Customers sometimes ask where to initially place data that will be tiering with FAST.
One suggestion is to set the population of data to be "highest first". That is actually a good idea after initial data load and movement because new data tends to be "hot" and therefore can justify sitting on expensive FLASH.
The very very cool thing about a "FLASH first" strategy is that the capacity of the FLASH tier will constantly be refreshed with new and hot data, whereas cold data automatically trickles down to cheaper seats. This can have significant impact to your realized service level - you'll get the benefit of having hot data always served from FLASH - but not having to shell out the cost for a 100% FLASH storage pool. Very cool!
This is an interesting way to think about FAST - it gives you close to 100% FLASH service levels at 5% the cost!
In effect, with "highest first" the FLASH tier acts as a FIFO (first in first out) buffer for new, hot data.
Let's say data stays hot for 90 days on average (I know this varies from customer to customer). As data gets cold, it will be demoted to lower cost disk, freeing capacity for new incoming data. This FIFO strategy allows us to recycle the FLASH capacity quite a few times during the system's service life. If data stays hot for, say, 90 days on average (blue dot below) and we keep the system online for 5 years, then the number of completed recycle turns is 1825days/90days = 20.27 times.
Wait a minute... this is important.
We have an expensive asset here - FLASH - that costs list ~$65/GB turned 20 times during its life thanks to dynamic data movement of FAST. If we compare that to old school static data placement, FAST effectively enables us to get 20X the service from FLASH resulting in a effective cost of $65/20/GB = $3.25/GB. Compared to $5.40/GB for traditional 15K FC drives used in static RAID sets in the past, FLASH + FAST has cut effective cost 40%.
So "highest first" can be a data management strategy that further increases service levels while delivering the highest possible return on assets. With FLASH in the pool, this becomes a "FLASH First" strategy.