Oracle Solaris and SPARC Performance, Part 1

In the wake of the SPARC T5/M5 launch last week, there's a lot we can discuss, especially related to performance.

To start with: we've just released an update to Oracle Solaris Studio, with compiler optimizations specifically designed to get the most performance out of applications on Oracle T5, Oracle M5, and Fujitsu M10 systems.

Oracle Solaris Studio compilers generate code that is up to five times faster than code compiled with open source alternatives, and this Platform Specific Enhancment (PSE) to Oracle Solaris Studio 12.3 can get you up to 10% more performance on the latest SPARC processors than code generated with the pre-update 12.3 release. That's a nice boost on top of the already awesome performance inherent to the new processors themselves.

This update is available on My Oracle Support for customers with a support contract; search for Article ID 1519949.1 in the Knowledge Zone for more information.


The problem is not one of performance, but one of adoption: in order to increase adoption, people have to be able to afford the hardware to play with at home, so that they can gain both expertise and confidence. It would certainly be naive to assume that one should do that in production at work, or that the employers will fork out the extra cash (and time) to set up test systems for people to play with, which neatly brings us to my point:

the Oracle hardware is so exorbitantly expensive for the average sysadmin, that they could not afford to buy it to get confidence and experience with it at home.

So, when a sysadmin faces a decision or is asked for input, (s)he will naturally gravitate towards that which (s)he already knows, and in this day and age that is not Solaris, let alone Oracle Solaris.

Blogs, presentations and expensive certification courses only go so far.

How does Oracle plan to deal with the issue of adoption and prices out of reach of the IT professional?

Posted by UX-admin on April 08, 2013 at 08:09 AM PDT #

It's a good point. Although Oracle systems are very cost-effective in a server environment (especially compared to our competitors), they're not in the price range of someone looking for a home system.

That's where a key differentiator for Oracle Solaris kicks in: we're the only enterprise UNIX that runs on both SPARC and x86 systems.

One very popular option is to run VirtualBox (see ) on your PC or Mac desktop or laptop. You can then set up and run a Solaris instance with virtually (see what I did there?) no hardware outlay, although the more memory you have, the better.

As for setting up test systems at work, you have even more options there, as it's again quite inexpensive to set up a SPARC logical domain, or a SPARC or x86 zone, whenever you need a test environment. We have customers who mix test and production environments on the same hardware with confidence -- and they can actually then instantly shift test instances to production, and vice versa.

Add to that capabilities such as Secure Live Migration (a feature of Oracle VM on both SPARC and x86 systems) and Oracle Solaris Cluster (which allows you to protect both physical and virtual instances, and you've got some very powerful, flexible, and cost effective options.

Posted by Larry on April 08, 2013 at 09:17 AM PDT #

Unfortunately VirtualBox does not perform at all when simulating LAN or WAN environments (I tried!): the "network" simply locks up, to the point where rebooting virtual instances does not help.

LDOM's are not an option because employers do not want to even give time to the employees to familiarize themselves with technology: projects are always late, over budget, and must go into production yesterday.

The practical knowledge of working with XSCF, or ALOM/ELOM/ILOM, fiber channel adapters, firmware upgrades and the like in conjuction with Solaris can only be gained by tinkering with the hardware at home. Doing so at the employer's expense is no longer an option (we have Dot-Com bust to thank for that).

Given the situation, most system administrators today opt for something they are comfortable with, which is usually redhat GNU/Linux on DELL. I have seen this scenario repeat itself over and over again in the past ten years, much to my dismay.

Besides, LDOM's are not a viable technology in a large server environment because they are different from zones: if I have Solaris on i86pc and sparc, it makes sense to administer them the same way. That same way is zones, rather than relying on specific hardware implementation which most people cannot even try out, and create a "salad" of environment management.

There is a problem, and something needs to be done about it. VirtualBox and LDOM's are not the solution to adoption problems. It would be a shame if Linux continued to gain momentum at Solaris' expense just because there is no plan in place to make Oracle hardware affordable to professionals. Those professionals are future revenue.

Posted by guest on April 08, 2013 at 01:13 PM PDT #

This could be a case of "your mileage may vary." We have customers who are actually using VirtualBox in production on top of Oracle Solaris, within zones, to provide secure virtual instances of non-Solaris operating systems.

That being said, if you haven't engaged with the VirtualBox team on the issues you've found, please do if you can. I've found them to be very responsive in the past.

If you're saying "LDOMs aren't an option because no one can take the time to use LDOMS" -- again, this is definitely not true in a large number of sites in production today.

If your case is that you would choose to use zones because they're available on both SPARC and x86 systems, that'a certainly one way to go, and why not? However, we do have customers who are enthusiastic about LDOMs -- one of the reasons that as of a few weeks ago, they're now a feature of Oracle's SPARC T-series and M-series processors, and Fujitsu's new SPARC processors as well.

Trying to push individual SPARC system costs down to be competitive with the lowest cost x86 or ARM systems is not going to work. What does work, for the vast majority of developers, and most administrative tasks, is the ability to run the same OS on low cost x86 systems, medium range x86 and SPARC systems, and high-end SPARC systems.

Medium range x86 systems are not inexpensive either, and their administrative experience is significantly different from that of the low-cost systems an admin could buy for home use. For most admins and most companies, this is not an issue.

Limiting server functionality to the capabilities an administrator can buy seems extremely restrictive to me. A systems company betting their future on that sort of limitation would be, I don't know, kind of a moonshot. (Which, to be fair, worked once. :-)

Posted by guest on April 08, 2013 at 02:10 PM PDT #

Post a Comment:
  • HTML Syntax: NOT allowed

Security. Speed. Simplicity.
An efficient, open, affordable cloud platform for SPARC and x86 systems.


« August 2016