Thinking of following Facebook and going DIY? Think again Microsoft is doing it, Apple is doing it – so is IBM. The giants are spending billions of dollars building fantastic data centres. But what about the rest of us? Do you walk in the footsteps of the giants and Do It Yourself (DIY) or buy something Commercial, Off The Shelf (COTS): it's an ages-old debate. The former demands …
The Fast Show was (IIRC) the first sketch show in Britain to (heavily or almost exclusively) feature recurring characters. While it was entertaining I think the combination of nostalgia and rarity (watchable comedy was rocking-horse shit rare back at the time) has inflated its status over the years.
While I agree with much of the story I can't agree with PC gaming being a niche market. A quick googling for some some figures seems to say that it's a bit less than half the size of the console market. Considering there are two main players in the console market that makes PC about the same as XBox or PS and I wouldn't call either of those niche.
Saying that prices have crept up is cherry picking facts slightly as well. Sonic 2 and their ilk cost about £50 from Woolworths in the 90's. Years after this Sony slashed prices on PlayStation games and prices have steadily crept back up after that point, but we're still shelling out roughly the same as we happily parted with for 16-bit titles decades ago.
for anyone who is even facing this question that is not at a big org.
I had clueless management a few years ago at another job, they wanted to a build a 140+ node hadoop cluster and wanted to do it 'on the cheap' by going with some random whitebox supermicro shop. When that decision was made I left the company within 3 weeks(the proposal I had was higher quality gear with NBD on site support I knew there would be a good amount of failures).
I was expecting bad things to happen but the shit really hit the fan it was great to hear the stories. Before they even received the first round of equipment the crap vendor they decided to go with said they had to be paid up front they didn't have the capital to front the customer to buy the equipment. Not only that the experience was so bad the new director of ops (hired a couple months after I left I never knew him) said they would never do business with this company again (this is before the first system landed in their data center).
So they got the gear and installed it. As I suspected they didn't burn anything in up front, so for the following I'd say at least six months they had roughly 30-35% of their hadoop cluster offline due to hardware failures. Because of the quorum hadoop operates in the head hadoop engineer told me basically half the cluster was down for six months while they tried to resolve stuff. It was HILARIOUS. How much business was lost as a result of trying to save about 20-25% on the project budget?
While the project was underway the CTO of the company said they would just hire a college intern to "go swap failed hard disks", as a response to my wanting on site support. As recently as a year ago they had not ever hired any interns to swap hard disks. I knew the quality was going to be crap and we weren't (at the time) staffed to support that. What actually happened proves to me I was right on every level all along, made me feel damn good.
The director in question quit within a year and the VP quit not long after(I believe he was on the verge of being pushed out). The technology side of the company basically fell apart after I left, they hired 9 people to try to do my job but still failed. Two of my former co workers BEGGED to get laid off(so they could get severance packages) when the company had a round of layoffs a few years ago, both got their wish.
Then they went the exact opposite, from super cheap to super expensive deploying a lot of Cisco UCS for a low margin advertising business. Not sure how long that lasted last I heard they refreshed everything onto Dell again.
I do miss that company for one reason though, the commute was literally about 800 feet from my front door. I had co-workers who parked further away than I lived in order to save on parking fees.
Now I live about 1 mile from my job, not too bad still.
I haven't worked for anyone with that little financial sense in quite some time. Trying to run Windows Server on a 30GB SSD with a projected install time greater than 6 years is called "being penny wise and pound foolish".
Examine the total cost of ownership over the life of the unit. This includes the cost of maintenance, downtime for upgrades, spares, electricity, bandwidth and so forth. I have spend my entire career as an SMB admin for the cheapest people alive, and I promise you that the sort of nonsense you advocate in that regard is far more costly over the life of the unit than simply buying a "sweet spot" drive and letting your OS grow.
I know it's very hard for some people to factor in the cost of manpower. THey think that being on salary makes their time cost nothing. For sysadmins this was true 15 years ago when there wasn't such a diversity of products to support. Now, the sheer volume of different kinds of hardware, software, networking etc that even the smallest of SMBs must support strains the ongoing maintenance capabilities of even the most strongly "pee in jars" sysadmin.
Try to make what you deploy as "fire and forget" as possible. That will require frontloading a few extra % in terms of hardware cost in order to recoup hundreds of % in operating costs.
Just finished one project where I did manage to get the hardware we needed without any fuss. Part of the justification was when I told the sponsor that the best hardware in the world would still be cheaper than the cost over run when he changed his mind because he hadn't understood the spec for the application (even though it had been explained, endlessly, as to what it did and did not do).
Project that is just about to go live we have outsourced to AWS. Best idea with hardware is to make it SEP.
I just bought a load of the cheapest 30GB SSDs I could find, but they'll be going into PoS terminals running linux, so I'm not worrying about the capacity.
I do however still harbour great hatred for whoever first propagated the idea that a Windows server only needed a 15GB c:/ drive, and that it was a good idea to partition it as such (rather than just using the whole disk). I can't count how many times I've had to try and screw every last byte of free space just to keep an old server from crapping out.
I don't even understand how you can run a modern Windows on a 15GB C:\ drive. 40GB VDI instances floweth over in 8 months. 75GB instances only go about 3 years before needing WinSXS purged. Server instances seem to be in the "80-100GB OS disk" range for usable installs of 3+years.
I sort of get the guy who can use 128GB disks on his endpoints and force users to keep data on the server, because he's augmenting the core storage with centralized storage. Even then, I'd look a little crosseyed at 128GB SSDs today, given that 240GB drives are the sweet spot pricing-wise, and Windows 10 looks to have the WinSXS problem in spades.
Why not buy a 240GB and "short stroke" it? Only using half the space means that the wear leveling algorithm has more cells to play with, making doubly sure you don't hit write life limits. And if you do need more space in the future it's not a rip and replace affair to get it.
As for the "30GB SSDs into PoS terminals", to me that falls under "embedded devices" rather than desktops. Single purpose devices that are probably never going to be updated and hang around until the end of time contributing to our IoT 33Billion device attack surface. Huzzah!
Seriously though, I've deployed a fair amount of PoS terminals, thin clients, etc...and even on those - even for Linux - I find 30GB is constraining. And if it's work to squeeze the OS and all relevant defenses in there today, 6 or 10 years from now space could be a real problem. Unless, of course, updates and maintenance aren't part of the plan.
I intentionally use 128GB SSD for my users computers, it's the easiest way to ensure that they keep their files on the server and not their bloody pc !
I can swap a pc or just tell them to log on to another one if they keep their files in the correct location ffs.
The economics of AWS etc really only works if you are completely Open Source with either the classic LAMP stack or Java plus PostgreSQL or similar. You can just about get away with Windows .net and SQLserver as there are standard images for this which don't cost too much.
However if your existing application uses anything licensed you could end up paying more, and, anything that requires license key based on hardware just plain won't work. (You fire up the AWS image get the pretend MAC address email it to your software supplier, get the purchase cost approved and a week later your instant image is up and running)
Most existing legacy systems will have some annoying technical glitch that prevent easy deployment they may require an outdated version of a particular library, the LDAP address may be hard coded in a program -- all sorts of things that are wrongish but don't really cause that many problems in your own data centre become show stoppers in the cloud.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
