Wednesday, January 09, 2008

Niagara-2 vs. Power6

Interesting observation:

"Siebel is one of the epidemic applications in the enterprise. Look in any company of a certain size and you find Siebel or a similar application. Customer Relationship Management was one of the big hype topics a few years ago. Thus benchmarks for this application are really important for server vendors.

When you dig around in the benchmark whitepapers for Siebel, you will find a really interesting gem of information: You can outperform three IBM p570 p6 4,7 GHz with two Sun Sparc Enterprise T5120 and two T5220.

According to the the benchmarking document from Oracle and IBM a configuration consisting out of one 8-proc p570, one 4-proc p570 and one 2-proc p570 was benchmarked. This configuration was able to serve 7000 users with a rate of 106,157 business transactions per hour. A few days ago Oracle certified the benchmark for our UltraSPARC T2 base servers. According to this document, a configuration consisting out of 4 systems with one UltraSPARC T2 each was able to serve 10000 users with a rate of 142,061 business transactions per hour.

I want to add some perspective to this result: The result is especially interesting, as the IBM Power6 based configuration is vastly more expensive than the one based on UltraSPARC T2. The capabilties of the T2 are really amazing."

[highlights introduced by me]

Update: more details

15 comments:

Ethan Anderson said...

If I can buy a PC with that processor, and it performs better than power (which crushes x86), then I think I might. As long as you have the source, you can use whatever ISA suits you best. For now, I prefer linux over opensolaris and freebsd..

A good processor is a good processor.
Is this in my price range, as an end user?

Anonymous said...

at the ibm webserver there must be a typo..
power 6 starts at ~3,5GHz so they have used a power5+ with 1,9GHz or something

what i also see is response time:
ibm: 0.091, 0.039
sun: 0.287, 0.154

and for the price diference
i dont think an HA feature ritch server like the p570 ist compareable to the T2 servers we have now...
set the T2 against something like a p510
and set the p570 against the hopefully coming ~M4000 T2 edition =)

but the raw power of the T2 is impressive anyway =)

milek said...

Power6 4.7GHz was used for database and application layer.

Response time - well, most of the time if response time is reasonable you don't care - all you care is how much concurent users you can serve. It doesn't matter if webpage is generated 0.1s longer, what matters is how much users can you serve.

HA - well, 2xT5220 + Sun Cluster is still waaaaay cheaper being faster at the same time and providing better resiliency than p570.

Anonymous said...

Please explain how the 2xT5220 + Sun Cluster provides better resilience than a Power 6 based 570 solution ?

milek said...

There are of course different scenarios when it comes to HA. However I've learned that having two separate boxes in real life gives you more resilience than one box no matter how redundand.

Then there's a cost factor and specific usage case. While POWER is a nice CPU it's overhyped when it comes to performance in real workloads and it's way too expensive.

Anonymous said...

Of course 2 seperate physical servers will be more redundant.

I was asking how a clustered T2 environment provides more resilience than a clustered Power 6 environment, which is what you were suggesting.

What I am suggesting, is that you are talking nonsense !

milek said...

You have misinterpretted me. I was only suggesting that 2x Niagara boxes clustered would not only provide better performance but still will be waaay cheaper than single, non-clustered p570.

Anonymous said...

and can you back that up with any facts, or is this just a huge assumption on your part ?

milek said...

Although IBM tries to hide prices for p570, they have to disclosure prices for some benchmarks they are publishing. Check thin benchmark published by IBM with prices published by them - http://www.tpc.org/results/FDR/tpch/IBM_570_10000GB_20071015_FDR.pdf

p570 with 2x 4.7GHz Power6, 32GB RAM costs about 219,000$ each! Keep in mind that for application layer in Siebel test IBM used server with 4 chips not two and with 64GB ram not 32GB so the config is still more expensive. Then you have to pay for AIX too...

Now let's check more pricey Niagara based server - T5220. You can check it on Sun web pages - they are not ashamed about they prices :)

T5220, 8-core, 32GB RAM costs about 32,000$. And you don't have to pay for Solaris - it's for free.

So you can buy almost 7x T5220 for the price of one p570 in a small configuration. And if you want to use Sun Cluster it is for free too.

Anonymous said...

Ok, I disgree with your thinking, and here is why.

1. Your basing your figures on IBM list prices, yet the SUN figures are after discounts. Who actually pays list prices anyway ?

2. Your comparing an enterprise class system like the p570, which can scale way beyond the capabilities of the T5220. By having such a small amount of resource in it in a server clearly designed to be used in configurations much larger than entry class. A 2 way 32 gb power 6 570 will never be cost effective. Thats like me moaning that an M8000 with 32GB of ram is too expensive.

3. The Siebel benchamrk even has a 2 way p570 as a web server. I mean, please, be real. Its even quoted as a Power 6 p570, despite having Power 5 processors.

4. Licensing costs. Lots of cores can and do make a difference.

5. I remain unconvinced about the difference in response times. With response times so much faster on IBM, i would be surprised if this did not mean that the P6 system was capable of running more users.

6. The IBM design used for the benchmark is also unrealistic. For example, why have seperate p570 systems, when 1 system which can be partitioned will do the job.

7. IBM comes into its own with partitioning and virtualisation, both of which are far superior to anything Sun has to offer. buit thats a different argument.

8. Cheapest isnt always best. if it was, we would all run on x86 linux.

Lastly, i will add that i do think the T2 systems are an impressive piece of kit. But i dont really see them as natural competitors for the Power 6 p570 range, and therefore it is wrong to read too much into the benchmarking (read benchmarketing) spiel that is out there.

milek said...

Ad.1 You're mistaken - I provided list prices for Sun and for IBM platforms. You're right - no one pays list prices - so niagara boxes are even cheaper in real life

Ad.2 I'm comparing configurations that IBM used itself for Siebel benchmarks - and for Application and Databse layer - so from the custtomer who needs Siebel it's a valid comparison - do I want to spent 10x more money on IBM and still get less performance or do I spent 10x less and yet get more performance. Now in different workload when you need to put much more in p570 perhaps current Niagara servers won't be good for comparison. Bu the blog entry was about Siebel benchmark.

Ad.3 Come on - for web serving POWER doesn't make sense at all. However they did put POWER6 4.7GHz p570 for Application and Database servers - and that is what interests more people as most cost and performance for entire solution is here and not at web server layer.

Ad.4 Well, Solaris is free, Sun Cluster is free - now check with IBM, even with lots cores, what the prices are for AIX and their clustering solutions - you could buy entire new servers just for a cost of their licensing.

Ad.5 You are not suggesting that IBM while doing the benchmark on purpose provided smaller numbers, are you? Be real - they did everything to get as high numbers as they could.

Ad.6 Because, regardles of what IBM's marketing is saying you, having different servers gives you more performance than partitioning one server and that benchmark is about performance (well, price/performance).

Ad.7 Moot point here. Neverthelesl it depends on exact environment. Quite often Solaris Zones approach is MUCH cost effective solution and easier to manage. Sometimes other solution is better. Anyway, not relevant to Siebel test.

Ad.8 I agree, cheaper is not always better. Nevertheless quite often people buying overprices p-series from IBM are just wasing their money.

I agree to some extent with you that comparing p570 to t5220 does not generally make sense. But then you are a customer who wants Siebel and wants performance... and best what IBM can do right now is to provide you solution which is 10x more expensive than Niagara boxes and still provides you with less performance.

That benchmark also shows that over-hyped POWER6 4.7GHz, which you can't still buy for anything else than p570, is not actually that fast in real life. Not to mention its bad price/performance ratio.

Anonymous said...

you'll be pleased to know that Power 6 will be rolled out across the rest of the P series range this quarter.

So, i think things will become clearer soon enough, but for now, i dont read too much into these comparisons.

milek said...

I agree. Nevertheless from a customer perspective they do make sense *now*. Since IBM wask talking about POWER6 4.7 for almost about a year and still hasn't delivered except one server.

Anonymous said...

Now that benchmarks for the entry level Power 6 p520 and p550 models are available, im interested in your thoughts.

http://www.ibm.com/developerworks/blogs/page/benchmarking?entry=winning_is_everything_with_power6

Anonymous said...

hmm....a no comment then ?

interesting !