Many would have argued in the past that a data warehouse was not a "real" data warehouse unless it existed the in big iron mainframe number crunching environment. A bit of a generalisation maybe, but when you look at where many of the big banks, insurance companies, retailers and telcos stick their transaction data for analysis …
SMP vs MPP all over again?
I've seen in the SMP vs MPP wars from both side - the hardware vendor, and the developer/implementor that need to make the technology work and give the customer that bottom line RIO.
Personally, I fail to see the deal with SMP or mainframes. No mistake, dating from the 80's, I loved mainframes. Still has a lot of affection for them. They allowed us to do amazing things on what were very primitive hardware.
But today, it is all about commodity hardware from a management view point. It is a lot easier getting management buy-in and approval for purchasing a bunch of small (Intel/AMD) servers, than what it is trying to motivate for a big iron.
And in most cases, that MPP platform is a lot more scalable and redundant than a single piece of big iron.
Case in point. All the hardware purchases we have made the last 3 years were small (AMD) servers for MPP. These are replacing the aging big iron Unix hardware boxes we have.. which in turn replaced the mainframes that used to be around a decade ago.
The actual problem that I've experienced is software pricing. Using commodity hardware and the associated reduction in hardware expenditure results in management expecting the same from the software. They expect (for historic reasons likely) that the software should not be more than the hardware.
The Mainframe still rules!
Not only is the IBM mainframe the best Data Warehousing platform - why have racks and racks of Linux servers all waiting to go wrong when you can replace them all with one IBM mainframe running hundreds of Linux images? Virtulisation is the future and bullet-proof 99.9999% reliable IBM mainframes are the perfect platform.
It goes deeper than that
The non functional capabilities count for a bit too !!!
If the warehouse is actually critical to your business that is. Or if it becomes crtitical down the line....how many times is that little aspect 'forgotten'...
Platform selection somes several steps down in the design. We all know this.
Initial costs not too relevant in my view but if they are a hot button for the company how about a benefits analysis ?
e.g. Just the cost differential between it being subsumed by existing DR on the mainframe compared with setting it up, testing it and making it actually work on the Intel path would be how much exactly ? ?
Lots of other e.g's possible... FTE supporrt costs, availability.. all the usual stuff.
Then again how many companies actually have a repeatable process for making these decisions that can be improved
as time goes by ? Not many at all - despite ITIL. or maybe because of it... but that is a whole other story
Plus the power bills, cooling and sheer storage space
One z990 can easily do the work of 750 or 1000 PCs - it'll and consume about 7.5-10% the electricity, and generate about 10% of the heat, while doing so. It also atkes up about 5% the floor space of the equivalent computing power, in the form of small iron.
1000 PC servers cost about as much as an entry level z-series, anyway, and if you consider the component failure, on 1000 PC servers, you can see that it would be a nightmare. Given the costs involved in storing, securing, cooling and powering 1000 PCs, compared to the costs of doing the same thing with a Big Ugly Fridge, unfortunately the Big Ugly Fridge wins it, easily.
It's notable that Google, which has long been the poster-child of the 'do it all with PCs' approach, have recently started building their own customised servers, to manage their work, lately. They could probably use a few mainframes, inf act, but they're far too secretive to start commiting their business into any kind of longterm relationship with the likes of IBM.
Horses for courses
There are pros and cons for each configuration. I have worked in both environments and found that they work equally well as Data Warehouses. I like DB2 as much as Oracle or MySQL. SQL-Server? Er... it has its uses and place.
Having multiple Linux servers is not a bad thing. They can be made bullet-proof as high-availability systems with mirroring software like Heartbeat which in my experience has given 100% reliability to the extent that a machine can be taken down for maintenance without users even noticing.
The problem is not so much the physical configuration but the software. Users of the very popular operating system from Redmond will generally or can only use software by them - ie. SQL-Server and Analysis Services. Third-party Warehouse providers are forced to work with these and it is the inadequacies of the software and operating system that gives rise to the impression that MPP platforms are not as good as mainframes.
Mainframe now obsolete?
Quote: "...the cost of these [commodity hardware platforms and the database management systems that run on them such as SQL Server, Oracle and MySQL] is a fraction of the premium you would pay for the mainframe equivalent, and that things like performance, scalability, manageability, and security are now generally on a par."
I dispute this statement, every little bit of it. Mainframes win hands down on any of the measures you point out.
Here's an example on the Manageability (which imo includes reliability) side - a Z series mainframe with more than one processor unit (called a book), say one book fails, it can be hot swapped with no downtime. These books can have their microcode updated on the fly (they are a kind of soft-configurable processor), i.e. no downtime. I/O racks can be hot swapped in their entirity, again with no down time. This leads to a >99.9% availability.
Would you show me a non mainframe system that actually is that robust please?
I also believe that the TCO for a mainframe is competitively comparable with the TCO for a "commodity hardware platform" that delivers the same performance. If you doubt it, please check for yourself (my info is from Gartner as well as IBM)
I state again that I believe you to be severely mistaken in pretty much every aspect of your assertion, which, as it forms the central plank of your contention would seem to leave you without a leg to stand on...
Note that I didn't stoop so low as to point out that you included at least one Microsoft product in your list - ROFLMAO! and in the same sentence as the word Security - is it April 1st again?
Where are the benchmarks?
Two thoughts on this:
1) I haven't seen any TPC data warehousing benchmarks vs. Oracle / Teradata / others. If the mainframe were truly faster and cheaper, they'd be publishing objective data. Show me the numbers!
2) A mainframe can only replace hundreds of PCs when the PCs are not busy. Consolidation and virtualization is great in this environment, and Xen / VMware / Solaris containers can also do this. There's no magic about a mainframe CPU, it's not really any faster than an x86 chip (I've benchmarked both).
"Note that I didn't stoop so low as to point out that you included at least one Microsoft product in your list - ROFLMAO! and in the same sentence as the word Security - is it April 1st again?"
A Touch of SunStroke, methinks.
Why don't you look at TPC-H Benchmark?-SMP still beats MPP on performance AND price/performance
I know there are no IBM mainframe benchmarks but I noticed Sun posted a new world record on the 3 Terabyte scale factor for TPC-H. The Sun Fire E25K SMP Unix system has beaten the best HP ProLiant BL25p clustered Linux system in both performance AND price/performance and as far as I can tell, the pricing includes software and hardware costs running Oracle. You can go to here to see the results: http://www.tpc.org/tpch/results/tpch_perf_results.asp
Horses for courses, methinks
Perhaps the real question is one of task. Mainframes excell at batch work, and always have. When a bank's entire business can be set up in a stack and constantly processed, 24/7, in a carefully managed queue that ensures the system is working flat out all of the time, then the throughput can be truely staggering.This is where mainframes wipe the floor with the opposition.
When you try to run one as non-batch server, however, doing ordinary work, as a result of realtime requests from users and processes that need immediate execution, on a non-predictable time schedule, then their competative advantage drops away quickly. I mean tasks like file serving or webserving.
Mainframe to Server Conversion
I dont run a really huge server so it is hard to say. I have seen pro's and con's to both environments because I have used them both, but just not on the scale of a giant corporation.
I work for a community college with an average of 12,000 students in attendance over a large area if you consider the Satelite campuses i.e. Community Education Centers spread over multiple counties. So there is this need to be accessable and secure. The Mainframe can do both. The problem is that for a medium sized business that the software and operating system and hardware licenses for a mainframe are somewhat prohibitive to ownership at the level we want to be at. However, this can be an illusion when you look at the big picture.
We just went from a Mainframe Environment to a server based environment. The reason for this was that the technology that was available was more tailored to the server environment and the fact that the product we wanted to purchase was not available on a mainframe. The type of database we wanted to use made a mainframe environment cost prohibitive and it just did not fit into the system model.
Others have claimed that a mainframe can do the work of hundreds of PC's. I say that is pure baloney. Even with the mainframe everyone in our organization still had a PC so there were no significant savings in that department. In fact at the time of the conversion, we were emulating the Mainframe on a Linux PIII Server, that had been customized to interface with the IBM Printer, and Controller. At best we were a low-end Mainframe user.
So we decide to install this new database, a prepackaged Application to run it, along with some third-party products. The nature and complexity of the new database design would have been next to impossible for us to duplicate by our 3 man IT department that supported our existing mainframe. We had a couple other people that did support like a Project Coordinator, and some Network people. However, a project encompassing every aspect of our business model from accounting, asset control, E-mail, Student Systems, Financial Aid, etc., kind of ballooned in size and cost.
Now we have a new-fangled database that is so large we are looking at about 2,000 files. It takes a Database administrator, about 7 System Analysts, and multiple Project managers to get this going. The new product requires traing funds that we did not use to spend and travel to training locations at a cost of around $1800 per person per class. The software keeps changing and we keep having to spend more training funds. The upgrades from one version to the next adds in support for SQL servers that requires and upgrade to the Database version as well. So the costs keep mounting. Database licenses and Sql server licenses, are not cheap. However, every few years we would have had to upgrade the mainframe anyway, and our software was so old that we would have had to find a replacement anyway.
So we did not really save any money. The real costs are not in the hardware or the operating system. The real costs in a conversion are in the devolopment and startup. We could have went to a larger mainframe and a more expensive operating system, installed DBMS and had some professional system analysts come in and set it up from scratch. However, it would have still probably have cost more and taken longer to accomplish. We still might have to go that route because we chose the cheaper database product.
MainFrame was never good for BI and DW work
The mainframe is still "king" in the OLTP market. If you look at systems that require a high throughput of transactions and security, those are on the mainframe. The NYSE is a good example. The primary trading system is still on the mainframe because it can deliver good throughput and incredible uptime.
Don't get me wrong, I'm really a Unix bigot, but you have to give credit where credit is due. The UNIX servers are not up to snuff.
But the mainframe isn't good for handling BI type queries or really ad-hoc queries. You have cost issues as well as a potential for the BI and DW work to impact on the OLTP applications.
So you have companies that do OLTP on the mainframe, and DW on Unix.
Nothing new there. This has been the model for many fortune 500 companies for years.
But there is a cost associated with running a mainframe. If you don't need the high volume OLTP and the security/uptime then you don't need a mainframe.
To one poster's point, trying to equate the mainframe to pcs is bullox. It doesn't make sense.
If we look at high end equipment from SUN/IBM/HP you are seeing SMP boxes that tend to mimic features that were once only found in a mainframe.
Being able to partition a Unix box in to acting like multiple boxes and distributing processors as needed, you have an environment that can replace the mainframe.
So while the mainframe isn't "dead", there are other more effective cost alternatives for SMB sized companies that can't afford the price tag of owning a mainframe.
Horses for Courses - revisited
A previous respondant said: "When you try to run one as non-batch server, however, doing ordinary work, as a result of realtime requests from users and processes that need immediate execution, on a non-predictable time schedule, then their competative advantage drops away quickly. I mean tasks like file serving or webserving."
I disagree with this - mainframe systems exist that serve 10000+ users each submitting real-time requests for immediate execution, often exceeding 50,000 transactions per second - an example of this type of installtion is a large or internantional Bank (yes, one was one I worked in).
Data "wherehousing" for large companies
In a large Pharma multi we have been living with change since Oracle 7 and client/servers challenged the mainframe. Applications and infrastructure have become pretty much independent over the last 20 years.
Oracle itself solved several of the problems with replication, distributed servers and OLAP (I can't remember who they bought out to do this). This enabled us to build distributed global applications.
The most important solution in our firm this decade has been a global SAP rollout - SAP comes with a business warehouse (BW) which is quite useful for extracts and reports from different SAP instances. It also, incidentally, solved a large weakness of the SAP mantra, based on the premise that non-SAP applications don't exist, or at least shouldn't.
Now, with BW, we can produce reports which show that every factory and sales outlet works differently. Before, the differences were larger, but we were not aware of them.
BW is, of course especially, strong on financial reporting - that at least could be unified over different sites.
SAP need not be a distributed installation, but allows the above-mentioned freedoms, so it appears to act as one.
The machines themselves are bigger than PCs, but I would class them as midframes - I'm not even sure that classical mainframes survived Y2K - perhaps supercomputers are the nearest thing today.
What the chicken farmers don't say.....
What I've noticed is that whenever someone attempts to sell you on the idea that 1000 chickens is better than 1 ox to pull your load, they never mention the fact that it takes a considerably larger enclosure and power to house 1000 chickens, whereas the OX only takes up one stall and eats far less.
I've seen the p690s come in and the blades in a shop which is primarily z/series based - we're now out of both space and power due to the exorbant infrastructure requirements created by these "modern" platforms.
Sad thing is, they could be running 10,000 virtual linux "chickens" on the z/series box, and save a hell of a lot of money.
Hello? Anyone out there? Am I alone?
One of your respondents said: "I'm not even sure that classical mainframes survived Y2K - perhaps supercomputers are the nearest thing today."
Um, what planet would this be on?
The Mainframe is alive and well, thank goodness - despite the best efforts of the wilfully ignorant new technology zealots whose ill informed opinions are liberally splashed about these days.
This person's use of the phrase "classical mainframe" reveals a belief that the mainframe is some kind of relic from a bygone age - a belief that would seem to be as out of place as faith in homeopathy, astrology, alchemy and a flat earth.
Go on, throw out the old, replace it with the new, everyone knows that shiny, ill thought out, badly designed, poorly built, over hyped, over priced, hard to maintain, prone to failure new technology offerings are clearly superior!
Even if your readers get it wrong, surely you should be preventing the spread of such misinformation? Or are you offering them up for mockery?
Why do we post comments we may not agree with?
I think the mainframe still exists - although I also wonder if the distinction is worth making. A mainframe is simply a computer using an architecture that doesn't change every month; and which handles multiprocessors and I/O properly.
I remember my first job in journalism after I'd walked out of a bank - the news editor refused to believe that mainframes still existed, despite me being sure I'd been using one a few days before.
But we don't post comments to attract mockery. We post them to invite discussion (as has happened in this case). We don't censor comments generally - except when they don't make any sort of sense; are dangerously wrong or spamming irrelevant URLs; or when they libel someone...