Tuesday Oct 09, 2007
I posted my previous blog entry when we introduced UltraSPARC T1 based systems. I guess it is time to do another blog now that we are introducing the next gen server.
We just launched UltraSPARC T2 servers (what we call as Niagara 2 servers internally). We have been working on it for 4 years or so and our intention was to integrate as many server capabilities on the chip and also to deliver systems with industry leading power/performance, price/performance & highly optimized for virtualization. We ended up delivering what we were aiming for and more. As you know, as an idea moves thru various phases and gets converted into a product, features get dropped along the way due to cost and / or time pressures. Not so for this baby!
We integrated 10GbE, crypto & PCI functionality on the chip. You may wonder how that helps datacenters. By integrating crypto functionality in the chip, we are able to process a HUGE amount of data at a fraction of compute costs without the need for add-on cards. Now, customers can secure their transactions and data at no cost. The 10GbE gives you terrific near line speed network bandwidth at a very low processing cost. All this means, you need less additional components (which implies better system reliability and availability). And, Solaris along with Sun xVM (virtualization) comes bundled in at no additional cost. This lower acquisition cost and cost of ownership for our customers. Did I mention that all the key pieces of the stack including UltraSPARC T2 processor are open sourced?
My team has been working on further enhancing the performance & scalability of this system and the benchmark results from bmseer as well as early access customer experiences indicate that we have a product which is setting new standards. This is what servers ought to look like. You bet it's worth making noise about!
When we introduced UltraSPARC T1, we had a lot of customers asking us if that system was suitable for running databases. They experimented with it and found out that they can get great database performance on that platform. With UltraSPARC T2 based systems, we have seen in excess of 2x transaction throughput running OLTP apps with databases like Oracle. And, we got that throughput with no special tuning. These apps actually scaled well as we enabled more cores on US T2. Even single threaded apps get a performance boost of around 25% going to US T2 systems.
BTW, for more info, please check Allan Packer's CMT Comes Of Age blog.
Tuesday Dec 06, 2005
I was talking to my neighbor this morning & here is how the conversation went.
"How is SUNW stock doing nowadays?" asked my neighbor.
"Flat" I replied.
"You know, there seems to be a tight relationship between the stock price of a company and employee morale" he said (you are right if you guessed that he is in management).
"Looking at SUNW stock price performance over time, I suspect that your employee morale is probably low. How do you as a manager deal with it?" he probed.
"Actually, Employee morale in my group is pretty good" was my reply.
I could tell that he was not expecting that answer.
"My group works closely and influences product design & development. We also work closely with customers who use these products. We believe we now have the STRONGEST product portfolio in the industry and today, we are announcing new servers which is going to change the datacenter world" I mentioned.
Naturally, his curiosity level increased, "How is the datacentre world going to change?".
"We are introducing servers based on a 32 thread processor chip in UltraSPARC T1. With these servers, we can process lots of simultaneous secure web online transactions needing web, app and database processing with less server footprint, power requirements and you can pack the datacenter footprint better. We even put together a metric called SWaP - Space Watt and Performance to let customers compare the servers".
"Hmmm... Faster, cheaper, smaller, more reliable. Sounds like how the industry evolves normally. What's so different about your servers?"
"Great Question. You are not looking at a 20% or 50% improvement here. You are looking at something like 5 fold improvement in these factors. A datacenter manager can use a LOT less of $$, space, power to do their job. This is enabled by a radical thinking and innovation which started almost 4 years back. So, it will do wonders to Total Cost of Ownership and we expect this platform to cause economic disruption in the datacenter".
"Sounds cool, any customer reactions?"
"Apart from demonstrating benchmark results which beat competition in terms of price/performance, my team have been working closely with many potential customers and they are blown away with the performance and footprint of these products. Some of them actually reran the apps a few times as they could not believe what they saw".
"Now I get why the morale of your team is good" he circled back.
I did not get a chance to talk to him about our how proud my team has been with complete server line refresh with the UltraSPARC IV+ servers as well as the Opteron servers we introduced recently and the positive reaction we have been getting from our customers.
You may want to check Richard's blog & benchmark data to get more info.
[ T:
NiagaraCMT
]
Monday Dec 05, 2005
While I have been reading blogs for some time now, I guess it is about time to show up and get active with blogs. As a brief intro, I manage a VERY talented team of engineers working on improving performance and availability of our products. My group is called PA²E which stands for Performance, Availability & Architecture Engineering.
Our job is to help Sun make better systems (processors, servers, software, networks etc) so that they do better than competition in satisfying customer needs. We work very closely with and influence product engineering teams starting from ideas, design, development through product release phases related to architecture, performance and RAS aspects of the product.
We work closely with our field to ensure that our systems are deployed correctly to meet customer requirements. We use the experience we gain to shape our products. As you can imagine, we would need some "workloads" to really understand what kind of systems we need to build. We participate in industry consortia like TPC and SPEC to develop benchmarks and we develop some benchmarks internally aimed at satisfying specific future customer work needs in addition to leveraging the industry effort.
|
|
|
|
Can you tell me a little more about xVM on SP...
Great englightening article about a great processo...
Sun xVM was recently announced and I have upd...