Silly question re Zeusram drives

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

Keljian

Active Member
Sep 9, 2015
428
71
28
Melbourne Australia
Ok I get that Zeusram drives are fast, what I don't understand is why someone would get one, versus 8 gb more ram if the latter is a possibility.

Could someone please enlighten me?
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
They have power loss protection. RAM does not. If you are writing data, that can be important (e.g. ZIL/ SLOG)
 
  • Like
Reactions: Keljian

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
Wouldn't a UPS and something that writes the ram to a drive, like a shutdown script, solve that issue?
Somewhat different. What if your shutdown script hangs? What if the issue is within the server not external power?
 
  • Like
Reactions: Keljian

MiniKnight

Well-Known Member
Mar 30, 2012
3,072
973
113
NYC
More like you can do it right or cheap.

Power loss protection on the ZIL means when the system thinks it is written on non-volatile media it (basically) is.

UPS + script means you hope that if your power fails that you have enough time for your RAM disk to flush through your CPU to the PCIe bus to a SAS controller to a backplane and then to the drives while the OS is properly completing a shutdown. Might it work most times? Sure. There's a reason people use these SAS or SATA RAM drives on larger systems, and it isn't because they're cheap.
 
  • Like
Reactions: Keljian

gea

Well-Known Member
Dec 31, 2010
3,140
1,182
113
DE
To be correct.
All ZFS writes are always collected in a rambased writecache to transform small random writes to a large and fast sequential write. On a powerloss some seconds of last writes are lost and an older filesystem (VM storage) may become corrupted or last database transactions lost. This is the case on any type of pool including SSD only pools.

To be safe, you can enable sync on a filesystem or disable writeback on an iSCSI logical unit. This will log last writes that are in ramcache and not yet on stable storage to a device called ZIL to make it possible to restore them on next reboot. This is similar in intention like a BBU + cache on hardwareraid. While regular disks are very slow on sync write logging, SSDs can be fast enough. (You must enable sync as well on SSD pools).

As an option, you can use an Slog on ZFS. This is a ZIL on a separate disk that is usually faster on samll random writes than the onpool ZIL. With SSDs this may be an option as well but only if the Slog is much faster as the SSD pool - mostly not the case.

Remains the problem with SSD where the firmware is reorganising data in the background (garbage collection). On a power outage data may be lost. To be safe, you need SSD with powerloss protection.
 
Last edited:

whitey

Moderator
Jun 30, 2014
2,766
868
113
41
For the nominal cost overhead (say $20-50 per device, sometimes not even that) I always prefer to stick w/ Ent Class PLP SSD's (call me a wuss or a man w/ tendencies to error on the side of caution heh)...used is fine as we all know here. Deals to be had!

Psst, I run my 8 device hussl4040ass600 ZFS pool w/out L2ARC/ZIL w/ 40+ VM's concurrently, seems pretty snappy/happy. For spinning media it's like night and day though as a lot of us know or have experienced.

EDIT: Side topic, I had a small vSAN hiccup recently that FreeNAS/ZFS came to the rescue of to restore a VM from snapshot. First time I've seen that, maybe I'll start a thread if I didn't destroy my logging/screenshots in the melee :-D
 
Last edited:
  • Like
Reactions: Patrick