RE: {Spam?} RE: oracle has customers over a barrel

> Date: Tue, 15 Sep 2009 14:06:31 +0100 > From: obnoxio@serendipita.com > CC: informix-list@iiug.org > Subject: Re: {Spam?} RE: oracle has customers over a barrel > > Ian Michael Gumby wrote: > > > > Mainframe MIPS are horrendously expensive, not to mention that the odds > > are that the customer hasn't upgraded his mainframe hardware in years. > > (Also expensive.) So Moore's law works against you on the z as well.... > > I'm not talking about existing MF customers. I'm talking about people > who have loads of Intel racks who want to consolidate. > Ah, but you didn't say that. This is why there's always the caveat of ymmv. ;-) Even still I'd still look at the price comparison when you consider the cost of everything. Looking at the Inel 2 socket quad core, if you were running windows, you could get 15-18 servers virtualized on one box. (Ok but we don't like windows but that shows a 15:1 compression ratio.) So you can put 48 servers in to one rack. Thats 24 1U high boxes which contain two 1/2 sized mother boards. See: http://www.supermicro.com/products/system/1U/6016/SYS-6016TT-IBXF.cfm Note: we didn't even talk about blade servers... The key point is that you can increase your power and density in to a smaller foot print. If you're not keen on disk speed you can use SATA or you can use a mix of SATA and SAS along with some SSD to get killer performance. I saw the article on Adaptec's new disk controller that has either a mini board or a connector to allow for an SSD to be used for page caching. (Why they don't use RAM instead is beyond me, but I guess its a cost factor.) > > IBM has recognized this and reduced the price of some of their MIPS but > > not all of them. Not to mention they are in a lawsuit with a company > > which allows one to run things like DB2 (read expensive MIPS) in a less > > expensive partition on the mainframe. (Thus protecting their revenue > > structure...) > > Linux is one of the cheaper MIPS. > True. Here's a link to an article on this topic.... http://www.theregister.co.uk/2009/08/17/ibm_mainframe_linux_cuts/ But still, does it make sense to go to the pig iron? I think you have to really run the numbers. Building custom white boxes may make more sense.... > > Also, you have to consider the fact that with Linux boxes you can > > distribute your application servers, web servers and database servers in > > to multiple safe silos. You really can't do that with your mainframe. > > This would allow for better fault tolerance and load balancing. If you > > cut the dark fiber running in to your machine room, you lose your > > mainframe. If your system is distributed, you might get bogged down by > > the loss of a node, but your other servers can pick up the slack. > > (Assuming that you built in the redundancy using ER and setting up > > 'live-live' peering. (Of course you could duplicate your mainframe for > > disaster recovery, however you better not use IBMers (IGS) to do this > > because I do know of one customer where they never tested for a > > catastrophic failure... ;-) Also be prepared to fight IBM on costs. The > > secondary system would have to be in a stand by configuration only. If > > you went live for load balancing... the costs get doubled. > > The costs get doubled? If you have two machines, each doing half the > work, the costs get doubled? I think you're a bit confused. > Uhm no, I don't think so. Look at your software licensing costs. If you create a secondary for disaster recovery, I don't believe you're charged for the licenses, or its at a discounted rate. If you make it a live peer to peer , then you are paying for two mainframes, and you have to have two locations that you have to maintain. With a distributed Linux environment, depending on your size, you can put 48+ servers in a rack + 1 rack for your disk. You can probably lease 2 racks in a shared data center for less than the cost of housing your own mainframe. I didn't look at the latest round of blade servers, but I think the density is even greater. Even if you don't scale 1:15, but even at 1:2, that's 96 virtual machines to run web, database, etc ... in a friggin rack. Mainframe aside, that's a shit load of computing power. Looking back to the mid 90's ~15 years ago, that would mean you had more horsepower in a single rack that you did in a large room filled with an SP2 running DB2. (TransUnion Dare project). To be fare, you could just put in 24 servers in to rack and still pound the crap out of the SP2. And that's pretty fscking amazing. So you really have to have a lot of intel boxes sitting out there. And with respect to the server consolidation, you have some examples where clients had gazillions of intel boxes running SQLServer that when they switched to IDS on Linux, they were able to reduce the number of servers by an order of magnitude. Thats a 10:1 ratio in reduction. (I'm not sure that it meant new hardware or repositioning the old boxes with both an OS and DB change. Go talk to Scott Pickett on that one. > And who gives a fuck about your one customer? I can show you fucking > hundreds who made the exact same mistake with Intel tin. Does that > invalidate your argument? > No. The point was that if you consolidate to a single mainframe on a single campus, you have some uptime risks. While the box may be up with 6+ 9's (99.9999%) uptime you can still have an outage because of a schlub wielding a backhoe and ignoring the cable markings. In this case (And I think it was also a customer of yours too.), the customer had redundant mainframes and servers. The bad thing is that this was a bank's credit card processing and it was in the period between Thanksgiving and X-mas. Can you say millions of dollars in potential damages and litigation? > > The truth is... you first have to figure out how much do you *have* to > > spend on a build out and what your performance metrics are before doing > > this and yes... ymmv so that in some cases a z box may win out. My gut > > instinct tells me that for most legacy apps still on the big iron, you > > can port them to Linux boxes and get better performance for much less money. > > Your gut may be wrong. The more boxes you have to consolidate, the > better the picture is for Linux on z. > Maybe. Maybe not. Don't get me wrong, I'm not anti z. I just don't see it except for potentially extreme cases and even there, I'm not sure that if you consider all of the costs, it works out. > > And there's another factor... Suppose you can replace your mainframe > > with hardware that will fill one rack, maybe two... > > Is it cheaper to lease/rent space in a data center than to build your own? > > Suppose you can, well, then you should. But how about if you could > replace 250 rack servers with a single mainframe? I work in plenty of > IDS sites where they don't have a lot of database servers, but they do > have a gazillion app servers and web servers and other servers ... all > running Linux. Then the picture looks a bit different. > No, not really. You would have to do a pretty serious exercise of identifying the app servers, loads and see what can and can't be consolidated. Also you have to consider distribution of app servers, web servers and database servers to see if it makes sense to distribute the work in to smaller foot prints and