Dude/Dudette, you're the one that said why not just buy over-provisionned run-of-the-mill x86 servers ? Say, 20 quad Xeon Dell machines ?
Please correct me if I am wrong. These are just opinions, I do not know the actual price of these Unisys "mainframe", so I might be totally wrong on the price/performance ratio. I just think it is impossible for proprietary hardware to match the ratio of commidity x86 stuff.
So I said to you that it costs about 15% more or less then then like commodity boxes would be per CPU. And by the time you over-provisioned (is it free to overprovision? Lets add 15% as the overprovision overhead) your commodity boxes the price would be very similar. So if you think 20 quads would do like work then the comparison is about 80 CPUs vs. a 32 cpu system. That's a bit of ballpark guestimation involved isn't it.
On the price, you got that right, not NDA, just a bad policy to say what you spent because you may not get a good deal next time if you go about blabbing price without consideration for the full package of what you buy - I don't talk about the price because on enterprise systems no one pays retail and the size of the discount on system is contingent on buying other things like services, storage, maintenance and whatever else is part of the solution (and what other vendor is bidding the solution so that I can clobber one vendor with the other).
Nothing at all to stop you from investigating the price on your own rather then speculating. Here's how you do it: Most of the large vendors have government contract schedules for hardware, software and services that are posted on the web. IBM, Sun, Unisys, Dell - most of them are out there so you just have to execute due dilligence and review them. The contract price of the hardware and services is in the published PDF documents that vendors provide to state government purchasing agencies. You can use those prices as a way of familiarizing how low the vendors would go on their hardware.
The last bit of consideration is this: If you like buying 20 overprovisioned servers and those servers give you a warm feeling, then by all means that would be the right idea. But if your application would NOT be helped by scaling out (because you have to redesign your database and rewrite the front tier software) then the scale out model may not be the right approach.
That's where the niche players are useful They can hide the problems of your software infrastructure for a few years with their hardware hooks and give you further breathing room while you are rewriting those apps to take advantage of your scale out strategy.
I know that the majority of datacenters have great, well behaved applications that last into perpetuity and don't need problem masking, but I'm in an odd case where the software is in constant need of maintenance and if that software throws up the revenue stream dries up.
I don't have an issue with anecdotes (as long as they're somewhat believable) and I don't mind reading them because I don't expect that threaded messages are authoritative in any remote way. I wouldn't use this thread as a reference in ANY purchasing decision. It's just interesting to hear what other viewpoints are - that's all.
I'm done on this post, but I am interested to see what the heck this "propriatary" aspect of the 7000 is that people refer to. I'm thinking they mean not commodity boxen, but neither are AIX boxen or Sun boxen.
They're worth about as much as people would pay for them - they started high, came down lower, and very few pay retail. I have several of these machines (BTW 3/4 of the gibberish about the performance in the thread is wrong and thirdhand) along with a sizeable number of quads that do ancillary services.
What I know is that I had a choice between Sun, AIX or Unisys. I chose to use the one that that was 40-50% less compared to them. Also looked at 440.
Costwise depends. If I was buying a good quad from a respected quad manufacturer (i.e. Dell or HP - no IBM thanks) the price per CPU on the 7000 is a little more (maybe 15% more) but the commodity boxen certainly would not do the work that was required. I needed above 4x performance and they didn't have anything worth purchasing. My only option was Sun or IBM for scalability and those involved RISC. No thanks, if I can stay away from RISC, I prefer to.
Your question about 32cpu mainframes makes it sound that you don't know that they have new models that don't require the mainframe "frame" (refrigerator box as someone called it in another post). They come in 8 processor cells similarly to the IBM 440 series. And they're priced about the same as those.
So pricewise by the time you purchased three commodity quads, you could get an 8x cell and a bunch of services to integrate it, and pretty good services. Of course all services are regional so you may run into some dogs but my experience has been positive. If your application was suitable for scale out the three quads would have been feasibile. In my case it wasn't.
Other gibberish I noticed in the posts. It's not NUMA. It's just not Profusion based so that confuses people.
Slashdot Mirror
Dude/Dudette, you're the one that said why not just buy over-provisionned run-of-the-mill x86 servers ? Say, 20 quad Xeon Dell machines ?
Please correct me if I am wrong. These are just opinions, I do not know the actual price of these Unisys "mainframe", so I might be totally wrong on the price/performance ratio. I just think it is impossible for proprietary hardware to match the ratio of commidity x86 stuff.
So I said to you that it costs about 15% more or less then then like commodity boxes would be per CPU. And by the time you over-provisioned (is it free to overprovision? Lets add 15% as the overprovision overhead) your commodity boxes the price would be very similar. So if you think 20 quads would do like work then the comparison is about 80 CPUs vs. a 32 cpu system. That's a bit of ballpark guestimation involved isn't it.
On the price, you got that right, not NDA, just a bad policy to say what you spent because you may not get a good deal next time if you go about blabbing price without consideration for the full package of what you buy - I don't talk about the price because on enterprise systems no one pays retail and the size of the discount on system is contingent on buying other things like services, storage, maintenance and whatever else is part of the solution (and what other vendor is bidding the solution so that I can clobber one vendor with the other).
Nothing at all to stop you from investigating the price on your own rather then speculating. Here's how you do it: Most of the large vendors have government contract schedules for hardware, software and services that are posted on the web. IBM, Sun, Unisys, Dell - most of them are out there so you just have to execute due dilligence and review them. The contract price of the hardware and services is in the published PDF documents that vendors provide to state government purchasing agencies. You can use those prices as a way of familiarizing how low the vendors would go on their hardware.
The last bit of consideration is this: If you like buying 20 overprovisioned servers and those servers give you a warm feeling, then by all means that would be the right idea. But if your application would NOT be helped by scaling out (because you have to redesign your database and rewrite the front tier software) then the scale out model may not be the right approach.
That's where the niche players are useful They can hide the problems of your software infrastructure for a few years with their hardware hooks and give you further breathing room while you are rewriting those apps to take advantage of your scale out strategy.
I know that the majority of datacenters have great, well behaved applications that last into perpetuity and don't need problem masking, but I'm in an odd case where the software is in constant need of maintenance and if that software throws up the revenue stream dries up.
I don't have an issue with anecdotes (as long as they're somewhat believable) and I don't mind reading them because I don't expect that threaded messages are authoritative in any remote way. I wouldn't use this thread as a reference in ANY purchasing decision. It's just interesting to hear what other viewpoints are - that's all.
I'm done on this post, but I am interested to see what the heck this "propriatary" aspect of the 7000 is that people refer to. I'm thinking they mean not commodity boxen, but neither are AIX boxen or Sun boxen.
ttfn
They're worth about as much as people would pay for them - they started high, came down lower, and very few pay retail. I have several of these machines (BTW 3/4 of the gibberish about the performance in the thread is wrong and thirdhand) along with a sizeable number of quads that do ancillary services.
What I know is that I had a choice between Sun, AIX or Unisys. I chose to use the one that that was 40-50% less compared to them. Also looked at 440.
Costwise depends. If I was buying a good quad from a respected quad manufacturer (i.e. Dell or HP - no IBM thanks) the price per CPU on the 7000 is a little more (maybe 15% more) but the commodity boxen certainly would not do the work that was required. I needed above 4x performance and they didn't have anything worth purchasing. My only option was Sun or IBM for scalability and those involved RISC. No thanks, if I can stay away from RISC, I prefer to.
Your question about 32cpu mainframes makes it sound that you don't know that they have new models that don't require the mainframe "frame" (refrigerator box as someone called it in another post). They come in 8 processor cells similarly to the IBM 440 series. And they're priced about the same as those.
So pricewise by the time you purchased three commodity quads, you could get an 8x cell and a bunch of services to integrate it, and pretty good services. Of course all services are regional so you may run into some dogs but my experience has been positive. If your application was suitable for scale out the three quads would have been feasibile. In my case it wasn't.
Other gibberish I noticed in the posts. It's not NUMA. It's just not Profusion based so that confuses people.
ttfn