best upgrade process for server

[Chris Gordon]

> On Apr 23, 2020, at 4:25 PM, David Christensen <dpchrist at holgerdanske.com> wrote:
> 
> When it was time to retire my previous desktop/ Linux SOHO server with an up-to-date FreeBSD server, I bought a lightly used Dell PowerEdge T30 with one Xeon E3-1225 v5 processor, one 8 GB ECC memory module, one 1 TB SATA HDD, and one DVD+/-RW drive.  (It is important that computers running ZFS have ECC memory.)

I would clarify the statement about ECC memory in that ECC memory is important for any system where memory errors are significantly detrimental to the applications running on the server whereas to justify the additional cost.  

As for file systems, ECC memory is no more or less important for ZFS, UFS, ext*, NTFS, etc.  The only case where I think the argument for ZFS could be stronger than other file systems is by virtue of ZFS typically using more memory for the ARC and thereby theoretically increasing the probability of a read serviced from ARC experiencing a bit flip. This thinking, though, applies to applications, too --  as more memory is used, the probability of encountering a memory error increases.  The idea that ZFS uniquely "requires" ECC memory to be "safe" often seems to stem back to an a blog post about the "scrub of death".  This has been debunked many times.  Allan Jude discusses this is far more detail and expertise than I can starting at about minute 57 of Episode 200 of the BSDNOw podcast:  https://www.jupiterbroadcasting.com/116226/getting-scrubbed-to-death-bsd-now-200/

That said, if you can afford ECC memory, you're better than without it.  You are even better with backups that are on something other than the machine you're backing up.  You're even better if you can periodically test those backups and validate that they are indeed good and recoverable.

Chris