Dell PowerVault 110T LTO3 ユーザーガイド

www.dell.com | support.dell.com

Dell™ PowerVault™ Systems

Performance Considerations

for Tape Drives and Libraries

____________________

Information in this document is subject to change without notice.

Reproduction in any manner whatsoever without the written permission of Dell Inc. is strictly forbidden.

Trademarks used in this text: Dell, the DELL logo, and PowerVault are trademarks of Dell Inc.; EMC and PowerPath are registered trademarks

of EMC Corporation.

Other trademarks and trade names may be used in this document to refer to either the entities claiming the marks and names or their products.

Dell Inc. disclaims any proprietary interest in trademarks and trade names other than its own.

June 2005

Contents 3

Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

General Host Backup Considerations

. . . . . . . . . . . . . . . . . . . . . . . 5

Tape Drive and Data Considerations

. . . . . . . . . . . . . . . . . . . . . 5

Hard Drive and RAID Array Configuration

. . . . . . . . . . . . . . . . . . 6

General Performance Considerations When Using Multiple Drives

in Tape Libraries

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

SCSI Configurations

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

SAN Configurations

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4 Contents

Performance Considerations for Tape Drives and Libraries 5

Introduction

W

ith recent improvements in tape drive transfer rates,

many host-side factors, such as RAID

(Redundant Array of Inexpensive [or Independent] Disks) configuration and hard-drive

specifications, must be considered when determining whether the host server and tape drive can

process data at the same rate. General configurations and attributes that may limit throughput

from the host server to the tape drive are discussed in "General Host Backup Considerations."

As multiple drives are placed into tape libraries, greater host bandwidths are needed to keep pace

with the potential throughput of multiple tape drives. Potential fibre limitations for multidrive

units, as well as recommended cabling configurations, are discussed in "General Performance

Considerations When Using Multiple Drives in Tape Libraries."

General Host Backup Considerations

The considerations in this section apply to both SCSI and storage area network (SAN) tape backup

configurations.

Tape Drive and Data Considerations

The following issues should be considered when evaluating performance:

Overhead from SCSI commands.

Command overhead on the SCSI bus restrict

all

SCSI devices

in achieving theoretical maximum transfer speeds. Tape backup software does not account for this

overhead; instead, the software only measures the rate at which data is written to the tape. For

example, the drive may be processing 80 MB/sec of data, but only writing 77 MB/sec of data.

The latter rate is what the backup software will report.

Tape block sizes.

64 Kb block sizes are optimal for most tape drives. However, some backup

applications allow the user to change block size, even though a larger size will not enhance

performance. Using block sizes less than 64 Kb can actually hinder performance. See your backup

software User's Guide for information on adjusting the block size of your tape device.

Backup software buffer size.

For optimal backup performance, backup software buffers should be

as large as possible. Some applications allow users to change the buffer size, which can help

maintain a steady stream of data to and from the drive and significantly increase transfer rates,

especially of small files. The larger the buffer, the more data it can hold and the less time the disk

spends seeking the data; however, this can affect memory and CPU performance. See your tape

backup application User’s Guide for specific details.

Drivers and firmware.

Always ensure that the SCSI or fibre controller and tape drive have the latest

drivers and firmware installed. Visit

support.dell.com

to download the latest drivers and firmware

for your Dell PowerVault tape product.

6 Performance Considerations for Tape Drives and Libraries

www.dell.com | support.dell.com

Attach tape drives and hard drives on separate controllers (internal or external) on separate host

bus adapters (HBAs).

This depends somewhat on the performance capabilities of your controller,

but best practice is to keep the tape drive HBA separate from the hard drive HBA to ensure

maximum throughput. Most onboard dual-mode SCSI/RAID controllers share one processor,

which must share bandwidth between the RAID array and tape drive. Thus, one controller is

handling reads and writes between the hard disks and the tape drive, as well as calculating and

writing any necessary parity information to the hard drives. See "Hardware RAID Configuration

Considerations" for specific information on RAID arrays and parity bytes.

Dirty drive heads or old media.

A dirty tape drive head or old media can cause high error rates and

a corresponding reduction in read/write speeds. Each time a drive attempts to rewrite or reread a

block on a tape, performance is degraded. Once a certain threshold of read/write errors is reached,

the drive will usually request cleaning. It is important to clean the drive heads on regular intervals,

or when requested.

The chance of encountering a bad block increases as media ages or is excessively used. The typical

lifecycle of a piece of LTO media is approximately 75 full tape read/writes.

Speed matching.

Newer LTO drives will match the speed of the data being provided to the drive,

down to approximately one-half of the maximum uncompressed data transfer speed. If data is

provided to the drive at less than the lower speed matching limit, the drive must stop, wait for the

buffer to fill, rewind, and then attempt to write the buffer (this is known as "back hitching").

For example, the Dell PowerVault 110T (LTO2 and LTO3) tape drive matches speed down to

30 MB/sec while writing to LTO-3 media. If the host server can only provide data at 20 MB/sec, the

drive will "back hitch" while waiting for its buffers to fill. In this situation, the effective throughput

will be something less than 20 MB/sec (probably closer to 15 MB/sec).

Confirming Performance of Your Tape Drive

Certain tape drive manufacturers have a performance diagnostic mode built into the drive that can

be used to confirm throughput. The PowerVault 110T LTO-2 and LTO-3 (firmware 53

XX

or later)

offer a diagnostic mode "F," which performs a quick read/write performance test on the drive and

media. If the performance rate is not within 6 percent of the maximum specified drive speed, the

test fails with an error message. No error message is displayed if the test passes. Consult your tape

drive User’s Manual for specific details on diagnostic mode "F."

NOTICE: Diagnostic mode "F" requires media that can be safely overwritten as part of the diagnostic test.

Do not use media containing critical data. Any data residing on the media used in the diagnostic test will

be lost.

Hard Drive and RAID Array Configuration

Several hard drive and disk array (both internal and external) attributes can affect backup or restore

performance. These attributes, as well as recommended configurations that help achieve

maximum backup and restore speeds, are discussed in the following subsections. If the tape drive’s

sustainable throughput exceeds that of the disk array, then the tape drive’s peak performance will

not be realized.

Performance Considerations for Tape Drives and Libraries 7

General Hard Drive Configuration Considerations

Data/operating system (OS) on different LUNs.

Backing up data on a logical unit number (LUN)

separate from the OS LUN ensures that the hard drive is not splitting access and overhead between

OS operations and backup operations. This can be accomplished by having one hard drive or disk

array contain the OS and a physically separate hard drive or disk array contain the data

to be backed up.

Figure 1-1. Single-Channel vs. Two-Channel Bandwidth

Hard Drive Performance

By design, tape drives write data sequentially and require a constant data feed to keep the drive

operating sequentially (avoiding back hitching). Conversely, hard drives are random access devices.

Therefore, hard drives can sometimes struggle to provide sequential data to tape drives if that data

is spread out over the drive platter. This forces the drive to continuously seek small blocks of data.

Additionally, other hard drive attributes can further affect the throughput of data to the tape drive.

Spindle speed.

Typically measured in RPMs (revolutions per minute), the hard drive's spindle

speed determines how many times per minute the drive platter assembly can perform a full

revolution. This has a direct effect on both random access times and sequential transfer rates.

The higher the spindle speed, the faster the drive can access data.

Single Shared LUN Separate LUNs

Single LUN with

Backup data

and OS

OS LUN

SCSI or RAID

Controller

Tape Drive

Backup Data

LUN

SCSI or RAID

Controller

Tape Drive

8 Performance Considerations for Tape Drives and Libraries

www.dell.com | support.dell.com

Random access time or seek time.

Usually measured in milliseconds, seek time is the length of

time a drive's heads take to find a piece of data on the disk. The seek time of a hard disk measures

the amount of time required for the read/write heads to move between tracks on the surface of the

platters. Because hard disks are random access devices, data can be stored on virtually any sector of

the disk. The longer it takes to access that data, the slower the overall throughput of the drive. This

attribute is very significant when a hard drive contains many small files. The smaller the files, the

more "seeks" the drive must make to read or write the file to disk; therefore, disks tend to read or

write very slowly when many small files are being transferred.

Sequential/sustained transfer rates (STR).

STR measures how fast a drive actually reads data from

and writes data to its platters. If the data being backed up is one large contiguous file, the sustained

throughput will be close to the drive's maximum STR. However, in real-world applications, data

becomes more scattered about the platter as data is deleted and written. Defragmenting a hard

drive can help the drive reach its maximum STR.

Buffer (cache).

The buffer is the amount of memory on the drive that holds the most recently

written or stored data. The bigger the buffer, the more data it can hold, resulting in less time

seeking data on the disk.

Hardware RAID Configuration Considerations

General overview of RAID

This section presents an overview of typical RAID configurations and how they affect backup and

restore rates. A RAID array is a set of hard disks that act as a single storage system or LUN. Data can

be potentially transferred through the channel of each hard drive at once, allowing for total

throughput to be a multiple of the total number of drives in the array, minus overhead and any

redundancy as described in the following sections.

In the case of a RAID configuration, the speed of the interface becomes important because the

drives share the bandwidth of the interface. For example, a single Ultra160 drive may only sustain

40 MB/sec. Thus, a five-disk RAID 0 array consisting of the same drive type should be able to

read/write at 200 MB/sec. However, the Ultra160 interface will limit the array to a maximum of

160 MB/sec.

External disk arrays, particularly in SANs, may offer significant levels of cache memory to improve

I/O performance. This cache will greatly improve performance when writing to the array and may

store frequently accessed data to improve read performance. With respect to its impact on tape

performance, the cache will mask most RAID limitations when restoring data to the array or

backing up data from the array. However, backup operations from external arrays with cache may

still feel the impact of RAID configuration limitations because the data still needs to be read from

the disks.

Performance Considerations for Tape Drives and Libraries 9

RAID 0

Commonly known as striping, RAID 0 allows two or more disks to be joined to create one virtual

drive in the fashion of a single LUN. It is referred to as striping because data is written across all of

the disks in the array, not just to one disk at a time. Thus, the throughput is spread across

n

channels (

n

being the number of hard drives in the array) instead of a single channel for a single

hard disk. This results in excellent read/write performance, but no fault tolerance.

Figure 1-2 shows four hard drives in a RAID 0 configuration. Data is striped across all four hard

drives, resulting in four channels for reading and writing to the array.

Figure 1-2. Example RAID 0 Configuration

SCSI or RAID

Controller

Tape Drive

D1

D5

D9

D13

D17

D2

D6

D10

D14

D18

D3

D7

D11

D15

D19

D4

D8

D12

D16

D20

Hard Drive 1 Hard Drive 2 Hard Drive 3 Hard Drive 4

D = Data Byte

10 Performance Considerations for Tape Drives and Libraries

www.dell.com | support.dell.com

RAID 1

Also known as mirroring, RAID 1 means data is written twice across two disks simultaneously. If

one drive fails, the system switches to the other drive without losing data. During tape drive

backups, the read rate from the RAID 1 array is approximately the same as a single drive because it

is reading from the primary drive. However, restore performance from the tape drive writing to the

RAID 1 array can be slower due to error checking/correction (ECC) included in writing to the

primary and mirrored disks. Much of this inefficiency is due to the fact that the mirroring is often

performed on the CPU or RAID controller. Thus, newer RAID controllers tend to be faster due to

newer and more capable processors.

Figure 1-3. Example RAID 1 Configuration

D1

D2

D3

D4

D5

M1

M2

M3

M4

M5

Tape Drive

SCSI or RAID

Controller

Write OnlyRead/Write

Read/Write

Hard Drive 1 Hard Drive 2

D = Data Byte

M = Mirrored Byte

Performance Considerations for Tape Drives and Libraries 11

RAID 5

With a RAID 5 array, data is striped across the disk array at the byte level and error correction data,

or parity data, is also striped across the disk array. RAID 5 arrays tend to have very good random

read performance; this read performance generally improves as the number of disks in the array

increases. With the larger disk arrays, read performance can actually outperform RAID 0 arrays

because the data is distributed over an additional drive. In additional, parity information is not

required during normal reads.

Restores from tape to a RAID 5 array tend to be nominal because it involves additional overhead

for calculating and writing the parity information.

Figure 1-4. Example RAID 5 Configuration

P1

D4

D7

D10

D13

D1

P2

D8

D11

D14

D2

D5

P3

D12

D15

D3

D6

D9

P4

D16

SCSI or RAID

Controller

Tape Drive

Read/Write

D = Data Byte

Hard Drive 1 Hard Drive 2 Hard Drive 3 Hard Drive 4

P = Parity Byte

12 Performance Considerations for Tape Drives and Libraries

www.dell.com | support.dell.com

In conclusion, RAID 0 tends to be the best overall configuration for read and write performance,

but does not allow for redundancy. RAID 1 is the worst performer overall, as all data written to

the array is mirrored and reads come from a single disk. RAID 5 tends to be a good read performer

but average write performer; however, RAID 5 improves if more disks are added to the array. If

the RAID is within an enclosure that offers significant levels of cache memory, then performance

limitations during restore operations may be abated. Backup operations will still be subject to

limitations of the RAID configuration.

In addition, the characteristics of the array still depend

heavily on the specific hard drive characteristics listed in "Hard Drive Performance."

General Performance Considerations When Using Multiple

Drives in Tape Libraries

When multiple tape drives are utilized simultaneously to perform data backups (such as in tape

libraries), additional aspects of the hardware configuration must be considered. By employing

simple performance-minded methods in setting up hardware and cabling configurations, additional

throughput bottlenecks can be limited.

SCSI Configurations

The latest high-performance tape drives offered in tape libraries support the Ultra160 specification

of the SCSI interface standard. Therefore, to achieve maximum performance, backup servers

utilizing SCSI must have an HBA installed that supports data speeds of Ultra160 or higher. A SCSI

HBA that meets this requirement will allow each tape drive to communicate with the host at a rate

of 160 MB/sec on the SCSI bus. The higher data rate of the SCSI bus compared to tape drive

speeds allows multiple devices to be connected to the same bus without sacrificing device

performance. But only to a point.

The 160 MB/sec data rate of an Ultra160 bus is the maximum possible data throughput rate to all

devices connected to the bus. Therefore, a single tape drive will not consume the full bandwidth of

the bus because it can read or write data to tape at up to 80 MB/sec (native). Multiple tape drives,

however, can combine to consume the full 160 MB/sec offered by the bus if each is operating at its

maximum native performance. Each additional drive connected to the same bus after this point

will reduce the average performance of each drive.

Therefore, to achieve maximum performance from a tape library, it is recommended to connect

no more than two tape drives to each SCSI bus. See "Recommended Cabling Configurations" for

specific details and illustrations. A SCSI HBA supporting at least Ultra160 should be used, but

upgrading to an Ultra320 HBA will not lead to an additional improvement in performance if the

tape drive's specification is Ultra160.

Performance Considerations for Tape Drives and Libraries 13

SAN Configurations

Fibre Channel (FC) offers many advantages over SCSI. First, it overcomes the distance limitations

of SCSI (12 meters for LVD SCSI versus 300 meters for a short-wave 2-Gb FC link)

and allows for

the transmission of data at higher speeds. As a serial network protocol rather than a bus-based

architecture like SCSI, FC has also become the protocol of choice for the implementation of SANs,

allowing for the consolidation of data storage resources. In addition, each FC connection is made

up of a transmit link and a receive link, allowing for full-duplex operation. This means that data can

be transmitted in two directions simultaneously. Therefore, during a backup operation across a

single FC connection, data can be read from a source and written to tape without taking turns in

communication, effectively doubling the bandwidth of a connection. See Figure 1-5.

Figure 1-5. Fibre Channel Link Diagram

When setting up tape libraries in a SAN, performance can still be affected by various factors. These

factors include FC link speeds, data flow between the source and tape library, and performance

limitations of external storage arrays. With an understanding of the overall setup and management

of the solution, many of these factors can be avoided.

Even with the high data bandwidth offered by the FC protocol in SANs, proper considerations

must be made for tape drives in order to avoid a situation in which the FC link may limit

performance. The data rate of a 2-gigabit (Gb) FC link is 200 MB/sec (that is, 200 MB/sec on the

transmit link and 200 MB/sec on the receive link). Therefore, attempting to operate multiple tape

drives across the same link can potentially exceed the full bandwidth of a link. If the host is

operating with a legacy 1-Gb adapter, backing up data to two drives may be sufficient to reveal

significant performance limitations.

Therefore, when using three or more tape drives simultaneously on a 2-Gb link, you may need to

distribute the backups across a number of connections, rather than relying on a single link. This is

where understanding the SAN solution's topology is beneficial. Following the data path during a

backup operation as it is read from the source and then written out to tape will help administrators

recognize any potential bottlenecks. If any bottlenecks are identified, measures may be taken

depending on the configuration. For example, if the backup solution requires multiple drives to be

in operation at once, splitting the tape hardware across separate fabrics may improve performance

by splitting the connections. See Figure 1-6.

HOST

Fibre

Channel

Device

Transmit

Receive

Transmit

2 Gb = 200 MB/sec

14 Performance Considerations for Tape Drives and Libraries

www.dell.com | support.dell.com

Figure 1-6. Single vs. Split Data Flow to Tape Library

At the same time, if a bottleneck exists in the data being read from an external FC disk array, the

use of I/O management software such as EMC

®

PowerPath

®

with an additional fabric connection

will automatically load-balance the data across multiple paths and increase availability through

path failover. See Figure 1-7. The left side of the figure represents a SAN disk array in which all of

the data is forced through a single link, creating a bottleneck that slows data flow to the tape

hardware. The right side shows how load balancing doubles the I/O bandwidth coming out of the

array by allowing the data to transmit across two links.

HOST

Fibre Channel

Switch

Fibre Channel

Disk Array

Fibre Channel

Switch

Fibre Channel

Switch

HOST

Fibre Channel

Disk Array

Tape Library Tape Library

Performance Considerations for Tape Drives and Libraries 15

Figure 1-7. Bottlenecked Data Flow vs. Load-Balanced Disk Array

Finally, FC disk arrays on the SAN can also experience the same performance limiters described in

"Hard Drive and RAID Array Configuration." Therefore, improving the performance characteristics

of the disk arrays will also have a direct effect on backup speed across the SAN.

SAN Configurations Utilizing the Library Fibre Channel Bridge

Certain tape libraries may be connected to a SAN by way of a Fibre Channel bridge module. The

module acts as a bridge between the SCSI and FC protocols and provides additional management,

security, and operational features unavailable in most native FC libraries. For details on these

features, see the Fibre Channel bridge User's Guide for your tape library.

In some tape library configurations, the Fibre Channel bridge module may act as a bottleneck and

decrease performance of tape drives. This is because the processing capability in the Fibre Channel

bridge module required to bridge the SCSI and FC communication cannot meet the aggregate

data throughput offered by certain multidrive configurations. Despite this, most data backup

solutions will not experience the Fibre Channel bridge module as the primary limiting factor in

tape performance. Dedicated backup servers will frequently encounter a situation where the

limitations at the host will be compounded by the exertion of feeding data to multiple tape drives.

This results in average drive performance below the level where the Fibre Channel bridge module

becomes a factor.

Fibre Channel

Switch

Fibre Channel

Switch

HOST

Fibre Channel

Switch

Fibre Channel

Switch

Fibre Channel

Disk Array

Fibre Channel

Disk Array

Tape Library Tape Library

16 Performance Considerations for Tape Drives and Libraries

www.dell.com | support.dell.com

Recommended Cabling Configurations

SCSI Cabling to the Host

Tape Library With up to Six Tape Drives

When the tape library is SCSI-attached to a host, ensure that no more than two drives are on a

single bus. Additional SCSI controllers are required for libraries with five or six tape drives to

ensure that no SCSI bus becomes a barrier to maximizing throughput.

Figure 1-8. SCSI Cabling for Library With up to Six Tape Drives

Tape Library With up to Two Tape Drives

The drives in a fully configured two-drive tape library can be cabled to a host on the same SCSI bus

without encountering significant limitations to backup performance. The backup rates for two

drives on a single SCSI bus will match the backup rates for two drives on separate buses. However,

customers who enable the verify feature in backup applications may wish to improve the verify

performance by splitting two drives onto two SCSI buses. By doing so, verify performance may

improve by up to 25 percent.

library-to-host

SCSI cable

library-to-host

SCSI cable

drive-to-library

controller SCSI

cable

terminator

Performance Considerations for Tape Drives and Libraries 17

Figure 1-9. SCSI Cabling for Library With up to Two Tape Drives

Cabling Drives to the Fibre Channel Bridge

Figures 1-10 through 1-17 illustrate how a tape library with up to six drives should be configured

with a Fibre Channel bridge module in order to optimize tape performance over FC.

terminator

18 Performance Considerations for Tape Drives and Libraries

www.dell.com | support.dell.com

Figure 1-10. Fibre Channel Bridge Cabling With One Tape Drive

library SCSI

interface

SCSI 1

terminator

drive 1

Performance Considerations for Tape Drives and Libraries 19

Figure 1-11. Fibre Channel Bridge Cabling With Two Tape Drives

library SCSI

interface

SCSI 1

SCSI 2

terminator

drive 1

drive 2

20 Performance Considerations for Tape Drives and Libraries

www.dell.com | support.dell.com

Figure 1-12. Fibre Channel Bridge Cabling With Three Tape Drives

library SCSI

interface

SCSI 1

SCSI 2

terminator

drive 1

drive 2

drive 3

ページを読み込んでいます...