- the same as previous according RAM: biggest volume for comparable price with not big drawbacks in speed. And prices for upgrade (because You need the same volume modules) also would be less;
Many workloads run MUCH better when there is a lot of RAM available. In particular those that have a wide file system working set, and benefit from file system caching. This is something that really needs benchmarking (with your real workload and your real file system / OS) to check.
Other workloads have a smaller working set (both of the program itself, and its file system cache usage), and then faster RAM is more important, since more RAM simply doesn't help.
The fastest single machine I've ever used at work had 1/2 TB of RAM (and several dozen PCI lanes); for the workload it was amazingly powerful (and I have no idea how much it cost, but I remember that each DIMM was $3K, and it had several dozen of them). My home server has 4 GB and a 4-core Atom CPU. The latter is optimized for low power consumption, small physical space, and reasonable cost.
- well manufactured rack mount servers from reputable brand (like IBM, Dell, Siemens/Fuji) would be ALWAYS MUCH STABLE in working;
- the same rack servers would be ALWAYS having much more lifetime;
- the QUALITY OF POWER in this rack servers ALWAYS would be better (even w/o Online-Interactive UPS). And in addition POWER source MODULES ARE DOUBLED (hotspot);
This is actually a very important observation. Amateur computers, built using desk side cases, cheap fans (often only 1 or 2 fans), and whatever power supply is on sale at NewEgg this week, tend to be somewhat unreliable. Enterprise-class servers may have less CPU power or slower RAM, but they may up for it by having about 10 or 12 fans (each individually pretty small), dual power supplies (which can be connected to two independent power sources, like one utility power + UPS, the other a generator-protected power source), and N+1 redundancy in fans and internal power distribution. Plus if you stay within a single brand (you buy all your expansion cards from the same vendor as the rack mount computer and motherboard), they tend to have very good BIOS support, for example for fan speed control. They are pretty much indestructible, exceedingly reliable (physical uptimes of a decade are common), and can be well managed (like integrating BIOS monitoring into an alarm infrastructure). The drawback is: bought new they are very expensive, and they tend to be very noisy.
Mandatory anecdote: When we got a set of new servers in the office, they had to be shipped without memory, because the server itself came from one color of money (capital investment), while the DIMMs could be bought from another color of money (as they were under $5K each, we bought them using the same pot of money used for office supplies like pencils). Which meant that someone had to volunteer to spend an afternoon in the computer room installing about a hundred DIMMs. Since I'm a fool, I did it. And one of the servers was installed at the top of the rack, so I had to use a ladder to install the DIMMs. Because getting the DIMMs up the ladder was painful, I unpacked them on the floor, then carried them up on the ladder a few at a time, without wearing an anti-static grounding strap. And managed to kill one of the DIMMs. Since we were an in-house customer, we had no warranty coverage, and my manager had to buy one extra DIMM for $3K, and he was REALLY MAD AT ME, since that looked bad on his budget. The lesson is: When dealing with expensive stuff, always wear anti-static wrist straps, connect them to the machine you're working on, and transport components in their original packaging until you're plugged in. Not sure that rule applies when using a $5 Raspberry Pi Zero though.
forum.netgate.com
