The details are a bit sketchy, but server maker Sun Microsystems is apparently getting ready to launch a rejiggered version of its Sparc T2 server platform that will implement an external network interface instead of using the on-chip (and very fast) networking in original Sparc T2 servers. Presumably, this engineering change …
Yet another example of Sun's inability to be profitable
The last thing we need is more model types, especially cheaper ones. When you are not profitable you need to offer less options with price premiums. With the exception of about 10 customers the T2 systems are only replacing USIII & USIV systems. The customers are locked into Solaris...demand a high price that is just below the migration cost. Does no one understand how to make a profit anymore? Stock is below $3 and dropping....or is that 75 cents....I'm confused.
More models and cheaper models.....
Hi Schaedler, either I'm in one the ten unique companies or I misunderstood your post. We're using a lot of T1 & T2 based boxes for new implementations as well as consolidation of older kit.
In one way your right, the consolidation potential is amazing so sure, T2 is replacing old 440's, 480's & 880's, virtualising the old OS's down onto the newer T2 kit gives a really good support cost saving. However, newer beefier systems are getting implemented onto 5240 and 5440 servers. 8gb dimms are expensive when we need loads of RAM but the 5240 offers very good value IMO.
We've not got a single system company wide using 10g ethernet (yet), we've got front end web and backend monitoring, backup & DB boxes but only aggr'd ethernet where needed. Hence if the newer CPU's make them cheaper still but with the same CPU perf the argument would be even more compelling when weighing up new purchases.
There has to be a cost for Sun to develop this extra CPU variant but I suspect any sale is a good sale at the moment so knocking a few % off to help deals probably benefits Sun more so than holding those few quid from the dev/engineering guys. Lets face it, they were a bit ahead with 10g support on every T chip as far as I can see. Two years time it might be defacto but not now.
Hey ho, gotta dash, Matt will be along soon to speak about other vendors.....
Vic Falls Chip?
I think the Victoria Falls chip dispenses with the on-chip 10GigE to make room for an SMP interface. So if there are Vic Falls chip bins with bad SMP interconnects, a uniprocessor platform would be a great place for them. The chip cost would essentially be zero. Combine it with a lower-cost platform (SATA vs. SAS disks, etc.), and Sun would have a nice entry box. It would also allow Sun to EOL the T1000 and T2000, which would reduce costs.
Speak for yourself
If this thing is affordable, I'm buying!!
Not a new chip?
Perhaps these are chips that failed QC but only in the network functions, and would otherwise be trashed.
I can think of two possibilities....
One; they have an awful problem with chips coming out of the fab with dud onbaord 10GbE controllers, enough to supply a whole line.... unlikley, otherwise news of such a large failure rate would likely have leaked by now. Second option; they are putting ordinary T2+ chips into a simplified motherboard, maybe with some of the socket pins simply not connected through and the external 10GbE ports removed. A simplified motherboard would be cheaper to produce - less tracks = less cost. And seeing how the systems are undercut by cheap'n'cheerful NAS boxen, something had to be done.
Of course, a simpler way to lower the price, increase the capacity of the system, not cripple the networking advantage, and generally make the 7110 more competetive, would be to simply switch to using a mix of SSD and SATA instead of 10k SAS disks, as suggested by Phillip Fayers on Sun's Mark Hamilton's blog. Mark's answer is that "compromises" had to be made and that the SSD/SATA mix is available further up the model chain! Surely it would make more sense to put the cost-saving options into the entry model from the start? If a Sunshiner can see these things, why can't Sun?
Is SPARC even relevant anymore?
Solaris is a good OS but ditch the SPARC gear. Those guys take forever to come out with new versions of their main processing units. Look for instance at how long it has been since they came out with their own Medium and High End servers, not a Fujitsu clone. Its been years!!
The T2 products are better then their T1 predecessors, But look how long it took for Sun to crank those out vs the many turns of chips from Intel. The architecture is a bit weird too with only two 10GE pipes for all those cores and threads that it can run is a major choke point. Not to mention, who knows how to really do the needed parallel programing for those kinds of systems?
Stick with the Intel products and run Open Solaris, or better yet go to Linux, where the really smart guys are dedicating their time and energy. That way you will have an OS that will be around for the next 20 years, on an architecture that will evolve, but still work on that old code. Who knows if Sun will even be around in the next 5 years??
Lots of failed chips at Sun
I knew a guy that worked on the T2 chips at Sun. He said that they had horrible problems with the fallout on the T1 chips, so that is why they made 6-core versions of the chips available. The new T2 chips are even more complex, so expect the failure rates to be high. What I don't know is how often the T1 products failed at Sun's customers sites. Higher then normal I'll bet.
Still think a whole product line being supplied by dud chips is unlikely, but the simplest way to check would be to check the part numbers for the CPU in the normal and budget models. Or just to take a CPU from the budget model and plug it into the normal model and see if the onboard 10GbE still worked. To actually cripple chips would probably be more expensive, so I'm sticking with the crippled m'board idea.
Oh, hold on a sec - sorry, I assumed that someone would actually buy some of these so we could compare the chip parts! :P
Probably does use Xaui
The expensive part of a 10Gig interface isn't the mere presence of the MAC inside the chip and the Xaui pins on the processor, its the XFP optics. I'd guess that the new server does use the Xaui in the Sparc chip but connects it directly to the 10 Gig Xaui on a Broadcom switch, which fans it out to multiple gigabit ethernet links which can use significantly cheaper copper ports.
"What I don't know is how often the T1 products failed at Sun's customers sites. Higher then normal I'll bet."
Actually, we have seen a much higher failure rate of x86 based kit. The T1s and T2s we use have been good as gold.
Surprised no one has mentioned IBM POWER processors. Well, now someone has.
We dumped SPARC for POWER a few years ago. It took a while to really learn AIX (which I now like much better than Solaris), but Sun just can't match IBM with their various virtualization options.
T1 is being withdrawn in January with USIV+
Will announce T2 at 1.6GHz and withdrawl the T1000/T2000.
The ethernet was removed from the T2+ chip along with two memory controllers to make room for the interconnect.
Reliability is certainly a problem but its usually "good enough" for the apps customers use these for. Customers do not use them for databases.
The 4core version was only created to limit the painful Oracle prices. Odds are a chip will only have two bad cores out of the eight without being total junk.
Still not sure when the big cuts will occur. RIF traning for the managers is almost complete.
Finally... a lower entry point CoolThreads
I have been waiting for a lower entry point on the CoolThreads.
This is really needed to compete with the now-in-existence quad-core from AMD and Intel.
A speed-bump is needed in the high-end, to pass the new 6 core Intel CPU's, as well.
It must be hard on the CoolThreads engineering staff, they finally have some competition that they have to deal with - I hope they perform well!
- Pics Indestructible Death Stars blow up planets with glowing KILL RAY
- Hands on Satisfy my scroll: El Reg gets claws on Windows 8.1 spring update
- Video Snowden: You can't trust SPOOKS with your DATA
- 166 days later: Space Station astronauts return to Earth
- What did you see, Elder Galaxies? What made you age so quickly?