A Few Quick Details on EMC FAST

28 09 2010

While digging around on PowerLink I happened upon a new whitepaper on EMC FAST so thought I would share.  There is a ton of buzz around this awesome new functionality for EMC’s CLARiiON and Celerra arrays (as well as Symmetrix/VMAX for the sake of accuracy) but being so new there has definitely been a lack of substance.

The main focus here is FAST Sub-LUN, aka the ability to migrate data at a sub-LUN level across multiple tiers of storage.  FAST Cache, the ability to add Enterprise Flash Drives to an array for increasing the SP cache, also gets an honerable mention.

First off, if you’re trying to determine exactly how the different tiers of storage are applicable to different workloads the following chart is very helpful.

I found it interesting that EFDs can be very beneficial for high I/O and super fast response for reads (no surprise there), but may not be the best bet for bandwidth intensive workloads based on cost per IOP, where FC may be the better choice.  But hey…that’s why FAST Sub-LUN is so great, you can just add the multiple tiers of storage into a pool and allow the array to make the best use of each specific type of disk.

Along those lines, FAST relies on heterogeneous pools of storage introduced in FLARE 30.  While the pools can contain different types of storage, meaning FC, EFDs, and SATA, another best practice is to ensure you maintain the same rotational speed within each pool tier. 

And if you have a CX4 based array but are still setup with traditional RAID Group-based LUNs, don’t fret…you can leverage LUN Migration to move to a Storage Pool topology therefore allowing for the adoption of FAST functionality.  Just keep in mind that while most environments won’t encounter any limitations with this technology, there are some facts to keep in mind, mainly the number of LUNs on a single array that can be configured to leverage FAST.  The limitations listed below are a 1-for-1 in terms of the maximum number of pool LUNs per system as well.

Once FAST Sub-LUN is enabled within a storage pool, there is an algorithm comprised of 3 main components, statistics collection, analysis, and relocation, which is used to determine how data is tiered to enable the best use of each type of storage.  The algorithm begins by gathering statistics on the activity level of each data slice within a storage pool.  This activity level is ultimately based on a culmination  of the writes, reads, and overall I/O’s against the 1GB slices of data.  The most recent activity level is given more weight in the overall analysis, and the longer the data is included in the analysis, the less consideration it is given in the overall equation.  Therefore, the most recent data always has the most impact in tiering decisions. 

The algorithm next reviews the performance of each 1GB slice of data on an hourly basis, and then based on the results of that analysis it begins to rank the slices from hottest, i.e. most heavily accessed, to coldest, least accessed. 

Once the analysis is complete, the array will begin to move the 1GB slices of data based on the previously determined ranking while always giving the higher priority slices precedence.  Therefore data will always be migrated to the highest performing tiers of storage first, therefore pushing less critical data to lower tiers enabling the most efficient use of the overall array.

In terms of FAST Cache, there was nothing too earth shattering but again a few tidbits of knowledge.  First off, the following chart helps with the determination between when FAST Cache should be used as opposed to FAST Sub-LUN.


And just in case you’re wondering if both can be used on the same array, the answer is of course, yes!  Along those lines, if you’re wondering if they will conflict, the answer is no.  If a slice of data has been promoted to a Flash tier of storage within a storage pool, FAST Cache will not promote that data to EFD-based Cache to ensure the system doesn’t waste resources copying from one Flash drive to another.  Likewise, if data is already being serviced by FAST Cache, that will drop its activity level ranking from storage pool and FAST Sub-LUN perspective, and it therefore will not be moved to an EFD tier thus, once again, preventing duplication in effort.

Essentially, FAST Cache is suited perfectly to improving the overall  performance of burst-prone data, while FAST Sub-LUN moves data to the appropriate tier of respective storage, therefore lowering the TCO for housing your data and also improving performance over a longer timeframe.  In the end…there are completely complimentary of one another.





3 responses

28 09 2010


So without the FAST enabler, there appears to be no advantage to creating multi-tier storage pools… correct? I’ve noticed that FLARE 30 will allow one to create a FC/SATA storage pool even without the enabler, but to what benefit? What happens if one creates a FC/SATA storage pool without the FAST enabler? Does it fill the FC, then fill the SATA?

10 10 2010

Hi Jeff,

First off, thanks for reading my post. While there are a lot of advantages to using storage pools including; the ability to leverage Virtual Provisioning which includes Thin LUN technology, increased performance through better LUN distribution, as well as removing the traditional constraints around RAID groups such as drive number limitations and negating the need to use Meta-LUNs which are inherently complex to manage, I can’t find any justification for using storage pools with multiple tiers of storage without FAST Sub-LUN enabled. There basically wouldn’t be anything to control distribution of a LUN across the underlying drives and I would imagine that inconsistent performance would be a result. If you have found any information to the contrary please feel free to share.

11 10 2010


That’s my take on it as well. My only wish is that EMC would provide FAST free of charge!

Other related question:
Storage pool – starts with 5 FC disks (4+1 raid 5). Thin luns created and data populated. I then add another 5 or 10 FC drives of same speed/capacity.

Does the system restripe the existing data across all disks as was the case in a expanded striped MetaLUN? I can find no EMC data on this.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: