Adam Leventhal wrote:
>> 1) How quickly and by what trigger is data flushed from the SLOG to
>> primary storage?  (I setup a test once just using normal disks and I
>> didn't see any data migration occurring.)
>>     
>
> Data isn't really flushed from the slog to primary storage -- indeed, the
> slog is never read from during normal operation. Instead data is streamed
> to primary storage asynchronously. This should happen within 5-10 seconds
> at regular intervals as transaction groups are closed.
>
>   
>> 2) How can I monitor SLOG fullness in order to determine whether
>> additional capacity is required?
>>     
>
> Rather than monitoring fullness, I'd suggest you look at the datarate to the
> slog device.

That seems very imprecise.   If I'm moving data to SLOG at 100MB/s does
that mean that 32GB is sufficient or not... it suggests some prior
wisdom and experience.

Thus, the following logical question would be: what is the behavior of a
full SLOG?  Is there any type of message sent to syslog or way by which
I could use DTrace to figure out if this was the case?


And, for good measure, I'll throw on one more question: I recall that it
is not possible to remove a SLOG from a pool once its been added.  Is
there an RFE to fix this?  Or, alternatively, is there a belief that
there should be?   I say this because I would assume that if the SLOG
was full, the SPA would refrain from using it and write directly to the
pool, in which case it should be easy to simply use the same function to
remove existing SLOGs.


Thanks for your help Adam!

benr.
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss

Reply via email to