On 8/1/2020 8:06 PM, David Christensen wrote:
On 2020-08-01 16:30, Leslie Rhorer wrote:I am a big proponent of having a separate boot drive+1no matter what the file system, I would definitely up the memory to the 16GB max,1. I try very hard not to spend money on obsolete technology.
What is "obsolete"? If a system does its job adequately, it isn't "obsolete", no matter how old it might be or how many newer bells and whistles it may fail to have. What's more, why spend $2000 or more on unnecessary hardware when $70 for a minor upgrade to older hardware will do the job perfectly well? To be sure, there comes a point where money spent on an older system that can no longer easily meet newer demands is poorly spent - maybe. Memory is often an example, as newer systems often cannot use older memory. Other upgrades, such as newer, larger drives, a new display, etc, can easily be used with a new system when the old one is finally retired.
I will admit one should be very cognizant of exactly how much one is actually saving in the long run by spending money on old technology.
2. I would wait until the computer is put into use and measurements indicate memory is insufficient.
I won't quibble with this. It's not the way I would probably handle it, but it is not an unreasonable position.
3. One or more small, fast SSD's used as a ZFS cache devices and/or intent log devices can dramatically improve performance of ZFS.
Or lots of other things. On something like a laptop, it is virtually insane not to employ an SSD. Even a high-powered server is very well served to boot from SSDs, however, and the cost has come down so drastically, there is little justification not to employ SSDs. If I were running ZFS, I definitely would employ SSDs as cache and intent log devices. For that matter, putting md intent bitmaps on SSDs is a pretty good idea, and booting from SSDs is a really good idea no matter what the mix of file systems.
4. ZFS and non-ECC memory risk the "scrub of death" scenario, which has been debated endlessly.
Yes. As I said, ZFS has many great features, especially for an enterprise server. For a simple NAS, not so much, I think.
I also ran desktop hardware as servers for many years, but decided to buy a "real" server (with ECC memory) when building a ZFS file server. The ECC is invisible, but the combination of Xeon processor, dual-channel memory, and server chips throughout provide obvious performance.
Or not. No matter how mighty, the server will not be able to out-perform the infrastructure. My fairly humble desktop servers can easily pump out more than 4Gbps to clients, and indeed the nightly sync between the main array and the backup array proceeds at just such a rate across the two 10G optical links into my servers from the switch. The links from the switch to the hosts, however, are all either 1G or 100M. While technically, all the hosts banging out full bore together might just be able to reach the transfer limit of the server, in practical terms, there is no way that amount of data could be digested or regurgitated from the limited number of hosts on the LAN. The fact the two motherboards are not server class and are over 16 years old really doesn't matter.