Best Practices of SmartTier and SmartThin in Oracle 11g Database Environments … · 2014-11-28 ·...

Huawei Proprietary and Confidential Copyright © Huawei Technologies Co., Ltd. 1

Best Practices of SmartTier and SmartThin in Oracle 11g Database Environments

Best Practices of SmartTier

and SmartThin in Oracle

11g Database Environments

This document describes why and how SmartTier and SmartThin should be applied to Oracle

OLTP applications. SmartTier, Huawei's dynamic storage tiering feature, identifies hot and cold

data by I/O monitoring and migrates data for performance acceleration. When SmartTier is used

for Oracle OLTP application, only six 200-GB SLC SSDs are used to compose the high

performance tier to store hotspot data, improving throughput by 46% and reducing latency by 62%.

SmartThin, a mechanism of allocating storage space on demand, allows users to create thin LUNs

for storage resource savings and initial investment reduction. Using SmartThin has a minor impact

on service throughput and latency.

Zou Yanping

[email protected]

Storage Solution Development Dept, IT, Enterprise Business Group

May 2013 V1.0

Huawei Technologies Co., Ltd.

Ltd.

mailto:[email protected]



HUAWEI T series unified storage systems (S2600T, S5500T, S5600T, S5800T and

S6800T) incorporate block- and file-level data with unified storage protocols and

management interfaces. With its industry-leading performance and multiple efficiency

improving mechanisms, the T series provides customers with comprehensive

high-performance storage solutions.

Features

Converged and Unified

Convergence of files and blocks: Unifies SAN and NAS protocols and co-processes

structured and unstructured data.

Converged storage protocols: Supports iSCSI, Fibre Channel, FCoE, NFS, CIFS,

HTTP, and FTP.

Central management interface: interoperates with an easy-to-use and wizard-driven

graphical user interface (GUI) for the central management on files and blocks and the

facilitated system configuration.

Flexible and Reliable

Easy upgrade: Supports easy upgrade from block storage to unified storage.

Hot swap design: Allows controllers, fans, power supplies, interface modules, and

disks to be swapped and expanded online without affecting services.

Multiple optional disk types: Includes SAS, NL SAS, SATA, SSD, and Fibre

Channel.

OceanStor T Series



High-reliability architecture: Provides all redundant components to eliminate single

points of failure. Data vault and file system mirroring technologies further improve

system reliability.

Leading I/O port scalability and flexibility: Supports up to 12 I/O modules, providing

48 I/O ports. Supported port types include 4 Gbit/s or 8 Gbit/s Fibre Channel, 1

Gbit/s iSCSI, 10 Gbit/s iSCSI (TOE), 10 Gbit/s FCoE, and 6 Gbit/s SAS.

Data protection: Provides seamless integration with Enterprise Vault and NetBackup

to achieve disaster recovery based on remote replication.

Application-aware optimization: Collaborates with the HostAgent software to

implement application-level backup, DR, and DR verification, and supports

mainstream application systems such as Oracle, DB2, Exchange Server, and SQL

Server.

Cost-effective and Efficient

SmartCache: Uses flash drives to extend cache, improving system performance and

boosting running speed of hybrid workloads by several times.

SmartThin: Supports automatic capacity expansion, improving disk utilization and

reducing the initial purchase cost.

SmartTier: Dynamically and transparently migrates data between various storage

media based on file access frequency or a time policy, reducing investments.

SmartQoS: Dynamically allocates storage system resources to obviate resource

contention among applications and meet performance goals of specified applications.

Proactive remote service: Offers the unique Cloud Service that provides the health

check, warning, and alarm notification functions, helping maintenance engineers

quickly troubleshoot system faults.



SmartTier is Huawei's automatic storage tiering technology used for its OceanStor T series

unified storage systems and OceanStor 18000 series enterprise storage systems. As shown

in the following figure, SmartTier allows three types of disks to reside in the same storage

pool and uses storage space from different disks to create LUNs for use by upper-layer

applications. By monitoring I/O access frequencies, SmartTier migrates frequently

accessed data blocks to the high performance tier and infrequently accessed ones to the

capacity tier. In so doing, SmartTier efficiently improves the performance of applications

running on the storage tiers and maximizes customer ROI.

Major Functions

Performance statistics collection: By calculating the values of certain performance

indicators (including read/write frequency and I/O size), SmartTier determines the

activity levels of data blocks and identifies them as hot or cold data blocks. All the

data blocks are monitored and analyzed in real time.

Performance analysis: Based on the collected performance statistics, SmartTier

analyzes data blocks and ranks them from hottest to coldest. The latest analysis and

ranking results determine the target tiers of data blocks.

Data migration: Based on the ranking results and data migration policies, SmartTier

migrates data blocks that rank high to the high performance or performance tier and

those that rank low to the capacity tier.

Performance prediction: SmartTier can check the performance statistics of a storage

pool to analyze the balance of workloads at the LUN level. This helps provide an

optimal policy model of storage layer capacity that strikes a balance between

performance and cost. Users can reallocate disks among all storage tiers based on the

What Is SmartTier



capacity policy. The performance statistics function must be enabled for policy model

to be displayed.



SmartThin is a storage feature that allocates storage space on demand. It is also used by

the OceanStor T series and OceanStor 18000 series. SmartThin employs virtualization

technologies. First, capacity-on-write technology allocates storage space from the storage

pool when receiving write requests. Then, direct-on-time technology writes data into the

allocated space. SmartThin enables users to create thin LUNs that are allocated storage

space on demand, saving considerable storage resources to reduce initial investments.

Major functions

On-demand space allocation: For example, when a host requests writing 1 MB data

into the storage array, the thin LUN is allocated only 1 MB storage space.

Online expansion: LUNs can be online expanded without affecting host services. The

space for expanding a thin LUN is allocated from idle space. Therefore, the expansion

has no influence on the original storage area.

Larger visible capacity than available capacity: When a thin LUN is created, it can be

allocated more capacity than it can actually use. For example, the storage pool

capacity is 10 TB, but SmartThin allows a thin LUN that has up to 64 TB capacity.

Space reclamation: SmartThin supports two methods of releasing storage space: SCSI

command (unmap) and zero data.

What Is SmartThin



SCSI command releasing: After a host sends an unmap command, the corresponding

thin LUN reclaims the specified storage space.

Zero data releasing: After a host sends all-zero data to the storage array, the

corresponding thin LUN reclaims storage space in the requested scope.



In an online transactional processing (OLTP) system, Oracle database needs to process

intensive random I/Os, and a number of access hot spots exist. However, the capacity

increase of traditional enterprise-class disks cannot keep pace with the performance of

them. The enterprise-class disks provide excessive capacity, inferior performance, and low

storage utilization, failing to meet the needs of Oracle database. Using SmartTier on the

storage for Oracle database can notably increase the performance of Oracle database.

The following table lists the top five waiting events in an Oracle OLTP database system

that runs on traditional disks (10,000 rpm SAS disks). As shown in the table, 96.87% of

the execution time is spent waiting for the db file sequential read event, which takes an

average of 15 ms. The high latency of the traditional disks undoubtedly causes the

performance bottleneck in the OLTP system.

Event Waits Time(s) Avg wait(ms) %DB time Wait Class

Db file

sequential read 426,694 6,492 15 96.87 User I/O

DB CPU 255 5.30

db file parallel

read 729 58 79 0.86 User I/O

log file sync 39,398 30 1 0.45 Commit

gc cr grant 2-way 56,407 18 0 0.27 Cluster

The following figure shows the read requests distribution of a LUN that stores a large

database tables and indexes. The LBAs of the LUN are evenly divided into 100 parts. The

read I/Os on each part are calculated to identify the read hot spots on the LUN.

Why SmartTier

OLTP

Bottleneck

OLTP Hot

Spots



The following table shows the changes to the capacity and performance of traditional

enterprise-class disks over the past five years. As shown in the table, the capacity

increased every year, but the performance stayed unchanged. If users choose traditional

disks, then a large number of disks must be purchased to provide high-performance

random I/O access. In addition, a large volume of storage space is excessive and cannot be

used by other services, because doing so adversely affects the database performance. Both

storage tiering and all-SSD storage can greatly improve database system performance.

Given the high costs of all-SSD solutions, storage tiering is the best choice for Oracle

OLTP databases.

Year Capacity RPM OLTP Throughput

OLAP Throughput

2008 73 GB 15k 200 IOPS 30 MB/s

2009 146 GB 15k 200 IOPS 30 MB/s

2010 300 GB 15k 200 IOPS 30 MB/s

2011 600 GB 15k 200 IOPS 30 MB/s

2012 900 GB 15k 200 IOPS 30 MB/s

The SmartTier feature can accelerate the performance of Oracle OLTP database system.

As the amount of data is around 1.5 TB, six 200-GB SLC SSDs can comprise a high

performance tier to store hotspot data of the OLTP system. The following figure compares

the database performance before and after SmartTier is enabled.

The following figure shows the transactions per minute (TPM) of the thick LUNs before

and after SmartTier is enabled. The TPM value peaks 245,336 when SmartTier is disabled.

After SmartTier is enabled, the TPM value soars to 360,262, an increase of 46%.

Performance

Acceleration

Capacity and

Performance of

Traditional Disks



The following figure shows the IOPS of thick LUNs before and after SmartTier is enabled.

The IOPS value peaks 8428 when SmartTier is disabled. After SmartTier is enabled, the

IOPS value soars to 14,878, an increase of 77%.

The following figure shows the SwingBench transaction average latency of thick LUNs

before and after SmartTier is enabled. The latency is 347 ms when SmartTier is disabled

and decreases by 62% to 99 ms after SmartTier is enabled.

The following figure shows the TPM of thick LUNs before and after SmartTier is enabled.

As shown in the figure, enabling SmartTier increases the TPM by 46%.



The following figure shows the latency of the db file sequential read wait event before

and after SmartTier is enabled. As shown in the figure, enabling SmartTier decreases the

latency of the db file sequential read wait event by 74%.



To enable SmartTier in an Oracle database, first add SSDs to the storage system, then add

them to the storage pool that needs performance acceleration, select a monitoring policy

and migration policy, and start the migration.

The following procedure describes how to enable SmartTier for a LUN on the ISM.

Step1 Import and activate the SmartTier license.

Log in to the ISM of the T series from Internet Explorer. Click the Settings icon on the

right and click License Management.

Choose License Management, click Inactive License, click Import, and click Open.

Click Activate.

Step2 Select the area where SmartTier is enabled.

Go to the Resource Allocation page, click Storage Pool, select a pool for which

SmartTier has been enabled such as USR, and click Expand.

Select the tier that requires capacity expansion. For example, select Tier 0 storage. Then

select a RAID policy and hot spare policy for the tier and add disks to the tier. The number

of disks must meet the minimum requirement of the RAID policy. Click Next.

Select a migration triggering mode, for example, Periodic. Select the number of days

on which migration is implemented and the migration time range on each day. Click Next.

Confirm the information summary and click OK.

Step 3 Enable I/O monitoring and data migration.

Go to the Resource Allocation page, click SmartTier, select storage pool USR, and click

Properties.

In the dialog box that is displayed, click the Advanced tab. Enable I/O monitoring. Select

the number of days on which I/Os are monitored and the I/O monitoring time range. Click

OK.

The level of Data Migration Rate can be High, Medium, or Low. Select a level based

on your requirements.

Click Start to implement the periodic data migration.

----End

How to Use SmartTier



As the digitalization drive advances, data is increasingly important to the operating of an

enterprise, and users impose ever higher requirements on the data storage system. The

space utilization and storage efficiency fall short of the requirements. Customers usually

go to considerable expense to purchase a number of storage devices, especially when data

growth is hard to predict and mixed applications have diverse storage needs. With

SmartTier, customers no longer need to estimate the data growth speed. SmartTier greatly

reduces initial investments while retaining the same high performance. If the amount of

data keeps increasing and requires capacity expansion, SmartTier can be used to expand

storage capacity online without having any effect on services.

When an Oracle OLTP system is initially deployed, the data growth for the next few years

must be evaluated so that a targeted storage capacity plan can be made. If thick LUNs are

used for the system deployment, customers have to buy sufficient capacity that satisfies

storage needs for the next few years. However, if customers choose thin LUNs, the

capacity to be purchased only need to meet the needs for a short term (for example, one

year). The following figure explains why thin LUNs are a better choice than thick LUNs.

Suppose that the amount of data grows by 6 TB every year and the customer has to

consider the storage needs for the next five years. If thick LUNs are used, the customer has

to buy 30 TB storage capacity initially. If the customer chooses thin LUNs, then only 6 TB

storage capacity is required for the first year, and the customer expands the thin LUN

capacity by 6 TB every year. In so doing, thin LUNs help customers greatly reduce the

initial investment.

As an online transaction system, the Oracle OLTP system requires a high service

continuity. The storage capacity requires expansion along with the increase in the amount

of data. If thick LUNs are used, the online system expansion will compromise the service

performance, incurring unnecessary costs. If thin LUNs are used to expand the storage

capacity, the service will stay unaffected and properly functional during the online

Why SmartThin

Initial Investment

Reduction

Online Expansion

Without Service

Interruption



expansion.

Applying the SmartThin feature to the Oracle OLTP system, customers can use thin LUNs

to store the database data for lower initial investment and minimum performance impact.

The following figure shows the TPM values when thin LUNs and thick LUNs are

respectively used. As shown in the table, the TPM value drops by less than 8% when thin

LUNs are used.

The following figure shows the IOPS values when thick LUNs and thin LUNs are

respectively used. As shown in the table, the IOPS value drops by less than 4% when thin

LUNs are used.

Impact of

SmartThin on

Performance



The following figure shows the average response time when thick LUNs and thin LUNs

are respectively used. As shown in the table, the response time increases by less than

15% when thin LUNs are used.



To enable SmartThin in an Oracle database, users only need to create a thin LUN. The

following describes the procedure:

Step 1 Import the SmartThin license. For details, see the "How to Use SmartTier" section.

Step 2 Create a thin LUN.

On the ISM, go to the Resource Allocation page, click LUN, and click Create.

Enter a name for the LUN, set its capacity and quantity, select a storage pool that allocates

the capacity, and click Advanced.

Check the box before Enable, set an initial capacity (for example, 1 MB), and click OK.

Click OK.

----End

How to Use SmartThin



This chapter describes how to best plan and configure SmartTier and SmartThin for an

Oracle database to accelerate the performance and reduce investments.

An Oracle database contains the following types of files: control file, online log file, data

file, and archive log file.

In an OLTP database, the data file features random small I/Os, and the bottleneck usually

occurs in random read operations. Therefore, the data file needs performance acceleration

most. The online log file synchronizes I/O operations frequently. Transactions must wait

until the log is written before they are submitted. Therefore, the latency of log writing

affects the response to transactions. The online log file also needs performance

acceleration. The control file and archive log file are the final areas that require

performance acceleration.

The data file of an OLAP database features multi-path random read operations and

relatively big-sized data blocks. Query operations are often concurrently processed in

multiple streams, making I/O operations more random. In most cases, the performance

bottleneck of an OLAP database occurs in the sequential reading temporary tablespace and

user tablespace. Therefore, the user tablespace and temporary tablespace need

performance acceleration most in an OLAP database.

SmartTier allows users to set a monitoring time range for each storage pool based on

application characteristics.

The service monitoring time range specifies a period during which performance data is

collected for analysis by SmartTier. Defining a service monitoring period helps distinguish

between the active period and the idle period of a storage pool. SmartTier monitors the

application performance during the active period only. The service monitoring time range

applies to a specific storage pool. The default period is 24/7 hours. Users must customize

this period to specific days or hours. After the I/O monitoring function is enabled,

SmartTier analyzes the collected performance data once every hour during the service

monitoring period. The service monitoring time range is changeable after it is set.

The monitoring time range setting must be aligned with the characteristics of the target

service. If the service runs from 9:00 to 21:00, it is better to monitor the service during the

same period.

If applications stay inactive or I/Os are generated by unimportant tasks (for example,

background backup and data synchronization) during a specific period of time, this period

from must be deleted from the service monitoring period in case the I/Os affect the

identification of hotspot data and non-hotspot data.

SmartTier Best Practice

I/O Monitoring

Policy

Database

Acceleration Area

Selecttion

Best Practice



Data Migration Granularity

SmartTier identifies LUNs based on a data migration granularity. The activity levels of

data blocks are identified by the granularity, and then the data blocks are migrated to

appropriate storage tiers based on activity level.

SmartTier supports a number of granularities, including 256 KB, 512 KB, 1 MB, 2 MB, 4

MB, 8 MB, 32 MB, and 64 MB, where the default granularity is 4 MB.

Choosing a smaller granularity helps improve storage resource utilization and reduces the

possibilities of migrating cold data together with hot data. However, big granularities are

recommended for streaming media and video surveillance applications. Typical

granularities are 32 MB and 64 MB.

The recommended granularity in most cases is 4 MB.

Migration Triggering Mode

Migration triggering mode indicates how data is migrated. SmartTier provides manual and

periodic migration modes, both of which apply to one storage pool. The default mode is

manual. The manual mode allows data migration at any time, where the periodic mode

permits data migration during a specific preset period. The migration triggering mode can

be changed after it is set.

Manual data migration may be required when a storage pool needs to be reconfigured (for

example, after a new tier is added), new LUN attributes need to take effect immediately, or

constantly changing hotspot data is distributed to all tiers. After a manual migration task is

initiated, SmartTier analyzes the current performance data and determines the destination

tiers of data based on the analysis.

If periodic migration is chosen, a migration schedule needs to be set first. The migration

schedule specifies when and where SmartTier migrates data.

Periodic migration is recommended, and data migration during idle service hours, for

example from 1:00 to 5:00, is preferable.

Migration Policy

A migration policy specifies the data migration direction and applies to one LUN in a

storage pool. Four migration modes are optional: automatic migration, migration to a

higher performance tier, migration to a lower performance tier, and no migration.

Automatic migration

This policy is the default policy set upon the creation of a LUN. SmartTier migrates

data blocks in a storage pool based on the ranking order of activity level. Automatic

migration is implemented based on the analysis data in I/O monitoring. Therefore, for

automatic migration to take effect, users must enable I/O monitoring and set a service

monitoring period.

Data Migration

Policy



Highest available

If a LUN has high requirements on response, IOPS, and bandwidth, the highest

available policy is appropriate for the LUN. This policy preferentially promotes data

blocks to the high performance tier and the performance tier. If the capacity of data

blocks to be migrated exceeds that of the high performance tier (or performance tier),

only data blocks with a higher activity level are promoted to the high performance

tier (or the performance tier).

Lowest available

The lowest available policy is recommended for the applications (such as file sharing)

insensitive to performance. This policy preferentially demotes data blocks to the

performance tier and the capacity tier, regardless of the activity levels of these data

blocks.

No migration

This policy does not migrate data blocks among storage tiers. Data can be migrated

only when the data relocation policy is changed.

It is recommended that the data migration policy be set and dynamically adjusted

based on the performance requirements of LUNs.

Migration Rate Selection

A data migration rate specifies how fast data is migrated among storage tiers and applies to

all storage pools. SmartTier supports high, medium, and low data migration levels, where

the default migration level is low. The following lists the migration speeds of the three

levels.

High: 300 GB/h to 350 GB/h

Medium: 60 GB/h to 72 GB/h

Low: 30 GB/h to 36 GB/h

As periodic migration has been selected to migrate data during off-peak hours, a medium

or high migration rate is recommended if users want to compete the hotspot data migration

in a short period of time. However, if the effect on performance is a more important

consideration during the migration, a low migration rate is preferable.

An initial capacity allocation policy specifies where new data is written into a LUN. The

policy applies to one LUN in a storage pool, and optional policies include automatic

migration, allocate from the high performance tier first, allocate from the performance tier

first, allocate from the capacity tier first. The default policy is automatic migration.

It enables the storage system to distribute new data to the capacity tier, the performance

tier, and then the high performance tier in sequence.

Initial Capacity

Allocation Policy



Allocate from the high performance tier first, allocate from the performance tier first, or

allocate from the capacity tier first enables the storage system to distribute new data to the

high performance tier (performance tier or capacity tier) first. It migrates data to other

performance tiers only when the capacity of the high performance tier (performance tier or

capacity tier) is insufficient.

When data is initially written, SmartTier has no collected performance statistics about the

data. Without performance analysis, SmartTier cannot identify the activity level of the

data. Therefore, the automatic allocation policy is recommended. Note that the initial

capacity allocation policy cannot be changed once it is set.

Huawei provides disks of five rates: 7200 rpm disk, 10k rpm disk, 15k rpm disk, eMCL

SSD, and SLC SSD.

The following table lists the performance of these disks.

Disk Type OLTP Latency OLTP IOPS OLAP Throughput

7200 rpm disk 20 ms 30 20 MB/s

10k rpm disk 15 ms 150 30 MB/s

15k rpm disk 10 ms 200 50 MB/s

eMLC SSD 2 ms 5000 80 MB/s

SLC SSD 1 ms 8000 100 MB/s

SmartTier classifies disks into three types by performance and use them to compose the

high performance tier, performance tier, and capacity tier. After these tiers are established,

SmartTier identifies hotspot data by monitoring I/Os and migrates the hotspot data to the

high performance or performance tier. Cold data is migrated to the performance tier or

capacity tier. In so doing, SmartTier accelerates application performance.

An Oracle OLTP database usually contains a great deal of hotspot data. It is recommended

that at least two storage tiers, namely the high performance tier and performance tier, be

established for the database. With automatic identification and migration of hotspot data,

SmartTier notably improves the database performance.

Archive logs are typically demanding in capacity not performance. Therefore, it is better to

store archive logs in the capacity tier, saving storage resource investments.

The control file and online logs of an Oracle database consume a predictable small amount

of storage space. As archive logs mainly involve write operations, NL SAS disks are

usually used to storage the logs, and thin LUNs are not recommended.

The data file of a database stores user data, and it is hard to estimate how much space the

data file requires. Allocating an inappropriate amount of storage space will result in

resource waste or cause online expansion to affect services. Therefore, it is recommended

that the data file be stored in a thin LUN, improving storage space utilization and ensuring

SmartThin Best Practice

Thin LUN Area

Selection

Disk Selection

for Storage

Tiers



flexible scalability.

When using SmartThin to create a thin LUN, users must set an initial allocation capacity.

The capacity must suit the characteristics of the application. Because SmartThin employs

copy-on-write (CoW) technology, allocating an initial capacity simplifies space

allocation when operations are performance, reducing the latency of write operations.

It is recommended that the initial allocation capacity satisfy estimated the storage demand

of the corresponding application. If the storage demand is unpredictable, users can set any

value. The minimum initial allocation capacity is 1 MB.

Initial Capacity

Allocated



Copyright © Huawei Technologies Co., Ltd. 2013. All rights reserved.

No part of this document may be reproduced or transmitted in any form or by any means without prior written

consent of Huawei Technologies Co., Ltd.

Trademarks and Permissions

and other Huawei trademarks are trademarks of Huawei Technologies Co., Ltd.

All other trademarks and trade names mentioned in this document are the property of their respective holders.

Notice

The purchased products, services and features are stipulated by the contract made between Huawei and the

customer. All or part of the products, services and features described in this document may not be within the

purchase scope or the usage scope. Unless otherwise specified in the contract, all statements, information,

and recommendations in this document are provided "AS IS" without warranties, guarantees or

representations of any kind, either express or implied.

The information in this document is subject to change without notice. Every effort has been made in the

preparation of this document to ensure accuracy of the contents, but all statements, information, and

recommendations in this document do not constitute a warranty of any kind, express or implied.

Huawei Technologies Co., Ltd.

Address: Huawei Industrial Base

Bantian, Longgang

Shenzhen 518129

People's Republic of China

Website: http://www.huawei.com

Email: [email protected]

http://www.huawei.com/

mailto:[email protected]

Best Practices of SmartTier and SmartThin in Oracle 11g Database Environments … · 2014-11-28 ·...

Documents

Transcript of Best Practices of SmartTier and SmartThin in Oracle 11g Database Environments … · 2014-11-28 ·...