by Anonymous Coward writes:
on Sunday August 23, 2009 @09:31AM (#29163039)
Sun sucks. Their overpriced hardware got eclipsed by Pentium 4 PCs which could do the same work for 1/10th the price. In the end, the only advantage Sun platforms offer over PCs is the capability to use massive amounts of RAM (64G, 128G, and beyond). But make no mistake about it, Sun got eclipsed by Intel. Moore's law has a harsh penalty for those who don't keep up.
On the other hand, Oracle having a say in OO, Java, and other projects is a bit scary. I'm not so in love with Oracle's embrace of FOSS or even the concept of FOSS and GNU.
Unfortunately said Pentium 4s also would fail 10x more often.
I don't know if you've worked (ie, have had direct administrative experience) with any of the larger Sun hardware such as E2900 and above, or even the Ex500's from back in the day, but if you did you'd also know that these servers have a knack for uptime and resiliency that x86 servers, even to this day, have never had. There was a reason for those higher costs.
Unfortunately said Pentium 4s also would fail 10x more often.
I don't know if you've worked (ie, have had direct administrative experience) with any of the larger Sun hardware such as E2900 and above, or even the Ex500's from back in the day, but if you did you'd also know that these servers have a knack for uptime and resiliency that x86 servers, even to this day, have never had. There was a reason for those higher costs.
At the same time, the application landscape changed to prefer scalability that allowed servers to be down without impacting the whole system. A single machine no longer was so important.
And up until a few years ago, people still went with Sun when they had a single important node.
The choice became between more servers that will crash slightly more often with less overall impact to the application, and more servers that will crash slighly less often.
Larger Sun hardware was never amazingly reliable anyway. Wh
Actually the advantage is a fast backplane, not the memory. You may remember they bought the rights to the
Cray asynchronous (really packet-switch-like) backplane quite a number of years ago, and have been expanding on it since.
It's easy to build a fast chip if it never has to maintain cache-consistency with anything off-chip. If it has to stay sane, even with only 64*4*2 = 512 threads banging on the same memory range, it not only takes an expensive bus, but it's also memory-transaction-rate limited.
That's why you read about transactional memory in Linux Weekly News: we all
need it, SPARC and Intel both.
Actually the advantage is a fast backplane, not the memory. You may remember they bought the rights to the Cray asynchronous (really packet-switch-like) backplane quite a number of years ago, and have been expanding on it since.
Just to clarify - faster can mean higher bandwidth or lower latency. Sun really screwed themselves by building machines that had bad memory latency, but good bandwidth. Real-world appplications like transactional databases care more about memory latency than bandwidth.
I absolutely agree with your last sentence, although I see most of the Sun efforts to be toward reducing the latency, by not making the bus wait in between the request and the reply. That speeded up small memory transactions a lot, targeting the low speed of locking operations that have to go to memory and/or achieve global cache consistency.
I would like to see latency bettered, too. To me, it's at least an order of magnitude high.
It certainly was at the time: the e25k was something like 72, and the approach has subsequently spread to smaller, faster crossbars, sometimes used in pairs to decrease latency and increase bandwidth.
Sun sucks. Their overpriced hardware got eclipsed by Pentium 4 PCs which could do the same work for 1/10th the price. In the end, the only advantage Sun platforms offer over PCs is the capability to use massive amounts of RAM (64G, 128G, and beyond). But make no mistake about it, Sun got eclipsed by Intel. Moore's law has a harsh penalty for those who don't keep up.
Coming from someone who couldn't tell you the difference between an Intel desktop processor and their Xeon line to save their life, much less know what RAS features SPARCs have over anything Intel has.
Don't forget massive amounts of CPU's, expansion cards, dynamic reconfiguration, partitioning, etc, etc.
Midrange computing is all a lie I guess, someone go tell IBM, HP and Fujistu.
These screamingly hilarious gogs ensure owners of X Ray Gogs to be the life
of any party.
-- X-Ray Gogs Instructions
Not sure if this is more funny or scary (Score:1, Insightful)
Sun sucks. Their overpriced hardware got eclipsed by Pentium 4 PCs which could do the same work for 1/10th the price. In the end, the only advantage Sun platforms offer over PCs is the capability to use massive amounts of RAM (64G, 128G, and beyond). But make no mistake about it, Sun got eclipsed by Intel. Moore's law has a harsh penalty for those who don't keep up.
On the other hand, Oracle having a say in OO, Java, and other projects is a bit scary. I'm not so in love with Oracle's embrace of FOSS or even the concept of FOSS and GNU.
Re: (Score:3, Informative)
Re:Not sure if this is more funny or scary (Score:4, Informative)
Unfortunately said Pentium 4s also would fail 10x more often.
I don't know if you've worked (ie, have had direct administrative experience) with any of the larger Sun hardware such as E2900 and above, or even the Ex500's from back in the day, but if you did you'd also know that these servers have a knack for uptime and resiliency that x86 servers, even to this day, have never had. There was a reason for those higher costs.
Re: (Score:3, Interesting)
Unfortunately said Pentium 4s also would fail 10x more often.
I don't know if you've worked (ie, have had direct administrative experience) with any of the larger Sun hardware such as E2900 and above, or even the Ex500's from back in the day, but if you did you'd also know that these servers have a knack for uptime and resiliency that x86 servers, even to this day, have never had. There was a reason for those higher costs.
At the same time, the application landscape changed to prefer scalability that allowed servers to be down without impacting the whole system. A single machine no longer was so important.
And up until a few years ago, people still went with Sun when they had a single important node.
The choice became between more servers that will crash slightly more often with less overall impact to the application, and more servers that will crash slighly less often.
Larger Sun hardware was never amazingly reliable anyway. Wh
Re:Not sure if this is more funny or scary (Score:5, Informative)
Actually the advantage is a fast backplane, not the memory. You may remember they bought the rights to the Cray asynchronous (really packet-switch-like) backplane quite a number of years ago, and have been expanding on it since.
It's easy to build a fast chip if it never has to maintain cache-consistency with anything off-chip. If it has to stay sane, even with only 64*4*2 = 512 threads banging on the same memory range, it not only takes an expensive bus, but it's also memory-transaction-rate limited.
That's why you read about transactional memory in Linux Weekly News: we all need it, SPARC and Intel both.
--dave
Re: (Score:2)
Actually the advantage is a fast backplane, not the memory. You may remember they bought the rights to the
Cray asynchronous (really packet-switch-like) backplane quite a number of years ago, and have been expanding on it since.
Just to clarify - faster can mean higher bandwidth or lower latency. Sun really screwed themselves by building machines that had bad memory latency, but good bandwidth. Real-world appplications like transactional databases care more about memory latency than bandwidth.
Re: (Score:2)
I absolutely agree with your last sentence, although I see most of the Sun efforts to be toward reducing the latency, by not making the bus wait in between the request and the reply. That speeded up small memory transactions a lot, targeting the low speed of locking operations that have to go to memory and/or achieve global cache consistency.
I would like to see latency bettered, too. To me, it's at least an order of magnitude high.
--dave
Re: (Score:2)
It certainly was at the time: the e25k was something like 72, and the approach has subsequently spread to smaller, faster crossbars, sometimes used in pairs to decrease latency and increase bandwidth.
--dave
Re: (Score:2)
Sun sucks. Their overpriced hardware got eclipsed by Pentium 4 PCs which could do the same work for 1/10th the price. In the end, the only advantage Sun platforms offer over PCs is the capability to use massive amounts of RAM (64G, 128G, and beyond). But make no mistake about it, Sun got eclipsed by Intel. Moore's law has a harsh penalty for those who don't keep up.
Coming from someone who couldn't tell you the difference between an Intel desktop processor and their Xeon line to save their life, much less know what RAS features SPARCs have over anything Intel has.
Don't forget massive amounts of CPU's, expansion cards, dynamic reconfiguration, partitioning, etc, etc.
Midrange computing is all a lie I guess, someone go tell IBM, HP and Fujistu.