FAST (FULLY AUTOMATED STORAGE TIERING)
FAST made a debut in the storage market yesterday (12/08/09). Finally after the market buzz we got a preview of the product in terms of its features, functionality, characteristics, possible shortcomings and use cases.
This blog post focuses on the features, the drawbacks and some applications around FAST. By no means is this a comprehensive or an exhaustive list of the above.
After all, FAST makes a debut, (The previous post on FAST).
NOTE: Out of the box thinking by EMC, imagine the flexibility one would have with a large infrastructure and FAST moving data based on policy. Its all about the big picture.
Here are some features, highlights of FAST and how it operates.
- FAST will perform data movement based on IOPS, average I/O size and write percentage. This is currently true for the Symmetrix V-Max, Clariion CX4 and Celerra NS.
- Three elements that define FAST: Storage Type, FAST policies and Storage Groups.
- FAST is based on user defined – configuration policies.
- The configuration of FAST is typically done through FAST wizards (Symmetrix Management Console) on Symmetrix V-Max, FAST LUN Migrator for Clariion CX4 and Rainfinity File Management Appliance or VE for Celerra NS.
- User defined analysis period for FAST. That will enable FAST to recommend or perform data analysis and then a data move based on policy.
- FAST created policies will associate with Storage Groups.
- FAST policies will be configurable at a LUN / drive / drive type / speed etc level.
- Data movement will take place based on a time of the date policy called “COLD” move or on usage policy called “HOT” move.
- Data movement will take place between various drive types, various LUN types and LUN sizes. (LUN types, LUN sizes will need to be same for the source and destination LUNs). For example a 9GB FBA LUN being migrated from Fibre to FLASH will need similar source and destination LUN properties.
- FAST data analysis will be performed in the background.
- For Symmetrix V-Max platform FAST will perform all analysis without the use of Symmetrix Performance Analyzer. Understanding is there will be some sort of API plugin available on the Service Processor of the V-Max that will enable Symmetrix Management Console (FAST plugin) to interface with the Symmetrix through the SYMAPI interface.
- For Clariions, the performance data of the array will be monitored and collected by Navisphere Analyzer.
- Based on the source LUN analysis, FAST will recommend the user to move the data either a faster speed drives or a slower speed drives.
- FAST will enable roll back based on user preferences (automated).
- FAST will be configurable by Symmetrix Management console wizards or SYMCLI
- FAST configurable by Clariion Naviseccli and installed on a host connected to Clariion, Host software called FAST LUN Migrator.
- FAST configurable by Rainfinity File Management Appliance GUI or CLI for Celerra NS “out-of-box data” movement. This enables the Celerra to move data to another Celerra or Centera or Atmos.
- FAST configurable by Rainfinity File Management / VE (Virtual Appliance) for Celerra NS “in-the-box” data movement.
- FAST can be installed non-disruptively on all platforms.
- FAST will operate both at an FBA and CKD level supporting open systems and mainframes for V-Max.
- FAST will operate at a LUN level on Symmetrix V-Max
- FAST will operate at a LUN level on Clariion CX4
- FAST will operate at a file level on Celerra NS
- FAST v1 users will be able to purchase a FAST v2 upgrade when it’s released in second half of 2010.
- FAST can be purchased as a FAST suite or part of an ATSM (Advanced Tiering Storage Management) suite with bundled discounting prices.
- For a unified storage system like a front end NAS (Celerra) with a backend SAN (Clariion), FAST can coexists at both levels. But it is not recommended to deploy FAST at a Celerra LUN level.
- FAST integrates with Symmetrix Management Console and with Rainfinity GUI (Celerra) for simple management. Though Clariion implementation will need one to specialize in CLI.
FAST introduction by EMC
Here are some drawbacks of FAST as I see it today.
- Does not support Virtual provisioning. So the Virtual provisioned LUNs will not be FAST enabled. Sub-LUN expected in second half of 2010
- FAST is not free and is charged based on RAW CAPACITY of the Storage Array.
- FAST will only work with similar LUN types (example FBA LUNs can be migrated to FBA) and LUN sizes (9GB LUN can be migrated to a 9GB LUN) only.
- FAST will require Professional Services and is not recommended for customer self implementation at least for the Clariion and the Celerra platforms. EMC is making a claim that FAST can be self provisioned on the Symmetrix V-Max platform.
- FAST works only on current generation systems like EMC V-Max Enginuity 5874, Clariion CX4 Release 29 and Celerra NS.
- With Clariion CX4 and FAST implementation, requirement is to have Navisphere Analyzer.
- With Celerra NS and FAST implementation, requirement is to have Rainfinity File Management /VE or Appliance.
- With Symmetrix V-Max FAST implementation, requirement is to have Symmetrix Management Console (Not free any more starting with the V-Max)
- For the Clariions, FAST only analyzes Fibre drives and LUN movement has to initiate from Fibre channel to FLASH or Fibre channel to SATA drives only. Movement from FLASH to Fibre or SATA to Fibre has to be initiated manually.
- A customer implementing Virtual Provisioning on Clariion CX4 and Symmetrix V-Max that is using 50% Thick Provisioning and 50% Virtual Provisioning. FAST will not work with Virtual Provisioned LUNs. But the customer will probably pay for 100% of Storage Raw Capacity when it comes to paying for FAST either non-bundled or bundled in an ATSM (Advanced Tiering Storage Management).
- FAST does not support IBM System I platform currently.
Some real world FAST applications
- Acadia Service Model can now add granularity with offerings around on-demand resources using FAST at its core.
- On Demand application and workload needs can be met for OLTP, Data Warehousing, Mainframe compute and Virtual compute using FAST driven policies.
- Multi-tenancy with Private Clouds
- Move high demand data on faster drives while rarely used data goes on slower drives.
- Sell ITaaS based on SLA’s. Higher SLA’s can mean higher price. All automated processes controlled by policy.
- Higher transactions typically mean low overall cost; mean higher efficiency means higher profits all achievable through FAST policy engine.
Need to see some real world implementations of FAST now. In theory and on paper, FAST looks pretty compelling but practically will it do the magic.
Some unanswered questions today
- How does it prevent from LUNs jumping between Fiber, FLASH and SATA if the application has un-expected performance that day?
- Manageability of FAST interfaces.
- Usability of FAST interfaces.
- Granular configuration policies associated with analysis.
- Future upgrades.
- Current implementations and how long and how effective are those.
- Can you improve performance without using SSD’s. Example with Fibre and SATA drives only.
- With future upgrades from FAST v1 to FAST v2 what happens with current user defined policies.
- Would Symcli scripts change with FAST implementation?
- During lockout periods (Where Symmetrix configuration cannot change, eg before a BIN file change) would FAST still operate.
To read some of the answers to the questions, as commented by Barry Burke on the StorageNerve Blog
FAST is a bit of a new subject for me. Any experts please feel free to correct me if my understanding of FAST is incorrect at any level.
Stay tuned for a series of FAST posts over the next few days talking about various topics and how FAST plays within those areas.