>I have always wanted to > know what is so special about SPARC hardware that > some people still prefer it over x86.
Unless you're really "jiggy" and "down" with the UltraSPARC processor, and know how to squeeze that extra performance out of an UltraSPARC based CPU, nothing really. If you look inside of any relatively modern Sun SPARC based hardware, you'll notice a striking similarity to an i86pc based server. For example, the new T5220 "Huron" system is a perfect example of this. The difference is, they designed new RAM chips, incompatible with the i86pc platform of course, for that particular system. So what differentiates a sparc based system from the i86pc competition? Well, it used to be the hardware redundancy and the hot swap capabilities. But nowdays, even the cheapest barebone server from ASUS, GIGABYTE OR TYAN has some sort of hot swap capability, and they all have models with redundant hardware (such as power supplies and fans). So, no differentiation there. Then it used to be, Sun systems had ALOM. And that really used to be a great, important differentiaror. But nowdays even the cheapest i86pc servers can, at the very least, send console output to ttya, and almost every single one of them can either use IPMI or have an IPMI module as an optional component which you can buy. And, ironically enough, Sun also sells ALOM as an optional component for certain models they sell. So, no differentiation there either. Then the differentiating factor was, Sun systems had built in SCSI and greater overall throughput. Which was all fine and well, until it was time to do some number crunching. The UltraSPARC processors happen to be notorious for being the slowest RISC processors ever designed; as I wrote at the beginning, special "love & care" is required to squeeze good number crunching performance out of an UltraSPARC processor. Especially the II and IIi series. And especially the UltraSPARC T1 and T2 family, which just so happen to be among the Sun's recent processors. If you can't do lots and lots of parallelism, the performance is dismal. But nowdays, the i86pc platform has been so advanced, that there really is no differentiation. The i86pc is like a big block V8: inefficient and gas guzzling, but simply so much brute force, that it just doesn't matter whether it's more primitive and less elegant than a SPARC based system. All that low level stuff has been abstracted, in no small part by Solaris himself, and especially by Oracle. Now, if you got here reading this far, you might think I'm really bashing Sun hardware and don't like SPARC. But the truth is, IF there were Sun V240s out there for a *sane*, decent price, I'd load up on them in a heartbeat. As they are still outrageously overpriced, even the used models on ebay, I turn back to TYAN for my computing needs. And hey, Solaris runs just as fast if not faster on those TYAN multicore Opteron systems! I also wouldn't mind having a T2000 in my rack, despite dismal single thread performance. Why? Because that would be that perfect compile server for all my SPARC based systems (yes, I actually do have quite a bit of SPARC hardware). I've used T2000s before and I can really get nice performance out of them for compiling with Sun Studio 12. But, at $9,000 for the most basic (and useless) configuration, and around $20,000 for a *sane* configuration, I'll be damned before I fork out that kind of cash for a bunch of processed sand and iron put together! This message posted from opensolaris.org _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
