will be updated with the same cadence, and moreover, Fowler said that Solaris 10 is now part of the nightly validation testing that Oracle does for its database and middleware stack.
"This is only the beginning of how we are going to improve the reliability and consistency of Solaris when running the Oracle stack," said Fowler.
Oracle's Solaris Cluster, formerly known by the "Full Moon" or Sun Cluster name, has been revved to a 3.3 release as well, and now Oracle WebLogic Server, Siebel CRM, and SAP ERP suites can be encapsulated in Solaris containers and clustered for high availability. Solaris containers can also use network attached storage arrays (including Sun's own NAS products) instead of direct-attached storage arrays. Solaris Cluster 3.3 can now use ZFS as its file system on Sun's Storage 7XXX NAS boxes, and the HA clustering product spanning Oracle servers can use InfiniBand as the connectivity to storage arrays as well as to the outside world.
Solaris Cluster 3.3 has been equipped with dynamic load distribution for applications running on the clusters and can actively monitor storage resources as it has been doing for server resources. (The real question is what place does Solaris Cluster have in a world where Oracle would seem to want to RAC everything).
The Solaris Studio compiler set has been updated with a 12.2 release as well, and C, C++, and Fortran applications compiled with the earlier 12.1 release can see as much as a per cent performance improvement if they are compiled under 12.2, according to Fowler. The compiler tools have a new memory debugger tool that works on executables, the performance analyzer has improved Java profiling and run-to-run comparison, and the compilers have tweaks that make applications more easily parallelized and run better on multicore/multithreaded platforms.
"This is critical," explained Fowler, "because the Sparc roadmap includes systems with thousands of threads and hundreds of cores."
Then again, Sun said the same thing in 1997. We'll see. What Sun has been promising with Solaris 11, due next year, is that the operating system will scale from a few terabytes of memory and hundreds of threads to "double digit" terabytes of memory and thousands of threads. As El Reg reported last month, five years hence, Oracle plans to kick out a Sparc box with a server with 128 cores, a stunning 16,384 threads, and supporting 64 TB of main memory and 256 logical domains.
Such a big bad box, Fowler said in August when discussing the Ellisonized Sparc server roadmap, would be able to crank through 120 million transactions per minute on OLTP workloads - forty times that of the forthcoming Sparc T3 machine - and do 50,000 Java operations per second - ten times that Sparc T3 box.
You can download Solaris 10 update 9/10 off the Oracle Technology Network here for free for either x64 or Sparc iron, but it is not intended for production purposes. If you put it into production, you are supposed to pay for Oracle Premier Support.
On non-Oracle iron, that support will run you $1,000 per socket per year on machines with from one to four sockets and $2,000 per socket per year on machines with more than four sockets. On Oracle iron with Solaris configured on it, Solaris 10 support costs 8 per cent of the invoiced price of the system you buy from Oracle per year.
The server chipheads over at Ideas International have just put out an analysis of Solaris 10 support costs here. By the way, Oracle is charging 12 per cent of the street price for the system as the annual support rate for the physical server, including the operating system.
Bootnote: This story originally said the annual hardware support was 12 per cent of the cost of the iron. This is incorrect. The 12 per cent covers the cost of both the hardware and the software in the Oracle system. ®
Sponsored: Hyper-scale data management