Slashdot Mirror
The End of Sun's Cobalt Servers
knighten writes "Sun Microsystems has taken the last of its Cobalt line of server appliances off the shelves in favor of the AMD based Sun Fire line." The article makes note of several relevant bits of history regarding Cobalt, the Appliance Server market, and Sun's Linux strategies.
The article reads like the cobalt was the only true "server appliance" left. As I see it this forgets all about blade servers, network attached storage and a nifty box I saw from IBM that allowed you to stack multiple servers to form a larger one footprint box. They have even moved into the consumer space with media servers and firewalls. The author obviously knows nothing about what is or isn't a buzzword.
I know I'm gonna get nailed for this, but the fact is that Sun should have never bought them. In a way, Sun was unworthy - Sun's CEO was too jealous of Microsoft to ever make a service based approach work, or at least be competitive price wise.
:)
From a data center perspective, yeah its true that Sun boxes can do some things better than x86 boxes running Linux, but I can't tell you how many times I've seen companies buy 100K worth of Sun servers to do services that I know darn well could just as well by an x86 box or two. It always amazed me to see the salesman talk "scalable" for systems that were really farmable. Yeah, experience with high end Sun boxes was great for my resume, thanks, but I wanted my career to have meaning too - and having a bunch of overpriced toys just for the sake of ego seems a little shallow, don't you think. (Sorta like Sun's CEO,
IMHO, the Sun just needs to set. Now that 64 bit Opetrons are out, they will have almost nothing to offer in the midrange. The lost the lowrange a long time ago, but are still in denial. And in the high range, the IBM and HP can beat them out in all categorises.
They originally paid about 2 Billion, yes with a B for the company and basically have nothing to show for it.
Sun servers scale from 1U/1CPU lower cost servers (5K) and developer stations to clusters of over 300 CPU servers, all with full binary compatiblity. I have yet to not be able to take software off a 1CPU low-end sun box and not be able to run it on the top-of-the-line servers without any recompiling.
This provides the capability to develop on low-end boxes without the headaches associated with recompiling on production servers and shortens our development cycle.
Most datacenters I've been in could not ignore the x86 and could not risk being locked into one vendor - so the benefits of scalability sorta wiped themselves out by the simple fact that Sun was too closed and forced companies to have multiple vendors. In addition, nobody can be cost competitive unless they can be farmable as well be scaleable ... Sun never could compete with the x86/BSD/Linux in that area. In theory scaleability sounded nice, but in reality I've never seen a truely scalable datacenter. Finally, now linux works on a wide variety of IBM, HP, Sun, and x86 boxes all the way from pda's to supercomputers. Ironically, it has delivered virtual "scalability" benefits that Sun could have never imagined.
IMHO, java still has too much overhead. We havent got the efficiency and equality that has been promised yet, but I assume we will get there someday.
Yea, that's not what he's saying though. Solaris SPARC runs only on SPARC, Solaris x86 runs on x86.
The article says that Solaris SPARC runs on x86, it's a contradiction.
- Kate
"DNA is life. The rest is just translation."
Your statement about a 'truly scaleable datacenter' does not make sense. You say you've never seen one, but then you mention that Linux has delivered scalabilty beneifts Sun could never imagine. Please make up your mind which it is.
Myself, I have seen scalability. I've have seen applications start out on 2cpu $20K database servers and migrate to E15-type servers without any code change. I'm not saying Linux (or AIX) cannot do this, but this is scalability that Sun does provide at a competitive price.
Having priced the cost of multi-CPU server-quality x86 platforms, there is very little cost benefit to going there. Multi-CPU server quality x86 boxes cost almost as much as the same Sun boxes.
As for Java, since you did not mention any alternative, I cannot provide any response. However, since the application servers and web server provide enough bandwidth, there is no reason to switch to anything else. I am not ready to jump on the 'let's switch everything to Linux' bandwagon yet, but I am on the 'let's review it as we deploy new products or grow existing ones.'
I am sure Linux will migrate into our datacenter, and eventually support production applications. It just doesn't make any sense to replace what works and is proven with something else unless there is a clear cost advantage.
I rarely read replies, it's my opinion and if you thought about your opinion a little more, I'm OK with that.
Risk analysis applications that process 10 million complex transactions in a few hours and produce hundreds of megs of output.
A large server may not be dedicated to just one database. There is a significant cost savings in purchasing a single high end system for several databases rather than splitting them over several smaller systems. With the partitioning technology available from Sun, you can start with a smallish (15 CPU) server and grow up as is needed. Using smaller servers to start with is cheaper, but the cost of swapping them out is expensive if the database requires more horsepower than the system can deliver. (Sun isn't the only one to support hardware/software partitioning, I'm just using it as an example.)
It is always a juggling act to find the appropriate cost/performance mix that provides both long term and short term advantages. Purchasing systems that are not expandable is often cheaper in the short term. Purchasing expandable systems is often cheaper in the long run, but the risk is the application may not grow enough to realize the savings before the technology becomes obsolete.
I hate estimating hardware requirements these days...
I rarely read replies, it's my opinion and if you thought about your opinion a little more, I'm OK with that.