Tip: To see changes to articles click on the 'Vault' tab and then the 'History' tab to compare article versions.
More discussion about the FAST announcement can be found at Backchannel Comments on FAST.
EMC’s delivery of its Fully Automated Storage Tiering (FAST) Version 1 raises some questions that EMC did not directly address in its public briefings. Here are 16 with initial answers from the EMC Symmetrix team:
Below each question is a response from the Symmetrix team...
1. What’s the performance impact of real time data movement?
Symmetrix: None - it operates as a low priority background task and the data is continuously available during the move (ie. it isn’t “real time,” moves can be concurrent with active workloads, or they can be scheduled for specific periods) A follow up question is how does FAST I/O on arbitrated loop impact other devices waiting to win arbitration on that loop. Is this an issue? With the Ultrapoint back-end on DMX4, V-Max and CX, loop congestion is almost never an issue, and when it is the duration is insignificant.
2. With the need to run the analysis tool for several days to get FAST smart, will it stay smart or do users have to keep running the tool?
Symmetrix: The analysis doesn't have to run for days - you define the "analysis window"...it can run continuously and make adjustments based on the most recent window.
3. What’s the performance impact of running FAST?
Symmetrix: Both analysis and moves run as low-priority tasks so as not to disturb active applications. As data is moved, performance improves (if being promoted) or else there is no impact (demoted).
4. Can FAST respond quickly enough to surges (e.g. month-end/quarter-end processing)?
Symmetrix: Quick reaction time is not the intent of FAST v1…but, if you know these things are coming, you can ”override” FAST by doing manual VLUN moves. With FAST v2, things will get a little more interesting…
Wikibon Note: This implies FAST is designed to move things 'safely' down the storage hierarchy to take advantage of lower cost devices and free up more expensive high performance real estate for more active data sets. 'Safely' in this context means FAST is smart enough to move low activity data sets to slower devices and leave high performance data alone or promote them to higher performance devices as appropriate. It's not designed to move data up and down the hierarchy heuristically based on usage patterns over a period of time to optimize performance in an anticipatory fashion-- which is the conceptual goal of autonomic storage systems. FAST v2, with sub-LUN level granularity, makes the FAST concept more attractive by further optimizing data placement for flash device users. Indeed, FAST v1 is primarily aimed at reducing cost (SATA) and leveraging flash drives more easily.
5. How does FAST work with automated hot spindle movements?
Symmetrix: Moves are coordinated.
6. What’s the tool set like to determine whether to run FAST in batch or real time mode?
Symmetrix: Difficult to describe, but your sales team can explain or provide a demo.
7. Does FAST investigate the entire array or only hot LUNs?
Symmetrix: FAST policies define the scope in terms of both LUNs and target devices.
8. How does FAST deal with lockout periods such as a BIN file change?
Symmetrix: It simply waits, but there are far fewer such lockouts with V-Max - much more concurrency than before.
9. When upgrading to FAST v2, what happens with FAST v1 user defined policies?
Symmetrix: FAST v1 policies apply to full "thick" LUNs, FAST v2 policies will apply to "thin" ones. The old policies stay in place.
10. When will FAST support virtual provisioning?
Symmetrix & CLARiiON: FAST v2 will bring sub-LUN support; on Symm this is built on virtual provisioning.
11. Will FAST support migration between different LUN types?
Symmetrix: If you mean thick to thin, the latest release of Enginuity offers this, although the migration method is currently disruptive - TimeFinder supports thick-to-thin clones.
12. Will FAST support older generation hardware?
13. When will customers be able to self-provision FAST on CLARiiON and Celerra?
14. When will FAST automatically move data on CLARiiON from Flash to Fibre and SATA to Fibre?
15. When will FAST support the IBM System I platform?
16. Question 16 was answered recently: Can FAST non-disruptively move data between different raid levels?
Action Item: Bottom Line: Is FAST v1 worth it for customers? Yes, if you can save money by moving things down the hierarchy onto cheaper disk or avoid using under-provisioned FC drives. Users should be aware that the savings, at least initially, will be on space and not have an immediate impact on CAPEX. Over time, CAPEX savings should be realized as space is freed. As well, in theory IT productivity should be improved as less manual intervention will be required to optimize data placement. The saving will be a function of the percentage of LUNs which go from high activity to low activity over time--which for many users is most LUNs--and time to learn how to exploit the feature.
Footnotes: FAST also simplifies the task of capitalizing on the performance benefits of Flash drives by identifying the busiest LUNs - customers have complained about the complexity of doing this analysis manually.