I feel like I've read as much as I can on the SLOG, but there is also possibly conflicting information that is likely from old versions.
I understand that SLOG is not worth the effort if you're system is not doing synchronous writes.
I also understand that there was a risk to using a SLOG such that if the drive went bad, your entire pool would no longer import because it broke the import process. I believe I have read this is no longer the case in zfs version 5000 and that if a SLOG is lost, it only affects the transactions not committed off the SLOG before losing the SLOG.
I also read that it is desirable to have a mirrored SLOG, but that seemed to me to be a way to reduce risk of losing your pool if the previous issue still was possible.
There is a tradeoff between size of SLOG and size of RAM. If you're capped at 32GB ram, you shouldn't have a SLOG larger than 128GB.
My questions:
* I had read somewhere you can run a command to see if you're system will benefit from a SLOG. Does this command help discover the frequency of synchronous writes?
* Is it no longer a risk that if the SLOG is lost, you lose your pool? I've avoided installing a SLOG for this reason even though I use VMs that perform synchronous writes.
* Am I correct that it may not be as worth the cost to run a mirrored SLOG?
* Since many of the faster SSDs are not coming in smaller sizes these days, could I partition off 64GB from at 256GB SSD and use that as my SLOG? Could I use the remaining 192GB for other storage?
* If the SSD has enough bandwidth, and the previous answer is yes, could I put a 64GB SLOG and a 192GB L2ARC on a 256GB SSD? I'm thinking something like a Samsung Pro 256GB.
* What are the reasons not to combine the SLOG and L2ARC?
Edit: Thank you. Was using the wrong term.
I understand that SLOG is not worth the effort if you're system is not doing synchronous writes.
I also understand that there was a risk to using a SLOG such that if the drive went bad, your entire pool would no longer import because it broke the import process. I believe I have read this is no longer the case in zfs version 5000 and that if a SLOG is lost, it only affects the transactions not committed off the SLOG before losing the SLOG.
I also read that it is desirable to have a mirrored SLOG, but that seemed to me to be a way to reduce risk of losing your pool if the previous issue still was possible.
There is a tradeoff between size of SLOG and size of RAM. If you're capped at 32GB ram, you shouldn't have a SLOG larger than 128GB.
My questions:
* I had read somewhere you can run a command to see if you're system will benefit from a SLOG. Does this command help discover the frequency of synchronous writes?
* Is it no longer a risk that if the SLOG is lost, you lose your pool? I've avoided installing a SLOG for this reason even though I use VMs that perform synchronous writes.
* Am I correct that it may not be as worth the cost to run a mirrored SLOG?
* Since many of the faster SSDs are not coming in smaller sizes these days, could I partition off 64GB from at 256GB SSD and use that as my SLOG? Could I use the remaining 192GB for other storage?
* If the SSD has enough bandwidth, and the previous answer is yes, could I put a 64GB SLOG and a 192GB L2ARC on a 256GB SSD? I'm thinking something like a Samsung Pro 256GB.
* What are the reasons not to combine the SLOG and L2ARC?
Edit: Thank you. Was using the wrong term.
Last edited: