Questions re swap-on-zfs

[Zaphod Beeblebrox]

While I remain convinced that swap on ZFS is bad, it also precludes getting
a dump, so I configure my ZFS systems with a partition for swap and a
partition for ZFS on each drive.  This has the additional bonus of making
swap effectively faster on a 2 drive system (2 swap partitions).


On Fri, Oct 4, 2013 at 12:35 PM, Volodymyr Kostyrko <c.kworr at gmail.com>wrote:

> 04.10.2013 12:50, Beeblebrox wrote:
>
>> I have 2 swap devices on my system. #1-tank0/swap on ZFS, #2-/dev/ada0p1
>> on
>> SSD 1GB. I have some questions for swap-on-zfs:
>>
>> 1. I had read in the past that swap-on-zfs can cause instability if memory
>> gets used to full capacity. Is this still the case, or has that issue been
>> corrected? The reason I created swap-#2 is precisely for this purpose; a
>> small backup resource.
>>
>
> Yes.
>
>
>  2. If the problem still persists, I would like to tweak the setup so that
>> swap-#2 is only used as last resort. I tried setting the priority (pri) in
>> fstab, but it does not seem to work - how can this be done?
>> /dev/ada0p1   none   swap   sw,pri=0   0  0
>> tank0/swap    none   zfs    sw,pri=9   0  0
>>
>
> Nope.
>
>
>  3. Is it necessary for the data on swap to go through the Intent Log? I
>> have
>> a separate ZIL device for tank0 pool. What risk is there if I disable the
>> log for the swap dataset? Is this done by modifying the "primarycache"
>> parameter for dataset?
>>
>
> There's no need for SWAP to go through any caches or ZIL.
>
>
>  4. Any other ideas to speed-up swap performance on tank0/swap?
>>
>
> In fact I even tried this:
>
>  * primarycache=metadata (that's would speed up finding blocks without
> caching them;
>  * logbias=throughput (don't use ZIL);
>  * sync=disabled (write data fast and right now);
>  * checksum=off (if it's gone it's gone, period).
>
> The machine still hangs from time to time so I think you wouldn't be lucky
> too.
>
> PS: compression=lz4 makes this host survive longer but this can be just my
> imagination...
>
> --
> Sphinx of black quartz, judge my vow.
>
> ______________________________**_________________
> freebsd-fs at freebsd.org mailing list
> http://lists.freebsd.org/**mailman/listinfo/freebsd-fs<http://lists.freebsd.org/mailman/listinfo/freebsd-fs>
> To unsubscribe, send any mail to "freebsd-fs-unsubscribe@**freebsd.org<freebsd-fs-unsubscribe at freebsd.org>
> "
>