Big Blue is adding to its Linux-only Power Systems server line, including a four-socket machine based on its Power7+ processors. These PowerLinux machines have a lower cost than plain vanilla Power Systems machines that are capable of also running its AIX Unix or IBM i proprietary operating system as well as Linux. IBM is also …
I must admit
I must admit there was a time I felt every IBM sells person would piss you right in the face. In fact my first introduction to AIX is one of the most theatrical/circus events I have ever experienced. Two guys dressed the same way introducing AIX to some two hundred persons so full of laughter after this circus event that I have never ever experienced anything like that before or after. Still I was offered a job as a programmer many many years ago and I have wondered about if I would be a different person to day had I accepted it. Any way IBM has changed,. it was one of the first to understand the force of Linux. Microsoft was the first to understand it. Regards to the IBM of to day.
You would have to be retarded to run anything IBM on a greenfield site these days. So who exactly buys these overpriced boat anchors with included IBM lock in??
How are you "locked in" with a server which runs all of the major Linux distros and any stack you want on top of those Linux distros? As for the price comment, see the main point of article.
The price comparisons are at list.
x86 gear gets discounted more heavily than Unix gear.
When you buy a Powr system, you have to buy into the ecosystem that goes with it - software, specialist hardware (SAN connectivity, etc.), which is also more expensive than the x86 world.
The FTE cost is higher too.
Thus, when you look at the entire system, x86 is always going to win out. And they're comparing to HP, which is one of the more expensive x86 vendors on the market, other than Big Blue of course.
Comparing to Dell or SuperMicro would offer event better price comparisons in favour of x86 over Power.
Seriously, this looks like a fair product, but price-performance is where it's at these days and a non-proprietary x86 architecture will always win out by a significant margin, both on acquisition and lifetime operational costs.
"x86 gear gets discounted more heavily than Unix gear."
Unix gear, from any OEM, is traditionally discounted much more aggressively than x86. x86 is bare bones whereas Unix, at least traditionally, had more margin to discount.
" buy into the ecosystem that goes with it - software, specialist hardware (SAN connectivity, etc.), which is also more expensive than the x86 world."
Fair point, although all of the Power components and utilities have come down significantly in cost.... and it all works together. You can buy I/O and third party software utilities from a wider range of providers in x86, but it is shoddy quality. No OEM in the x86 world has the margin, meaning staff, to do testing on the 8,000 potential configurations that people could use... so you get a lot of random I/O connection drops or 10g NICs that are running at 2g. Depends on what you prioritize.
Facts about Power
Power servers deliver more value than any x86 server on the market not to mention all other Unix offerings. Power's value prop is that it delivers performance, reliability, flexibility and efficiency at competitive prices. Long gone are the Big Iron days. Today, with IBM's Power7+ lineup they have 1, 2 and 4 socket servers that are price comparable to x86 for a similar configuration. Thankfully IBM continues to deliver a premium product to meet customers needs at varying levels. Similar to car manufacturers that offer sub-compact, compact, mid-size, large and luxury cars with varying features and price points IBM has the same in support of IBM i (good 'ol 400), AIX and Linux on servers ranging from 4 - 256 cores running at speeds up to 4.42 GHz and memory capacities from 4 - 16 TB.
Many if not most x86 customers have been trained to buy commodity servers lacking in reliability features requiring horizontal scaling in data centers. Not a big deal until you run out of power, cooling and space not to mention the FTE's required to support that. It's interesting to read the comment earlier posted about the cost of a Unix FTE costing more. I say UNIX because HP-UX, Solaris and AIX are all quality operating systems and it is often the case their platforms run a large majority of a businesses operations. I have a customer that runs 80% of their business on Power with 2 FTE's and the 20% balance is using x86 that requires 6 FTE's. This is attributed to the inherent reliability of the OS and server not to mention their ability to efficiently run significantly larger and more workloads.
IBM could play the "I'm cheapest" game like I guess Dell and SuperMicro do as stated above. If that works for your organization that is great. However, Power servers are price competitive. Where x86 requires 2, 4 or more servers for many workloads often requiring expensive clustering software you often only need 1 Power server. Some businesses demand a cluster as part of their risk mitigation strategy and with Power servers they only need two servers then. Not because the servers aren't reliable but because the business dictates that reasonable measures to avoid outages be taken. Because fewer Power servers are required the TCA and TCO continue to favor Power.
Lastly for this comment you should know that Power servers not only generally outperform x86 servers on a core for core basis but that takes a back seat to their real value which is their efficiency. For customers it isn't about the price of a $15k x86 servers when you have to buy 4 to do the job of 1 Power server. The software on those 4 servers may cost $1M compared to $250k on the single Power server. Or, in the case of EnterpriseDB there is just the support cost on a 8, 16 or 32 core Power server that I would put up against 64 - 128...even 256 x86 cores to do the same work....I've even seen it take 300 x86 (latest x86 chipsets) cores that only take 30 Power cores to do. This isn't hyperbole it is what Power servers are - inherently reliable, flexible and efficient.
Re: Facts about Power
[...AC because I consult on all vendors, all O/S's...]
Good post and worthy of a detailed read.
I have to mention some of the points you bring up
Firstly, the RAS features on a Power server such as this really don't vary much from an equivalent x86 vendor. the CPU is different, but the memory DIMMs have the same MTBF, as do the power supplies, motherboards, etc. When I see x86 vendors now discussing MTBFs on latest generation hardware of 8-11 years, this reliability question has largely gone away.
Almost all customers are building in RAS using software anyway, such as VMware, Oracle RAC, etc. which is why the market for Unix continues to decline.
Your example of a customer running a mixed shop with just 2 FTE's on the Power side is irrelevant. I have a customer with over 600 x86 servers and just two FTEs maintaining the install.
You only need two x86 servers for a cluster too - the same as Power - so what's your point?
your 4-to-1 ratio is unrealistic - even the article states that you are getting about 20% more performance from a Power system - and that's only if your application allows you to run the server at 100%, otherwise it's unusable.
The TCA and TCO will never be lowest on the Power side. If that were the case, everyone would be migrating there - but they are not. I've had countless conversations and assisted with migrations from proprietary to industry standard over the last 5 years. Not one customer I've worked with has migrated from industry standard to proprietary because the numbers just don't work. The lowest cost model is always going to be industry standard x86 platform.
I get that you have a particular point of view and work for a company that values its worth through its relationship with IBM. IBM's strengths continue to be in its software, services and channel relationships - hardware is largely a vehicle for IBM to sell software and services.
This product line was IBM's attempt to differentiate itself in a market where it is losing ground across the board in hardware: its x86 share is declining, Unix is going the same way as the mainframe and the iSeries is slowly sunsetting. IBM's storage lines are struggling against the big vendors too, if IDC is to be believed. The product line is a worthy try and will be interesting to IBM shops, but the rest of the IT industry will take a pass.
Re: Facts about Power (Memory RAS)
The memory RAS between POWER and x86 is not equivalent (POWER being better). This is especially true if the x86 system is not configured in "lock step" mode. Keep in mind "lock step" mode on x86 carries a significant bandwidth tax. I have never seen published benchmarks in "lock step". In addition, most x86 benchmarks do not include virtualization, which is always on for POWER. This adds to the POWER performance advantage once an apples to apples testa are made.
Curious, how many people out there run "lock step" on their Intel x86 boxes? Has anyone compared performance w/wo lock step for real applications?
Re: Facts about Power (Memory RAS)
Good point on Lockstep.
Lockstep memory mode caries a bandwidth tax regardless of which platform you're on (x86 or Power) because of the memory channels being used. It also adds to the cost of a system - half as much memory bandwidth for an addressable sum. Power7 has 2 memory controllers - 8 DDR3 lanes - so has twice as many lanes to use.
Performance impact? No performance impact at all if you're not taxing the box. Impact to box if it is being taxed beyond 50% of its capability on proc cycles or mem scaling, I'd wager.
Some x86 customers I know use some of the advanced DDR3 options - memory sparing, ECC, lockstep, but not many. RAS is good enough on the platform overall not to require this.
Re: Facts about Power
I think you haven't really got it.
I don't blame you, I didn't for many years either. Back in the days when I was a *NIX consultant I killed off many a bit mini or mainframe systems. But one thing I learned was that whenever you killed of a mainframe or mini the scale out system you replaced it with always exploded in size. And most often were 2-4 times bigger than originally sized.
So as a *NIX guy that today mostly are involved when it's refrigerator sized machines, I have to say to all the mainframers and mini people out there.
Sorry you were right, scale up is for a huge chunk of the workloads out there, the only way to go.
And when it comes to scale up, then IMHO POWER is most likely the best platform out there right now, but I wouldn't buy the boxes that IBM is peddling here. I'd go for the larger more expensive boxes. I know that in TCA when just looking at the hardware then these machines are hugely expensive. But IMHO they are worth it, if you have enough need for capacity.
And I must admit I think it is a bit of a joke, when people talk about commodity boxes having RAS that is in the general same range as scale up servers.
And you mention Oracle RAC as a good way of securing RAS, it sure is, but when you buy an extreeeem expensive piece of software like RAC, then you don't want to deploy it on commodity hardware, you want something that is fast and reliable.
Re: Facts about Power
Just replaced a customer's aged x86 environment with another one. They went from 29 servers to 3.
2 full racks to less than half.
Your comparisons of Unix/Minis to x86 is very 4 generations ago. the latest gen Intel stuff is high performance and for almost all the customers I work with,
RAS is just IBM Unix FUD. RAS is plenty good enough in x86 for almost all customers, unless you're controlling a nuclear power station.
More Facts about Power
Power servers are about "balanced" computing. There is a balance between the cores, memory and I/O. This is the reason a 16 core server can often (and regularly does) do the workload of a 64 core x86, Itanium or SPARC server. Power memory doesn't need to run at 1600 MHz like x86. - this is an example of the shiny object that distracts consumers to buy off the spec sheet.
Power RAS capabilities far exceed x86 in every case except for the very high end x86 servers. These high end x86 servers differ from the standard 2 & 4 socket x86 servers.
I'm rambling - I over explain stuff but try to avoid confusion. Regards
snipped by mod for brevity
Re: More Facts about Power
What you're missing here however is that scalability these days means virtualization, which means memory scalability in the largest amount of cases.
The 20% extra CPU performance is irrelevant when scalabilyt is a funcition of the memory build in a virtual environment. The market's moving ever further away from Power and this a last gasp attempt by IBM to cling onto some market share in the non Unix space.
Lot of IBM marketing in these posts
"Based on published SPECjbb2005 Java benchmark tests and extrapolating back from 4GHz processors to the 3.5GHz chips used in the above comparison, the Power machine does about 20 per cent more work than either of the four-socket machines in the table."
But fact is, you get only 20% more performance than a x86 server. But these POWER7+ servers are running at 3.5GHz, not at 2.4GHz as the x86 servers do. So these POWER7+ servers are 46% higher clocked, and still only gives 20% more performance. If you overclocked the x86 servers, you would get higher performance that the POWER7+ servers. This means x86 are faster, clock for clock. POWER7+ are not a good cpu, no. And the article states that Intel will unleash their new IvyBridge E7 cpus soon, so the "performance gap will close soon".
If you buy POWER7+ servers you need to invest in another architecture, SAN, software, etc. You will be locked in, for a measly 20% higher performance. Is it worth it? You better go with HP instead, they have superior higher clocked x86 servers beating these POWER7+ servers. Even Itanium will beat P7+ under right circumstances. And where is POWER8? IBM is loosing it. The price of POWER servers are lower than ever. Soon POWER will not carry the cost. IBM is transitioning to a services company, hardware is going down the drain.