24GB+ of RAM?

ddrueding said:
What is the cheapest way to get a machine with 24GB+ of RAM?

Should I be looking for motherboards with more than 6 slots?
Click to expand...
The answer to your second question is Yes.

As to your first question, IMO it's the wrong question. You're going to be running multiple server images; hardware reliability is of higher importance than for a single server as there's more impact to the business if the platform goes down. You're already saving a ton of capital from not buying multiple physical servers and a lot of operating expense by not having to pay for electricity & cooling for those extra machines. Use some of that savings to invest in high quality, robust components. Quality over cost. You probably saved $15K or more on the reduced box count; spend a few hundred of that on the best components you can get. The better question would be what's the best memory to pair with the system board & CPU?

I know you're rolling your own for the build, but as a sanity check you might look at comparable servers from the big guys like IBM & Dell to validate that you can do better for less. FWIW, the IBM System x3650 M2 has 16 memory slots & 12 2.5" disk bays in a 2U chassis.
 
Don't forget that you can also buy a BrandX server with a boat load of DIMM slots and load them up with 3rd party RAM from someone like Crucial where the memory is validated for use with the server.

I know many big box manufacturers would price gouge memory upgrades. I used to find that it was cheaper and I could get better RAM from someone like Crucial.
 
Good point. My employer has had good luck buying Kingston RAM for Dell servers in the past. I'm not sure what we do nowadays but we are buying servers with 16-32GB standard since they are almost all serving as VMware hosts.
 
Mercutio said:
OK. Say it with me:
C-L-U-S-T-E-R
Click to expand...

Dammit, I had a great response that got eaten, so you'll get the short version.

1. If you don't have the expensive licenses (vMotion), having one big pool is the most efficient use of your resources, and makes it much easier to manage.

2. If you do have the expensive licenses, you are paying per CPU socket, and should be spreading that cost across as many VMs as possible.
 
Fushigi said:
I know you're rolling your own for the build, but as a sanity check you might look at comparable servers from the big guys like IBM & Dell to validate that you can do better for less. FWIW, the IBM System x3650 M2 has 16 memory slots & 12 2.5" disk bays in a 2U chassis.
Click to expand...

That isn't bad, actually. Just spec up the CPUs a bit, add the second PSU and the support for all the drive bays that are all already there (can't believe this isn't standard), and that is only at $3k. Not bad at all.
 
blakerwry said:
Don't forget that you can also buy a BrandX server with a boat load of DIMM slots and load them up with 3rd party RAM from someone like Crucial where the memory is validated for use with the server.

I know many big box manufacturers would price gouge memory upgrades. I used to find that it was cheaper and I could get better RAM from someone like Crucial.
Click to expand...

blakerwry said:
btw, I would stick with Kingston or Crucial. I've had trouble with other brands - Corsair, Buffalo, etc
Click to expand...

Fushigi said:
Good point. My employer has had good luck buying Kingston RAM for Dell servers in the past. I'm not sure what we do nowadays but we are buying servers with 16-32GB standard since they are almost all serving as VMware hosts.
Click to expand...

Kingston is the plan, they are just cheap and seem to always work. Crucial is good as well, but seems to be more expensive. It made sense when they were the best, but they share that rep now.
 
I think I'd rather split the workload to a second system for redundancy's sake, rather than building some obscenely beefy (and overpriced) single system, but this Tyan board looks like a pretty good deal.

I do not like having all my eggs in one basket though.
 
Make sure that RAM is compatible on that board's HCL.

Anyway, I'm still thinking vMotion is a worthwhile investment for you. It certainly adds a great deal to your ability to manage your Guest OSes, and since you're not planning to buy any more Windows servers, you can use the money you're saving on not buying more licenses and CALs for them.
 
Last edited:
Pradeep said:
If you have dual psus, then dual pdus feeding from duals gens is the next step :)
Click to expand...

Already have the dual PDUs, each drawing from their own on-line UPS, that then draws from the single beast UPS, and the single massive generator. Not flawless, but pretty damn good. ;)
 
I just don't like seeing things like that because there is always something really important somewhere that someone forgot about that is either single corded or non redundant on its two power supplies the way it is loaded. Anytime we do maintenance or a shutdown on a piece of one of those systems someone is always scrambling because something went down. That's the good scenarios, the bad ones have everything switching to the "B" side, overloading that breaker and tripping it because someone somewhere trusted an amp clamp to do their load calculation for them not realizing that when the A is in use there is no load on the B side till A goes away on some equipment. It takes an extraordinary amount of planning and cooperation to pull this off successfully, something that just doesn't happen over the years including personnel turnover. I know several places that are doing this and had to resort to lock and key to prevent unauthorized plugins, they just keep getting burnt. I feel so much better with a multi module UPS system with redundancy or a true n+1 system with static switches.
 
Speaking of power failures...just having our first major rain of the year. The rain started at 4:15AM, the first gensets started turning on at 4:30AM (after 5 minutes of being on battery). PG&E (our power company) is crap.

Now I get to drive to some of our more remote locations in the rain to find out what went wrong with their battery systems.
 
From my point of view PG&E is awesome, really drove business for us around 2001 time frame with those rolling blackouts. We could watch the news for the times to plan when we needed more techs on call. Really sort of odd when you can plan for failures like that, most of the time we are emergency response driven when working on power equipment.
 
Yeah with redundant power you have to overprovision by 2x. Costly but what is uptime worth to a critical process? Dual Cat 1.5MWs was the biggest I got to play with. Keep a good eye on the batts Dave.
 
Horribly inefficient to have a standby unit too, just looking at a unit that has 0% load on it and is pulling 30amps per phase at 480 volt can't be cheap. Super expensive battery charger till it is needed.

Also see it completely the other way around, a data center with no UPS or generator at all. They just had conditioned power via some datawaves (special transformer capacitive filter). It was a render farm and they tell me worst case they loose whatever frame was being worked on and maybe 10 minutes setup time. Wouldn't tell me what they were working on though but there was probably 40 standard 19" racks all filled with 1U machines. Mind boggling power in there.
 
Can you really notice the difference using the SSDs with ESXi? It just seems like the money would be better spent on more spindles dedicated per VM rather than SSDs...
 
Handruin said:
It just seems like the money would be better spent on more spindles dedicated per VM rather than SSDs...
Click to expand...

That doesn't make sense to me. The way I see it, either disk access matters or it doesn't. I'm not that sure it does (other than for backup, which should rock). But if disk access matters, surely faster disks are better? It isn't a capacity thing, as most of my VMs are under 6GB.
 
If you have the money to spare, sure they're going to be faster. If the disk access doesn't matter, why not save $1950 and get one 2TB drive for all your VMs or put the money into more RAM?
 
Handruin said:
If you have the money to spare, sure they're going to be faster. If the disk access doesn't matter, why not save $1950 and get one 2TB drive for all your VMs or put the money into more RAM?
Click to expand...

We'll find out ;) I'll leave one of the X25-Es out and replace it with an HTS541040G9SA00 (Hitachi 5400RPM 2.5" 40GB). If I can't tell the difference, the SSDs go elsewhere. But I suspect that it will make a difference.
 
You must log in or register to reply here.
Back
Top