IBM hosted a briefing earlier this week to talk about some new members of its PureSystems offerings, but only one of the lineup they introduced was truly new – and it took The Reg a couple of days of digging to uncover pricing information. The "Troy" Flex System modular servers from Big Blue and their PureSystems stacks for …
DIY wins, when it comes to hardware
The longer I have to read these hardware offers, the more I want to take out my screwdriver and build those servers myself. Why should I leave the self-made server to Google?
I once made a calculation where I found I can build a 16 core Xeon E5-2670/256GB memory/2TB SSD/20TB HD server myself for £10K - delivery and labour included.
Let's assume one of these manufactured servers cost £300K after one year - that's including hardware, external maintenance contract, delivery and installation, license fees, SAN etc., and let me tell you £300K is a conservative estimate, you can easily pay £2million.
For £300K I can build 30 servers - that's 480 cores, 960 threads, 7.5TB memory, 60TB SSD, 600 TB storage in total, and I haven't mentioned memory and CPU overallocation yet, but I need to include clustering and hot sites.
Can any of these premium hardware manufacturers give me the above processing and storage power, duplication and high availability for £300K?
I have to pay for monitoring tools, virtualization and Linux or Windows, databases, etc. in the VMs either way. This leaves the hardware maintenance. What if something breaks?
Well, I take out my screwdriver and replace the broken part. That is me, or my colleague, and we do it on the spot and we don't need to wait for some "home call" or an engineer callout. And we go onto the open market to get the best deal and don't need a vendor specific replacement. And I can shred the part to make sure no sensitive data gets taken away. There shouldn't be a loss in availability, as the reduced costs for the hardware allowed me to design the system with resilience and duplication in mind, and I've freed myself from a vendor lock-in and increased security.
Why should I leave the self-made server to Google?
Re: DIY wins, when it comes to hardware
> Why should I leave the self-made server to Google?
Because you are not Google.
I understand the concepts of "overestimating oneself" and "underestimating the time/money/effort/debugging between where one is now and where one wants to get by at least an order of magnitude" because I have been burned too many times.
These problems are out there. They will bite you. And then your whole rig crashes and you don't know why.
Did you include your salary and that of your colleague (and probably the other colleague who is on 24/7 call), btw?
Re: DIY wins, when it comes to hardware
"Because you are not Google." - so only big and great companies can do the DIY, but flexible SMEs may not build their systems themselves, although there is a lower level of complexity?
"underestimating the time/money/effort/debugging" - "And then your whole rig crashes and you don't know why."- yeah, you're right there, the many times I was better than a vendor's 3rd level support, and their standard reply "uh, maybe you upgrade to the latest version first and call us back?" Like I can do upgrade and testing all accross from sandbox to pre-prod to prod systems all in one night.
"Did you include your salary and that of your colleague" - We sysadmins are there anyway,regardless where the hardware comes from. If we build our servers ourselves, then there are only internal maintenance costs; we can catalogue which components are deployed where, and why we have built it this way, with what firmware/hypervisor/software level, and have a detailed CMDB from the start.
With vendor supplied hardware there are internal and external maintenance costs and we have no clue. We have to pay for a discovery tool to fill the CMDB and close security holes.
Pure data for transactions
We are looking at the Pure data for transactions solutions. Our company has parallel sysplex on the mainframe and have been looking to implement that in the Unix environment, but given the effort to implement a pre-integrated systems would be perfect. Interestingly, the name for the technology on Power was called PureScale. Looks like they used the Pure name for everything but have abandoned the term purescale. It will take forever but we are looking to dump oracle RAC.
Re: Pure data for transactions
Whilst it looks cheaper empirically to do it yourself, as an experienced infra/DW architect who has been through this several times before I can tell you its rarely that straight forward.
Firstly, the H/W costs are the tip of the iceberg - the software costs are thos which really bump up the price. Ive only even seen one client pay list price, but generally you can wipe 30% off that for starters.
Secondly, having access to free open source software is one thing, having the skills & experience to build that into a tuned, fully integrated and manageable solution is another. Dont underestimate how difficult that is.
if we do it your way ive got one and a half tons of iron, silicon and cabling splashed all over the datacentre floor, with probably a 12 month wait to get something that actualy works - while my business splashes cash on resources, hosting etc & gets no benefit from it - potentially costing them millions of pounds in the process or.....
yet anonther name. change
What is the root of IBMs obsession with rebranding and renaming.
It would not be so bad except their web site and product documentation lags several years behind. You need to remember what a product was called three years ago when searching for the reference manuals.
My favourite was when good old IEBCOPY, the worlds worst surviving copy utility got rebranding as Tivoli.
- Infosec geniuses hack a Canon PRINTER and install DOOM
- Boffins say they've got Lithium batteries the wrong way around
- Phones 4u slips into administration after EE cuts ties with Brit mobe retailer
- In a spin: Samsung accuses LG exec of washing machine SABOTAGE
- Game Theory Half a BILLION in the making: Bungie's Destiny reviewed