I worked for Lenovo at their plant in Research Technology Park, North Carolina, developing C++ install/driver/preload software for the ThinkXxxxx line they’d just bought off IBM. Overpriced hardware with a jillion idiosyncracies necessitating a slew of obscure software/driver patches. That contract, along with first-hand experience with the sorts of issues you describe in this post, convinced me never to buy from Lenovo.
L M A O
The term ‘shooting oneself in the foot’ comes to mind.
I try to figure out what it is that Lenovo is trying to accomplish here, but I can’t decide whether they’re evil or incompetent. Incompetency is assumed, but really, could even marketing people be this stupid? This is the kind of thing which even the CEO of a large company could understand and get angry about, which generally means that for an IT techy it’s possible to make the company buy from another vendor next year.
I guess this is why more and more companies just go generic, or even put their own systems together 
This was very predictable. If you go back to the comments on your original buy vs. rent blog post, there were several people who recommended pre-installed, expensive SAS drives. When I saw your total expenditure for these boxes I thought, there’s no way that’s enough money for what you need. Just bite the bullet and buy a box with 6 3.5 SAS drives, as big as you can get them at 15k RPM. Don’t cobble the system together with random parts. These things need to be tested as a unit, with hardware and firmware on the RAID controllers and hard drives all agreeing with each other.
If you think Jeff has a problem with just rigging it up to work, then you must not have seen this website of his called StackOverflow
[RIMSHOT]
…kidding, kidding…
LOL… just… LOL
A few geeks raving and ranting on about how we will never buy our 17 servers from Lenovo again !!11!. I bet they’re shaking in their boots big time 
Surprising how few people here actually got the Lenovo doesn’t care about or even want customers who only buy 10 machines point. Most people here don’t seem to get that a real enterprise datacenter works a little differently than their 10 rack geek playground.
I dismantled an 8 drive lenovo server a couple of weeks ago. I believe the scrap parts (intended for the trash) are still at the office. I will check at the office today, maybe they would be interchangeable. They do look identical.
Ah, duct tape and zip ties, my friend. lol
Years ago I used erector-set parts to mount a hard drive. 4 holes were drilled into an emptyish spot on the case.
Jeff,
As many said–SuperMicro is your friend here. You can get redundant PSUs. You’ll have to put in your own RAID card, but that’s what you probably should have done with the Lenovo anyway.
There can be big perf differences–at least for RAID5 or 6–so this is one area you probably DON’T want to skimp and rely on what Lenovo would have supplied. (The perf differences may not be too much if going the RAID 10 route, but then a cheapie onboard RAID from SuperMicro would probably have been just fine since there’s no parity calculation involved).
If you’re going for a large GB array, I’d still suggest RAID 6 for simplicity and rebuild safety (db perf purists be-damned).
Jeff -
It’s slightly amusing you’re stymied by the lack of a few pieces of bent metal. I doubt a dud BIOS flash would impede you in the same way. Go to the basement, select sheet metal stock of the appropriate thickness, cut to size, put into your metal bending machine, drill necessary holes, and you’re done.
If you don’t have a basement (etc.), there’s someone in your neighborhood who cuts and bends metal either for a living or as a hobby. The cost should be: not much.
How did you cope in the days of soldering and wirewrap?
I’m going to second the SuperMicro idea that several others have talked about.
We have probably 50 or more supermicro servers and can easily recommend them. As long as you buy the right motherboard for the case (or just buy it bare bones) you should be good to go. The build quality is just as good or better than most of the other OEMs (Dell, HP, etc.).
We haven’t had to use customer support many times but they were always helpful when we did. Once we had a raid card that wouldn’t work with one of their motherboards and they gave us custom bios builds to fix the problem until they could release a new official bios. Try getting someone like dell or lenovo to do that. (Not that i have add a problem will dell in the past, but getting a custom bios doesn’t sound like something they would/could do).
** Batteries not included
As an insider, I would not recommend Azure. From my interactions with those fellows, the group seems to be a huge mess. Way too many architects, no clear vision, no business plan, way too many senior (principal in Microsoft lingo) people who think they’re smarter than everybody else on the team, and, as often happens when all these ingredients are present, version 1 is WAY, WAY too huge in scope. As a result, it will only become usable by V3 SP2 (or whatever the service counterpart to it would be called). The only way to build a large system that works is by first building a smaller system that works. Someone needs to beat this simple wisdom into their heads.
Why not fabricate your own? If not from measurements, then possibly buy one from the manufacturer and then copy them. I see you’re in Berkeley. I wish I was only 35 miles from a TechShop. http://www.techshop.ws/ But they keep postponing opening one in Austin.
Jeff, I may have some of these on hand that come from failed HDDs. If you’re interested, let me know!
The comments indicate that these drives have an extremely high failure rate
Not to disregard the statistical significance of three random newegg.com users, or anything, but you’ll find the very same carping for every drive manufacturer and model. The only people that care enough to post anything are the tiny percent with failures on their hands. For more actual data, try here: http://www.storagereview.com/map/lm.cgi/survey_login
Third, how much of your time have you already spent on this issue? What is your time worth?
Well, I’m on holiday, and doing this essentially for fun, because I enjoy it. The usual rules don’t really apply in this case.
If you want to use generic hardware, I recommend you buy your servers from somewhere like Supermicro.
Once I factored in the true hardware raid and (for the RD120) the dual redundant power supplies, the SuperMicro stuff actually wasn’t that great of a deal. But I agree it’s incredibly nice not to have to deal with the drive caddy used car salesman routine.
The fact that you’re talking about running the site with just two servers is slightly scary.
See the articles I wrote on plentyoffish.com , which runs a #13 US site on five servers and a MSFT stack, with probably two orders of magnitude more traffic than we get.
You’ve got zero redundancy.
How is this any different than my current rental situation? I have zero redundancy. I’d have to re-provision a server from scratch (beyond the OS, I mean) and restore data backups myself if there was a catastrophic failure.
How is this any different than my current rental situation? I have zero redundancy. I’d have to re-provision a server from scratch (beyond the OS, I mean) and restore data backups myself if there was a catastrophic failure.
One step at a time. Next thing you know you will be purchasing a SAN!
You have been nave, indeed.
Most hardware vendors want you to buy their own drive mounts…