emc fast vp for unified storage systems · pdf fileemc fast vp for unified storage systems a...

EMC FAST VP for Unified Storage Systems

A Detailed Review

Abstract

This white paper introduces EMC® Fully Automated Storage Tiering for Virtual Pools (FAST VP) technology

and describes its features and implementation. Details on how to work with the product in the Unisphere™

operating environment are discussed, and usage guidance and major customer benefits are also included.

March 2011


A Detailed Review 2

Copyright © 2010, 2011 EMC Corporation. All rights reserved.

EMC believes the information in this publication is accurate as of its publication date. The information is

subject to change without notice.

THE INFORMATION IN THIS PUBLICATION IS PROVIDED “AS IS.” EMC CORPORATION

MAKES NO REPRESENTATIONS OR WARRANTIES OF ANY KIND WITH RESPECT TO THE

INFORMATION IN THIS PUBLICATION, AND SPECIFICALLY DISCLAIMS IMPLIED

WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Use, copying, and distribution of any EMC software described in this publication requires an applicable

software license.

For the most up-to-date listing of EMC product names, see EMC Corporation Trademarks on EMC.com

All other trademarks used herein are the property of their respective owners.

Part Number h8058.1


A Detailed Review 3

Table of Contents

Executive summary ............................................................................................ 4

Audience ...................................................................................................................................... 4

Introduction ......................................................................................................... 4

Storage tiers ........................................................................................................ 5

Extreme Performance Tier drives: Flash drives (for VNX and CX4 platforms) ........................ 5 Performance Tier drives: SAS (VNX) and Fibre Channel (CX4) .............................................. 5 Capacity Tier drives: NL-SAS (VNX) and SATA (CX4) ............................................................ 6

FAST VP operations ........................................................................................... 7

Storage pools ............................................................................................................................... 7

FAST VP algorithm ...................................................................................................................... 8

Statistics collection ................................................................................................................... 8 Analysis .................................................................................................................................... 8 Relocation................................................................................................................................. 8

Managing FAST VP at the storage pool level .............................................................................. 8

Automated scheduler ............................................................................................................... 9 Manual relocation ................................................................................................................... 10

FAST VP LUN management ...................................................................................................... 10

Tiering policies ....................................................................................................................... 11 Initial placement ..................................................................................................................... 12

Using FAST VP for file ...................................................................................... 12

Management .............................................................................................................................. 12

Best practices............................................................................................................................. 14

Automated Volume Manager rules ............................................................................................ 15

Guidance and recommendations .................................................................... 16

FAST VP and FAST Cache ....................................................................................................... 16

What drive mix is right for my I/O profile? .................................................................................. 17

Conclusion ........................................................................................................ 18

References ........................................................................................................ 19


A Detailed Review 4

Executive summary Fully Automated Storage Tiering for Virtual Pools (FAST VP) can lower total cost of ownership (TCO)

and increase performance by intelligently managing data placement at a sub-LUN level. When FAST VP is

implemented, the storage system measures, analyzes, and implements a dynamic storage-tiering policy

much faster and more efficiently than a human analyst could ever achieve.

Storage provisioning can be repetitive and time-consuming and produce uncertain results. It is not always

obvious how to match capacity to the performance requirements of a workload’s data. Even when a match

is achieved, requirements change, and a storage system’s provisioning will require constant adjustment.

Storage tiering puts drives of varying performance levels and cost into a storage pool. LUNs use the storage

capacity they need from the pool, on the devices with the required performance characteristics. FAST VP

collects I/O activity statistics at a 1 GB granularity (known as a slice). The relative activity level of each

slice is used to determine which slices should be promoted to higher tiers of storage. Relocation is initiated

at the user’s discretion through either manual initiation or an automated scheduler. Working at such a

granular level removes the need for manual, resource-intensive LUN migrations while still providing the

performance levels required by the most active dataset.

FAST VP is a licensed feature available on EMC® VNX™ series and CLARiiON

® CX4 series platforms.

The VNX series supports a unified approach to automatic tiering for both file and block data. CX4 systems

running release 30 and later are supported for the tiering of block data. FAST VP licenses are available a´

la carte for both platforms, or as part of a FAST Suite of licenses that offers complementary licenses for

technologies such as FAST Cache, Analyzer, and Quality of Service Manager.

This white paper introduces the EMC FAST VP technology and describes its features, functions, and

management.

Audience This white paper is intended for EMC customers, partners, and employees who are considering using the

FAST VP product. Some familiarity with EMC midrange storage systems is assumed. Users should be

familiar with the material discussed in the white papers Introduction to EMC VNX Series Storage Systems

and EMC VNX Virtual Provisioning.

Introduction Data has a lifecycle. As data progresses through its lifecycle, it experiences varying levels of activity.

When data is created, it is typically heavily used. As it ages, it is accessed less often. This is often referred

to as being temporal in nature. FAST VP is a simple and elegant solution for dynamically matching

storage requirements with changes in the frequency of data access. FAST VP segregates disk drives into the

following three tiers:

Extreme Performance Tier – Flash drives

Performance Tier – SAS drives for VNX platforms and Fibre Channel drives for CX4 platforms

Capacity Tier – Near-Line SAS (NL-SAS) drives for VNX platforms and SATA drives for CX4

platforms

You can use FAST VP to aggressively reduce TCO and/or to increase performance. A target workload that

requires a large number of Performance Tier drives can be serviced with a mix of tiers, and a much lower

drive count. In some cases, an almost two-thirds reduction in drive count is achieved. In other cases,

performance throughput can double by adding less than 10 percent of a pool’s total capacity in Flash

drives.


A Detailed Review 5

FAST VP has proven highly effective for a number of applications. Tests in OLTP environments with

Oracle1 or Microsoft SQL Server

2 show that users can lower their capital expenditure (by 15 percent to 38

percent), reduce power and cooling costs (by over 40 percent), and still increase performance by using

FAST VP instead of a homogeneous drive deployment.

FAST VP can be used in combination with other performance optimization software, such as FAST Cache.

A common strategy is to use FAST VP to gain TCO benefits while using FAST Cache to boost overall

system performance. There are other scenarios where it makes sense to use FAST VP for both purposes.

This paper discusses considerations for the best deployment of these technologies.

The VNX series of storage systems delivers even more value over previous systems by providing a unified

approach to auto-tiering for file and block data. FAST VP is available on the VNX5300™ and larger

systems. Now, file data served by VNX Data Movers can also use virtual pools and the same advanced data

services as block data. This provides compelling value for users who wish to optimize the use of high-

performing drives across their environment.

Storage tiers FAST VP can leverage two or all three storage tiers in a single pool. Each tier offers unique advantages in

performance and cost.

Extreme Performance Tier drives: Flash drives (for VNX and CX4 platforms)

Flash drives are having a large impact on the external-disk storage system market. They are built on solid-

state drive (SSD) technology that has no moving parts. The absence of moving parts makes these drives

highly energy-efficient, and eliminates rotational latencies. Therefore, migrating data from spinning disks

to Flash drives can boost performance and create significant energy savings.

Tests show that adding a small (single-digit) percentage of Flash capacity to your storage, while using

intelligent tiering products (such as FAST VP and FAST Cache), can deliver double-digit percentage gains

in throughput and response time performance in some applications. Flash drives can deliver an order of

magnitude better performance than traditional spinning disks when the workload is IOPS-intensive and

response-time sensitive. They are particularly effective when small random-read I/Os are part of the profile,

as they are in many transactional database applications. On the other hand, bandwidth-intensive

applications perform only slightly better on Flash drives than on spinning drives.

Flash drives have a higher per-gigabyte cost than traditional spinning drives. To receive the best return,

you should use Flash drives for data that requires fast response times and high IOPS. The best way to

optimize the use of these high-performing resources is to allow FAST VP to migrate data to Flash drives at

a sub-LUN level.

Performance Tier drives: SAS (VNX) and Fibre Channel (CX4)

Traditional spinning drives offer high levels of performance, reliability, and capacity. These drives are

based on industry-standardized, enterprise, mechanical hard-drive technology that stores digital data on a

series of rapidly rotating magnetic platters.

The Performance Tier includes 10k and 15k rpm spinning drives, which are available on all EMC midrange

storage systems. They have been the performance medium of choice for many years. They also have the

highest availability of any mechanical storage device. These drives continue to serve as a valuable storage

tier, offering high all-around performance including consistent response times, high throughput, and good

bandwidth, at a midtier price point.

1 Leveraging Fully Automated Storage Tiering (FAST) with Oracle Database Applications EMC white

paper 2 EMC Tiered Storage for Microsoft SQL Server 2008—Enabled by EMC Unified Storage and EMC Fully

Automated Storage Tiering (FAST) EMC white paper


A Detailed Review 6

The VNX series and CX4 series use different attach technologies. VNX systems use a 6 Gb/s SAS back

end while the CX4 series uses a 4 Gb/s Fibre Channel back end. Despite the differences in speed, the drive

assemblies on both series are very similar. When sizing a solution with these drives, you need to consider

the rotational speed of the drive. However, the interconnect to the back end does not affect rule-of-thumb

performance assumptions for the drive.

This also applies to the different drive form factors. The VNX series offers 2.5-inch and 3.5-inch form-

factor SAS drives. The same IOPS and bandwidth performance guidelines apply to 2.5-inch 10k drives and

3.5-inch 10k inch drives.

FAST VP differentiates tiers by drive type. However, it does not take rotational speed into consideration.

We strongly encourage you to use one rotational speed for each drive type within a pool. If multiple

rotational-speed drives exist in the array, multiple pools should be implemented as well.

Capacity Tier drives: NL-SAS (VNX) and SATA (CX4)

Using capacity drives can significantly reduce energy use and free up capacity in higher storage tiers.

Studies have shown that 60 percent to 80 percent of the capacity of many applications has little I/O activity.

Capacity drives can cost about four times less than performance drives on a per-gigabyte basis, and a small

fraction of the cost of Flash drives. They consume up to 96 percent less power per TB than performance

drives. This offers a compelling opportunity for TCO improvement considering both purchase cost and

operational efficiency.

Capacity drives are designed for maximum capacity at a modest performance level. They have a slower

rotational speed than Performance Tier drives. NL-SAS drives for the VNX series have a 7.2k rotational

speed, while SATA drives available on the CX4 series come in 7.2k and 5.4k rpm varieties. The 7.2k rpm

drives can deliver bandwidth performance within 20 percent to 30 percent of performance class drives.

However, the reduced rotational speed is a trade-off for significantly larger capacity. For example, the

largest Capacity Tier drives are 2 TB, compared to the 600 GB Performance Tier drives and 200 GB Flash

drives. These Capacity Tier drives offer roughly half the IOPS/drive of Performance Tier drives.


A Detailed Review 7

Table 1. Feature tradeoffs for Flash, Performance, and Capacity drives

Flash drives Performance (SAS, FC) Capacity (NL-SAS, SATA) P

erfo

rman

ce -High IOPS/GB and low latency

-Sole use response time <1-5 ms

-Multi-access response time < 10

ms

-High bandwidth with contending

workloads

-Sole use response time ~5 ms

-Multi-access response time 10-50

ms

-Low IOPS/GB

-Sole use response time 7-10 ms

-Multi-access response time up to

100 ms

-Leverage storage array SP cache for

sequential and large block access

Str

eng

ths

-Provide extremely fast access

for reads

-Can execute multiple sequential

streams better than SAS/FC

-Sequential reads can leverage

read-ahead

-Sequential writes can leverage

system optimizations favoring

disks

-Read/write mixes give predictable

performance

-Large I/O is serviced fairly

efficiently

-Sequential reads can leverage read-

ahead

-Sequential writes can leverage

system optimizations favoring disks

Lim

itat

ion

s -Writes slower than reads

-Heavy concurrent writes affect

read rates

-Single-threaded large sequential

I/O equivalent to SAS/FC

-Uncached writes are slower than

reads

-Long response times when under

heavy write loads

-Not as good as FC at handling

multiple streams

FAST VP operations FAST VP operates by periodically relocating the most active data up to the highest available tier (typically

the Extreme Performance or Performance Tier). To ensure sufficient space in the higher tiers FAST VP

relocates less active data to lower tiers (Performance or Capacity Tiers). FAST VP works at a granularity

of 1 GB. Each 1 GB block of data is referred to as a “slice.” When FAST VP relocates data, it will move

the entire slice to a different storage tier.

Storage pools Heterogeneous storage pools are the framework that allows FAST VP to fully utilize each of the storage

tiers discussed. LUNs can then be created at the pool level. These pool LUNs are no longer bound to a

single storage tier; instead, they can be spread across different storage tiers within the same pool.

Figure 1. Heterogeneous storage pool concept


A Detailed Review 8

LUNs must reside in a pool to be eligible for FAST VP relocation. Pools support thick LUNs and thin

LUNs. Thick LUNs are high-performing LUNs that use contiguous logical block addressing on the

physical capacity assigned from the pool. Thin LUNs use a capacity-on-demand model for allocating drive

capacity. Capacity usage is tracked at a finer granularity than thick LUNs to maximize capacity

optimizations. FAST VP is supported on both thick LUNs and thin LUNs.

RAID groups are by definition homogeneous and therefore are not eligible for tiering. LUNs in RAID

groups can be migrated to pools using LUN Migration. For a more in-depth discussion of pools, please see

the white paper EMC VNX Virtual Provisioning - Applied Technology.

FAST VP algorithm FAST VP uses three strategies to identify and move the correct slices to the correct tiers: statistics

collection, analysis, and relocation.

Statistics collection

A slice of data is hotter (more activity) or colder than another slice of data based on the relative activity

level of the two slices. Activity level is determined by counting the number of I/Os for each slice. FAST

VP maintains a cumulative I/O count and “weights” each I/O by how recently it arrived. This weight

deteriorates over time. New I/O is given full weight. After approximately 24 hours, the same I/O carries

about half-weight. After a week, the same I/O carries very little weight. Statistics are continuously

collected (as a background task) for all pool LUNs.

Analysis

Once per hour, the collected data is analyzed. This analysis produces a rank ordering of each slice within

the pool. The ranking progresses from the hottest slices to the coldest slices relative to the other slices in

the same pool. (For this reason, a hot slice in one pool may comparable to a cold slice in another pool.)

There is no system-level threshold for activity level.

Relocation

During user-defined relocation windows, 1 GB slices are promoted according to both the rank ordering

performed in the analysis stage and the tiering policy set by the user. During relocation, FAST VP

relocates higher-priority slices to higher tiers; slices are relocated to lower tiers only if the space they

occupy is required for a higher-priority slice. This way, FAST VP fully utilizes the highest-performing

spindles first. Lower-tier spindles are utilized as capacity demand grows. Relocation can be initiated

manually or by a user-configurable, automated scheduler.

Managing FAST VP at the storage pool level FAST VP properties can be viewed and managed at the pool level. Figure 2 shows the tiering information

for a specific pool.


A Detailed Review 9

Figure 2. Storage Pool Properties window

The Tier Status section of the window shows FAST VP relocation information specific to the pool selected.

Scheduled relocation can be selected at the pool level from the drop-down menu labeled Auto-Tiering.

This can be set to either Automatic or Manual. Users can also connect to the array-wide relocation

schedule using the button located in the top right corner. This is discussed in the “Automated scheduler”

section. Data Relocation Status displays what state the pool is in with regards to FAST VP. The Ready

state indicates that relocation can begin on this pool at any time. The amount of data bound for a slower

tier is shown next to Data to Move Down and that amount of data bound for a faster tier is listed next to

Data to Move Up. Below that is the estimated time required to migrate all data within the pool to the

appropriate tier.

In the Tier Details section, users can see the exact distribution of their data. This panel shows all tiers of

storage residing in the pool. Each tier then displays the amount of data to be moved up and down, the total

capacity allocated (user capacity), and the consumed capacity.

Automated scheduler

The scheduler launched from the Pool Properties dialog box’s Relocation Schedule button is shown in

Figure 2. To this end, relocations can be scheduled to occur automatically. It is recommended that

relocations be scheduled during off-hours to minimize any potential performance impact the relocations

may cause. Figure 3 shows the Manage Auto-Tiering window.


A Detailed Review 10

Figure 3. Manage Auto-Tiering window

The data relocation schedule shown in Figure 3 initiates relocations everyday at 10:00 PM for a duration of

eight hours (6:00 AM). You can select the days that the relocation schedule should run. In this example,

relocations run seven days a week, which is the default setting. From this status window, you can also

control the data relocation rate. The default rate is set to Medium so as not to significantly impact host I/O.

This rate will relocate data up to 300-400 GB per hour3.

Manual relocation

Manual relocation is initiated by the user through either the Unisphere GUI or the CLI. It can be initiated

at any time. When a manual relocation is initiated, FAST VP performs analysis on all statistics gathered,

independent of its regularly scheduled hourly analysis, prior to beginning the relocation. This ensures that

up-to-date statistics and settings are properly accounted for prior to relocation.

Although the automatic scheduler is an array-wide setting, manual relocation is enacted at the pool level

only. Common situations when users may want to initiate a manual relocation on a specific pool include

the following:

When reconfiguring the pool (for example, adding a new tier of drives)

When LUN properties have been changed and the new priority structure needs to be realized

immediately

As part of a script for a finer-grained relocation schedule

FAST VP LUN management Some FAST VP properties are managed at the LUN level. Figure 4 shows the tiering information for a

single LUN.

3 This rate depends on system type, array utilization, and other tasks competing for array resources. High

utilization rates may reduce this relocation rate.



Figure 4. LUN Properties window

The Tier Details section displays the current distribution of 1 GB slices within the LUN. The Tiering

Policy section displays the available options for tiering policy.

Tiering policies

There are four tiering policies available within FAST VP:

Auto-tier (recommended)

Highest available tier

Lowest available tier

No data movement

Auto-tier Auto-tier is the default setting for all pool LUNs upon their creation. FAST VP relocates slices of these

LUNs based on their activity level. Slices belonging to LUNs with the auto-tier policy have second priority

for capacity in the highest tier in the pool after LUNs set to the highest tier.

Highest available tier The highest available tier setting should be selected for those LUNs which, although not always the most

active, require high levels of performance whenever they are accessed. FAST VP will prioritize slices of a

LUN with highest available tier selected above all other settings.

Lowest available tier Lowest available tier should be selected for LUNs that are not performance- or response-time-sensitive.

FAST VP will maintain slices of these LUNs on the lowest storage tier available regardless of activity

level.

No data movement No data movement may only be selected after a LUN has been created. FAST VP will not move slices

from their current positions once the no data movement selection has been made.



Initial placement

The tiering policy chosen also affects the initial placement of a LUN’s slices within the available tiers.

Initial placement with the pool set to auto-tier will result in the data being distributed across all storage

tiers available within the pool. LUNs set to highest available tier will have their component slices placed

on the highest tier that has capacity available. LUNs set to lowest available tier will have their component

slices placed on the lowest tier that has capacity available.

LUNs with the tiering policy set to no data movement will use the initial placement policy of the setting

preceding the change to no data movement. For example, a LUN that was previously set to highest tier but

is currently set to no data movement will still take its initial allocations from the highest tier possible.

Using FAST VP for file In the VNX Operating Environment for file 7, file data is supported on LUNs created in pools with FAST

VP configured on both the VNX unified and EMC Symmetrix® with gateway systems.

Management The process for implementing FAST VP for file begins by provisioning LUNs from a pool with mixed tiers

(or across tiers for Symmetrix) that are placed in the protected File Storage Group. Rescanning the storage

systems from the System tab in Unisphere starts a diskmark that makes the LUNs available to VNX file

storage. The rescan automatically creates a pool for file using the same name as the corresponding pool for

block. Additionally it will create a disk volume in a 1:1 mapping for each LUN that was added to the File

Storage Group. A file system can then be created from the pool for file on the disk volumes. The FAST VP

policy that has been applied to the LUNs presented to file will operate as it does for any other LUN in the

system, dynamically migrating data between storage tiers in the pool.

Figure 5. FAST VP for file



FAST VP for file is supported in Unisphere and the CLI. All relevant Unisphere configuration wizards

support a FAST VP configuration except for the VNX Provisioning Wizard (known pre-VNX as the

Celerra Provisioning Wizard). FAST VP properties can be seen within the properties pages of pools for

file (see Figure 6), and property pages for volumes and file systems (see Figure 7 on page 14) but they can

only be modified through the block pool or LUN areas of Unisphere. On the File System Properties pages

the FAST VP tiering policy is listed in the Advanced Data Services section along with whether thin,

compression, or mirrored is enabled. For more information on thin and compression reference the white

papers EMC VNX Virtual Provisioning and EMC VNX Deduplication and Compression.

New disk type options of the mapped disk volumes are as follows:

LUNs in a storage pool with a single disk type

o Extreme Performance (Flash drives)

o Performance (10k and 15k rpm SAS drives)

o Capacity (7.2k rpm NL-SAS)

LUNs in a storage pool with multiple disk types (used with FAST VP)

o Mixed

LUNs that are mirrored (mirrored means remote mirrored through MirrorView™ or

RecoverPoint)

o Mirrored_mixed

o Mirrored_performance

o Mirrored_capacity

o Mirrored_Extreme Performance

Figure 6. File Storage Pool Properties window



Figure 7. File System Properties window

Best practices VNX file configurations, for both Symmetrix and CLARiiON, will not expressly forbid mixing LUNs with

different data service attributes, although users will be warned that it is not recommended to mix due to the

impact of spreading a file system across, for example, a thin and a thick LUN or LUNs with different

tiering policies.

Note: VNX file configurations will not allow mixing of mirrored and non-mirrored types in a pool (if attempted,

the disk mark will fail).

Unless you are managing file systems that are used for archival data, it is not recommended to use block

thin provisioning or compression on VNX LUNs used for file system allocations. If thin provisioning for

block is used, closely monitor back-end physical space usage from the block perspective to avoid a

situation where space runs out.

While Automated Volume Manager will attempt to ensure consistent performance for file systems that span

LUNs with different data service attributes, it is recommended that this be managed in one of two ways:

Optimize ease of use on the file side; configure LUNs with different service attributes in separate

pools and make these pools available independently so they can be assigned individually to

specific file systems per the demand of each file system.

Optimize ease of use on the block side and optimize the use of resources on the whole system;

define one or two large pools for use across all users of the system, which will typically have

different service attributes associated with different LUNs.

When used for file systems, these different LUN types will appear in a single file-side pool and can be

managed with Manual Volume Management to appear independently such that homogenous file systems



can be configured in the manual pools. This approach allows users to leverage the aggregation effect of

large pools.

For best performance it is recommended to put file data in a separate pool from block data. This makes the

pool have a more common I/O profile that typically performs better and is easier to manage and

troubleshoot if performance issues arise.

Automated Volume Manager rules Automated Volume Manager (AVM) rules are different when creating a file system with underlying pool

LUNs as opposed to file systems with underlying RAID group LUNs. The rules for AVM with underlying

pool LUNs are as follows:

For VNX, the following rules apply:

1. Concatenation will be used. Striping will not be used (striping is done at the block pool

level).

2. Unless requested, slicing will not be used. Slicing is turned off by default because slicing

allows multiple file systems on one LUN. Since data services can only be changed on the

LUN level, all file systems on a LUN are affected by a single change. Slicing can cause

the loss of granularity to change data services on the individual file system level.

Turning slicing off forces the file system to use the entire LUN. LUNs should be sized

with consideration to the size of the file system that will be created on them.

3. AVM checks for free disk volumes, and sorts them by thin and thick disk volumes.

4. AVM checks for free disk volumes:

If there are no free disk volumes and the slice option is set to no, there is not

enough space available and the request fails.

If there are free disk volumes:

a. AVM first checks for thick disk volumes that satisfy the size request.

b. If not found, AVM then checks for thin disk volumes that satisfy the

size request.

c. If still not found, AVM combines thick and thin disk volumes to find

ones that satisfy the size request.

5. If one disk volume satisfies the size request exactly, AVM takes the selected disk volume

and uses the whole disk to build the file system.

6. If a larger disk volume is found that is a better fit than any set of smaller disks, then

AVM uses the larger volume.

7. If multiple disk volumes satisfy the size request, AVM sorts the disk volumes from

smallest to largest, and then sorts in alternating SP A and SP B lists. Starting with the

first disk volume, AVM searches through a list for matching data services until the size

request is met. If the size request is not met, AVM searches again but ignores the data

services.

For VNX gateway with Symmetrix, the following rules apply:

1. Unless requested, slicing will not be used.

2. AVM checks for free disk volumes, and sorts them by thin and thick disk volumes for the

purpose of striping together the same type of disk volumes:

If there are no free disk volumes and the slice option is set to no, there is not

enough space available and the request fails

If there are free disk volumes:

a. AVM first checks for a set of eight disk volumes.

b. If a set of eight is not found, AVM then looks for a set of four disk

volumes.

c. If a set of four is not found, AVM then looks for a set of two disk

volumes.

d. If a set of two is not found, AVM finally looks for one disk volume.

3. When free disk volumes are found:

a. AVM first checks for thick disk volumes that satisfy the size request, which can

be equal to or greater than the file system size. If thick disk volumes are



available, AVM first tries to stripe the thick disk volumes that have the same

disk type. Otherwise, AVM stripes together thick disk volumes that have

different disk types.

b. If thick disks are not found, AVM then checks for thin disk volumes that satisfy

the size request. If thin disk volumes are available, AVM first tries to stripe the

thin disk volumes that have the same disk type, where “same” means the single

disk type of the pool in which it resides. Otherwise, AVM stripes together thin

disk volumes that have different disk types.

c. If thin disks are not found, AVM combines thick and thin disk volumes to find

ones that satisfy the size request.

4. If neither thick nor thin disk volumes satisfy the size request, AVM then checks for

whether striping of one same disk type will satisfy the size request, ignoring whether the

disk volumes are thick or thin.

5. If still no matches are found, AVM checks whether slicing was requested.

a. If slicing was requested, then AVM checks whether any stripes exist that satisfy

the size request. If yes, then AVM slices an existing stripe.

b. If slicing was not requested, AVM checks whether any free disk volumes can be

concatenated to satisfy the size request. If yes, AVM concatenates disk volumes,

matching data services if possible, and builds the file system.

6. If still no matches are found, there is not enough space available and the request fails.

Managing Volumes and File Systems with VNX AVM provides further information on using AVM with

mapped pools.

Guidance and recommendations The following table displays the total number of LUNs that can be set to leverage FAST VP based on the

array model. These limits are the same as the total number of pool LUNs per system. Therefore, all pool

LUNs in any given system can leverage FAST VP.

Table 2. FAST VP LUN limits

Array model VNX5300

CX4-120

VNX5500™

CX4-240

VNX5700™

CX4-480

VNX7500™

CX4-960

Maximum

number of pool

LUNs

512 1,024 2,048 2,048

FAST VP and FAST Cache FAST Cache allows the storage system to provide Flash drive class performance to the most heavily

accessed chunks of data across the entire system. FAST Cache absorbs I/O bursts from applications,

thereby reducing the load on back-end hard disks. This improves the performance of the storage solution.

For more details on this feature, refer to the EMC CLARiiON, Celerra Unified, and VNX FAST Cache

white paper available on Powerlink.



The following table compares the FAST VP and FAST Cache features.

Table 3. Comparison between the FAST VP and FAST Cache features

FAST Cache FAST VP

Enables Flash drives to be used to extend the

existing caching capacity of the storage system.

Leverages pools to provide sub-LUN tiering,

enabling the utilization of multiple tiers of storage

simultaneously

Has finer granularity – 64 KB Less granular compared to FAST Cache – 1 GB

Copies data from HDDs to Flash drives when

they get accessed frequently

Moves data between different storage tiers based on

a weighted average of access statistics collected

over a period of time

Is designed primarily to improve performance While it can improve performance, it is primarily

designed to improve ease of use and reduce TCO

FAST Cache and the FAST VP sub-LUN tiering features can be used together to yield high performance

and improved TCO from the storage system. As an example, in scenarios where limited Flash drives are

available, they can be used to create FAST Cache and the FAST VP sub-LUN tiering feature can be used

on a pool consisting of Performance and Capacity disk drives. From a performance point of view, FAST

Cache will dynamically provide performance benefit to any bursty data while FAST VP will move warmer

data to Performance drives and colder data to Capacity drives. From a TCO perspective, FAST Cache with

a small number of Flash drives serves the data that is accessed most frequently, while FAST VP sub-LUN

tiering with Fibre Channel and SATA drives can optimize disk utilization and efficiency.

As a general rule, FAST Cache should be used in cases where storage system performance needs to be

improved immediately for burst-prone data. On the other hand, FAST VP optimizes storage system TCO as

it moves data to the appropriate storage tier based on sustained data access and demands over time. FAST

Cache focuses on improving performance while FAST VP focuses on improving TCO. Both features are

complementary to each other and help in improving performance and TCO.

The FAST Cache feature is storage-tier-aware and works with the FAST VP sub-LUN tiering feature to

make sure that the storage system resources are not wasted by unnecessarily copying data to FAST Cache if

it is already on a Flash drive. If FAST VP moves a chunk of data to the Extreme Performance Tier (which

consists of Flash drives) , FAST Cache will not promote that chunk of data into FAST Cache, even if FAST

Cache criteria is met for promotion. This ensures that the storage system resources are not wasted in

copying data from one Flash drive to another.

A general recommendation for the initial deployment of Flash drives in a storage system is to use them for

FAST Cache. However, in certain cases FAST Cache does not offer the most efficient use of Flash drives.

FAST Cache tracks I/Os that are smaller than 128 KB, and requires multiple hits to 64 KB chunks to

initiate promotions from rotating disk to FAST Cache. Therefore, I/O profiles that do not meet this criteria

are better served by Flash in a pool or RAID group.

What drive mix is right for my I/O profile? As previously mentioned, it is common for a small percentage of overall capacity to be responsible for

most of the I/O activity. This is known as skew. Analysis of an I/O profile may indicate that 85 percent of

the I/Os to a volume only involve 15 percent of the capacity. The resulting active capacity is called the

working set. Software like FAST VP and FAST Cache keeps the working set on the highest-performing

drives.

It is common for OLTP environments to yield working sets of 20 percent or less of their total capacity.

These profiles hit the “sweet spot” for FAST and FAST Cache. The white papers Leveraging Fully

Automated Storage Tiering (FAST) with Oracle Database Applications and EMC Tiered Storage for

Microsoft SQL Server 2008—Enabled by EMC Unified Storage and EMC Fully Automated Storage Tiering

(FAST) discuss performance and TCO benefits for several mixes of drive types.



Other I/O profiles, like Decision Support Systems (DSS), may have much larger working sets. In these

cases, FAST VP may be used to deploy Flash drives because DSS workloads are not typically FAST

Cache-friendly. Capacity Tier drives may be used to lower TCO. The white paper Leveraging EMC Unified

Storage System Dynamic LUNs for Data Warehouse Deployments on Powerlink offers analysis on the use

of storage pools and FAST VP.

At a minimum, the capacity across the Performance Tier and Extreme Performance Tier (and/or FAST

Cache) should accommodate the working set. However, capacity is not the only consideration. The spindle

count of these tiers needs to be sized to handle the I/O load of the working set. Basic techniques for sizing

disk layouts based on IOPS and bandwidth are available in the EMC VNX Fundamentals for Performance

and Availability white paper on Powerlink.

As discussed above, using FAST Cache as your first line of Flash-drive deployment is a practical approach

when the I/O profile is amenable to its use. In circumstances where an I/O profile is not FAST-Cache

friendly, Flash can be deployed in a pool or RAID group instead.

Performance Tier drives are versatile in handling a wide spectrum of I/O profiles. Therefore, we highly

recommend that you include Performance Tier drives in each pool. FAST Cache can be an effective tool

for handling a large percentage of activity, but inevitably, there will be I/Os that have not been promoted or

are cache misses. In these cases, Performance Tier drives offer good performance for those I/Os.

Performance Tier drives also facilitate faster promotion of data into FAST Cache by quickly providing

promoted 64 KB chunks to FAST Cache. This minimizes FAST Cache warm-up time as some data gets hot

and other data goes cold. Lastly, if FAST Cache is ever in a degraded state due to a faulty drive, the FAST

Cache will become read only. If the I/O profile has a significant component of random writes, these are

best served from Performance Tier drives as opposed Capacity drives.

Capacity drives can be used for “everything else.” This often equates to comprising 60 percent to 80

percent of the pool’s capacity. Of course, there are profiles with low IOPS/GB and or sequential workload

that may result in the use of a higher percentage of Capacity Tier drives.

EMC Professional Services and qualified partners can be engaged to assist with properly sizing tiers and

pools to maximize investment. They have the tools and expertise to make very specific recommendations

for tier composition based on an existing I/O profile.

Conclusion Through the use of FAST VP, users can remove complexity and management overhead from their

environments. FAST VP utilizes Flash, performance, and capacity drives (or any combination thereof)

within a single pool. LUNs within the pool can then leverage the advantages of each drive type at the 1 GB

slice granularity. This sub-LUN-level tiering ensures that the most active dataset resides on the best-

performing drive tier available, while maintaining infrequently used data on lower-cost, high-capacity

drives.

Relocations occur without user interaction on a predetermined schedule, making FAST VP a truly

automated offering. In the event that relocation is required on-demand, FAST VP relocation can be

invoked through Unisphere on an individual pool.

Both FAST VP and FAST Cache work by placing data segments on the most appropriate storage tier based

on their usage pattern. These two solutions are complementary because they work on different granularity

levels and time tables. Implementing both FAST VP and FAST Cache can significantly improve

performance and reduce cost in the environment.



References The following white papers are available on Powerlink:

EMC CLARiiON Storage System Fundamentals for Performance and Availability

EMC CLARiiON Best Practices for Performance and Availability

EMC CLARiiON, Celerra Unified, and VNX FAST Cache

EMC Unisphere: Unified Storage Management Solution

EMC VNX Virtual Provisioning

An Introduction to EMC CLARiiON and Celerra Unified Platform Storage Device Technology

EMC Tiered Storage for Microsoft SQL Server 2008—Enabled by EMC Unified Storage and EMC

Fully Automated Storage Tiering (FAST)

Leveraging Fully Automated Storage Tiering (FAST) with Oracle Database Applications

Leveraging EMC Unified Storage System Dynamic LUNs for Data Warehouse Deployments

emc fast vp for unified storage systems · pdf fileemc fast vp for unified storage systems a...

Documents