The Impact Of Disk Fragmentation On Servers

The Impact of
Disk Fragmentation
on Servers
By David Chernicoff
Published: May 2009

The Impact of
Disk Fragmentation
on Servers
Testing Server Disk Defragmentation
IT professionals responsible for server hardware well
understand the value that professional grade disk
defragmentation software brings to their servers.
Storage servers can experience high levels of disk thrashing
(the constant writing and rewriting of small amounts of data)
caused from excessive ﬁ le fragmentation.
Problems in delivering services to users however are difﬁ cult
to directly trace to server fragmentation issues. Network and
application issues have a much more visible impact on the
performance of network-based services, especially when
problems with those functions are encountered. But with the
negative impact on ROI that network performance problems
cause, IT pros would be ill-advised to overlook the advantages
➔ Contents
that assuring the optimization of the underlying hardware
infrastructure can bring. Optimal disk performance translates
Testing Server Disk Defragmentation ..... 2
into better ROI. Testing will bear this out.
The Testing Environment ........................ 3
The Tests ................................................ 4
File
Copy
........................................ 4
We tested the impact of server disk defragmentation by look-
Document
Open
............................ 4
ing at common tasks that network servers, both physical and
Backup
........................................... 5
virtual, encounter, ranging from maintenance tasks such as
Anti-Virus
Scan
.............................. 5
server backup and anti-virus scans, to basic knowledge worker
VHD
Start
...................................... 5
tasks involving opening ﬁ les stored on the host server and
VHD
Save
...................................... 6
virtual machines, and manipulating email. We also looked
Server Application Tests ......................... 6
Exchange
Test
One
......................... 6
at tasks that are more taxing on the server, such as database

Exchange Test Two .......................... 7
queries, index creation, and bulk updates. Each test was per-

SQL Server Bulk Insert.................... 7
formed as the sole task on the server.

Table Key Creation ......................... 7

SQL Query 1 .................................. 7
When considering the results of our testing keep in mind

SQL Query 2 .................................. 8
that a production environment will see signiﬁ cantly heavier
Conclusion ............................................ 8
server use, which results in much greater potential for ongo-
ing disk fragmentation. In your production environment with
dozens, if not hundreds, of users touching your server storage
simultaneously, your disk fragmentation can become severe
in a very short time. Preventing this fragmentation from af-
fecting server performance is an ongoing process.
2
THE IMP

A
THE IMP CT OF DISK FRA
A
GMENT
CT OF DISK FRA
A
GMENT TION ON SER
A
VERS
TION ON SER
sponsored by
sponsored b Diskeeper

The Testing Environment
We tested three levels of fragmentation, described
For our benchmark tests we used an HP ProLiant
herein as low, medium, and high. We used the
DL380 G5 equipped with dual quad-core 2.83 GHz
Diskeeper Diskcrusher fragmentation utility to create
Xeon processors, each with a 2x6MB L2 cache, 16
fragmented ﬁ les and directories. We ran all tests a
GB of RAM and seven 72 GB 10,000 RPM SCSI
minimum of three times with the results reported here
drives attached to an HP Smart Array P400 control-
being the average of all test runs.
ler that has a 256 MB cache
and that supports both serial-
Low
Medium

High
attached SCSI and SATA drives.
Number of files
101,652
1,220,660
2,087,158
The volumes we tested against
were 30 GB, 80 GB, and 175
Avg. Number of Fragments
3.21
1.69
2.30
GB. We used a 500 GB 7200
per File
RPM locally attached SATA
Number of Fragmented Files 99,074
613,221
1,994,117
drive for backup only. The
Number of Excess Fragments
225,216 840,076 3,005,400
server operating system was
Percent Fragmented –
40%
50%
84%
Microsoft Windows Server
Volume
2008 Enterprise; the applica-
tion server software installed
Percent Fragmented – Data
51%
58%
91%
in VHDs was Microsoft SQL
Free Space
22%
15%
15%
Server 2008 and Microsoft Ex-
Table 1: Fragmented disk test conﬁ gurations
change Server 2007. All server
software was updated with service packs, patches and As shown in Table 1 the level of fragmentation and
hotﬁ xes current as of February 2009. The disk defrag-
the number of affected ﬁ les increases with each test-
mentation software used was Diskeeper Server.
ing tier. The level of fragmentation you’ll encounter in
production environments is dependent upon the level
The seven SCSI drives attached to the array controller were of use and types of applications the server deals with.
conﬁ gured as two physical drives. We used the ﬁ rst physi-
In all likelihood, if your server storage levels are con-
cal drive, comprised of two drives conﬁ gured as a RAID 0
sistently exceeding 75 percent or so, you’ve begun
stripe set for maximum performance, for the installation of
aging data off of the servers or you’re planning to add
the operating system and all related ﬁ les. We conﬁ gured
additional storage. While fragmentation isn’t a direct
the remaining ﬁ ve drives as a RAID 5 stripe set to be rep-
result of reduced capacity, the chances for fragmenta-
resentative of the type of hardware storage conﬁ guration
tion increase as free storage space decreases and the
found in most business environments. We performed all
operating system is forced to write data into an ever-
applications, VHDs, and tests on the RAID 5 stripe set. The
increasing number of non-contiguous spaces.
volume size was dependent upon the test level.
By using an automated defragmentation process, the
As an example of the effect fragmentation can have, the
same disk volume sees absolutely minimal fragmen-
screen capture in Figure 1 shows the Diskeeper fragmen-
tation even though it is in continual use by applica-
tation analysis of a severely fragmented disk. The severe
tions and users (Figure 2).
fragmentation documented here will have a negative
impact on storage performance.
Figure 1: Fragmentation map of a heavily fragmented disk
Figure 2: Fragmentation map after automated defragmentation by Diskeeper.
sponsored by Diskeeper
THE IMPACT OF DISK FRAGMENTATION ON SERVERS
3

We ran each set of tests for
three iterations, and then de-
Low
Medium

High
fragmented the storage using
Diskeeper to reduce or elimi-
Number of files
101,652
1,220,660
2,087,158
nate the disk fragmentation.
Percent Fragmented –
0
0
0
We repeated each test (also for
Volume
three iterations) and averaged
Percent Fragmented – Data
0
0
0
the results. In the following
test descriptions and analysis,
Avg. Number of Fragments
0
0
0
the comparisons are all before
per File
and after defragmentation at
Number of Fragmented Files 0
1
1
each speciﬁ c fragmentation
Number of Excess Fragments
0
2
4
level tier. We did not do cross-
tier comparisons. All test times
Free Space
22%
15%
15%
are reported in seconds.
Table 2: State of fragmentation after Diskeeper has been run
The Tests
While the limiting factor in doing a ﬁ le copy from the
In our ﬁ rst set of tests we look at common server
server to the client might be the available network
tasks that are likely to be affected by disk fragmenta-
bandwidth, as technologies such as Gigabit Ethernet
tion. These tasks are all primarily storage related; that
become more common, the base limiting factor will
is, the performance of the storage media will have a
be how fast the operating system can feed data to
primary impact on the performance of these tasks.
the network request, which is directly impacted by
fragmentation of the data on the local drive.
File Copy
Document Open
In the ﬁ le copy test, a folder containing 5 GB worth
In this test, a 100-page Microsoft Word document
of ﬁ les and sub-directories was copied from the test
was opened from the server to a Windows XP client
volume to the boot volume of the server. To minimize
running Microsoft Ofﬁ ce 2007. The size of the docu-
variables, the copy was done locally, not across the
ment was 3.3 MB.
network. We timed the test using a stopwatch. This is
one of the most basic tasks done with server data and,
Document Open Tests (measured in seconds)
in a severely fragmented environment, showed some
of the most signiﬁ cant performance improvements.
Low – Fragmented
11.7
Low – Defragmented
10
File Copy Tests (measured in seconds)
Medium – Fragmented
12.7
Low – Fragmented
44
Medium – Defragmented
10.7
Low – Defragmented
39
High – Fragmented
14.7
Medium – Fragmented
72
High – Defragmented
10.3
Medium – Defragmented
60
Our test results showed performance improvements
High – Fragmented
97
of upwards of 30 percent. In the case of any ﬁ le load
from server to client the performance improvement
High – Defragmented
54
will be determined by just how badly fragmented
The basic task of moving data from one location to
is the ﬁ le located on the server. In our tests, the ﬁ le
another on the server shows that a fragmented disk
was clearly badly fragmented, signiﬁ cantly so at the
has a major negative impact on the ﬁ le copy. Even
highest level of fragmentation testing. To prevent this
the lightly fragmented low-level test showed an im-
type of ﬁ le fragmentation, the best methodology is
provement in copy time of over 11 percent, while the
an ongoing background ﬁ le defragmentation pro-
copy that was done from the very highly fragmented
cess, the beneﬁ ts of which are clearly demonstrated
drive improved in time by almost 45 percent. Given
by this test. And given how often this type of task is
how common the ﬁ le copying task is the beneﬁ t is
performed in most business environments, the value
clear. Defragmented disks are a signiﬁ cant time saver
of the defragmentation cannot be understated. As
for common user tasks.
shown in this and the File Copy test, basic data ma-
nipulation is much faster on defragmented storage.
4
THE IMPACT OF DISK FRAGMENTATION ON SERVERS
sponsored by Diskeeper

Backup
Anti-Virus Scan Tests (measured in seconds)
In the ﬁ rst test, we backed up the test volume using
disk-to-disk backup as supported by Windows Server
Low – Fragmented
256
Backup, which is a component of Windows Server
Low – Defragmented
238
2008. Backup was done using the VSS copy method,
Medium – Fragmented
1485
which is designed to work with other backup tools
Medium – Defragmented
1359
that would require that the archive and backup in-
formation in the ﬁ les remain unmodiﬁ ed. We backed
High – Fragmented
4428
up to a SATA-attached dedicated hard drive that was
High – Defragmented
4004
reformatted between tests. Timing was done using the
backup application.
Many factors will have an impact on the speed of a
complete anti-virus scan of your storage. The way the
Backup Tests (measured in seconds)
scanner works, the total number of ﬁ les that need to
be scanned, the size of the ﬁ les, and the fragmenta-
Low – Fragmented
1193
tion level of the disk all have a direct impact on the
Low – Defragmented
1130
length of the AV scan process. In our tests with the
Kapersky Lab AV solution, the disk defragmentation
Medium – Fragmented
2787
resulted in upwards of a 10 percent performance
Medium – Defragmented
2300
improvement—with the improvement being more
High – Fragmented
6960
signiﬁ cant as the test drives increased in size, number
High – Defragmented
6620
of test ﬁ les, and fragmentation.
While different backup tools will be differently affect-
VHD Start
ed by disk fragmentation, our tests showed one simple
This test measured the amount of time it took to
fact; defragmented disks back up faster. Individual runs
launch the saved test virtual machine. The VM was
demonstrated performance improvements of up to 20
launched from a saved state and timing stopped
percent with our test data set and the built-in Windows when the Hypervisor manager reported that the VM
Server backup. Our least effective test result, a large
was successfully started.
data backup that can represent a signiﬁ cant amount of
time, still showed an improvement of 5 percent. Our
VHD Start Tests (measured in seconds)
highest report results, which averaged a 17 percent
reduction in backup time, shows that reducing or
Low – Fragmented
62.3
eliminating disk fragmentation prior to backup will
Low – Defragmented
51
allow larger amounts of data to be backed up, espe-
Medium – Fragmented
60.7
cially if time is a constraint in your backup process. If
backup is run as a background application, reduced
Medium – Defragmented
58
fragmentation will allow for lower resource consump-
High – Fragmented
55.3
tion necessary for the backup process, minimizing
High – Defragmented
47
further the impact of the backup on active users of the
storage.
With as much as a 17 percent improvement in the start
time of the test virtual machine, the effects of fragmen-
The single,
Anti-Virus Scan
tation on the VHD are clear. This fragmentation will
consistent
For the AV scan test, we
also impact the performance of the VM itself, because
result that
performed a complete
all of the additional I/O necessary to read from a
appears in
scan of the test volume
severely fragmented VHD will reduce the performance
using the Kapersky Lab
of the virtual computing environment. Fragmentation
all of our
AntiVirus Version 6
must also be watched if your VMs are conﬁ gured with
tests is that
Windows Server software,
the dynamic disk option, which allows the virtual ma-
defragmented
current as of February
chine to grow the size of its storage as necessary. This
server drives
2009. The default con-
means that as the size of the VHD grows it will con-
ﬁ guration of the AV soft-
tinue to fragment into the available space on the hard
using
ware was used with only
drive. Making sure that the host machine hard disk is
Diskeeper
the test volume selected
regularly defragmented and managed will improve the
deliver better
for scanning. Timing was
performance of virtual machines running on the host
performance.
done using the AV ap-
and allow for the use of dynamic disk allocation within
plication.
the VM without danger of disk performance issues.
sponsored by Diskeeper
THE IMPACT OF DISK FRAGMENTATION ON SERVERS
5

Even with signiﬁ cant free space of the disk, as shown
by the white space in the fragmentation map (Figure
3), major fragmentation can still occur even without
VHD test volume.
Figure 4: Fragmentation map after automated defragmentation
by Diskeeper.
Server Application Tests
In the server application tests we looked at the
impact of fragmented storage on server-based ap-
Figure 3: Fragmentation map of VHD volume
plications. Other factors will have an impact on the
VHD Save
overall performance of these applications; optimiz-
This test measured the length of time required to save
ing storage strategies, including defragmentation,
the test virtual machine. From the Hypervisor manager,
reduces the impact of storage performance on the
the running machine was saved and timing stopped
overall application performance.
when the manager reported the save complete.
Exchange Test One
VHD Save Tests (measured in seconds)
In this ﬁ rst Exchange test, the client, a Windows
XP Professional Workstation running Ofﬁ ce 2007,
Low – Fragmented
365.3
uses Outlook to open 100 messages from the server.
Low – Defragmented
271.7
One hundred messages are highlighted then opened
simultaneously. Timing starts when the open is
Medium – Fragmented
409.3
launched and stops when all of the messages have
Medium – Defragmented
402
been opened and console control returns.
High – Fragmented
447.7
Exchange Test One (measured in seconds)
High – Defragmented
390.3
Low – Fragmented
7.7
With test results indicating as much as a 25 percent
performance improvement after defragmentation, the
Low – Defragmented
7
VHD Save tests show quite clearly the effect of writing
Medium – Fragmented
10.7
a very large ﬁ le to a fragmented hard drive. The more
Medium – Defragmented
8.6
fragments on the drive the less likely it will be that a
large ﬁ le can be written contiguously. And in the world
High – Fragmented
18.4
of virtualization, large ﬁ les are the standard, and the
High – Defragmented
11.6
need to be able to read and write those ﬁ les with a
minimum of fragmentation is a requirement to meet
While the impact of server fragmentation gets signiﬁ -
the basic ROI needs of the enterprise.
cantly greater as the disk becomes more fragmented,
even the common lower levels of fragmentation will
Automated background defragmentation results in a
have a large impact on user response time when you
major reduction in fragmentation even with an active
consider that hundreds of users may be accessing the
VHD (Figure 4). Regular use of the background de-
data store at the same time. Delayed response time
fragmenter will continue to minimize fragmentation.
for email users is a generator of a large percentage
of help desk calls, and implementing a defragmenta-
tion strategy can help to solve the problem. As our
tests show, allowing the data to become seriously
fragmented can have a major negative impact on the
6
THE IMPACT OF DISK FRAGMENTATION ON SERVERS
sponsored by Diskeeper

Exchange user experience with a 40 percent reduc-
mented environment not only improves load times
tion in performance in our highly fragmented test
but reduces the amount of disk thrashing necessary to
environment. Good defragmentation strategies result
manipulate the data and the amount of work that is
in fewer help desk calls.
necessary to later defragment the database.
Exchange Test Two
Table Key Creation (measured in seconds)
In this test, the contents of an existing folder were
moved to a new folder. Time to complete was mea-

Table 1
Table 2
sured from the client side.
Low – Fragmented
12.5
15.9
Low – Defragmented
12
14.9
Exchange Test Two (measured in seconds)
Medium – Fragmented
14.1
18.23
Low – Fragmented
9
Medium – Defragmented
12.4
17.1
Low – Defragmented
8
High – Fragmented
25.5
32.4
Medium – Fragmented
13.8
High – Defragmented
20.6
25.3
Medium – Defragmented
9

Table 3
Table 4
High – Fragmented
24.9
Low – Fragmented
26
35.4
High – Defragmented
12.3
Low – Defragmented
24.2
33
A new folder was created and the contents of the
Medium – Fragmented
32.3
49.1
Inbox were moved to the new folder. With our heav-
Medium – Defragmented
30.4
43.8
ily fragmented test environment showing a greater
than 50 percent performance improvement after
High – Fragmented
51
68.8
defragmentation it’s clear that this test was extremely
High – Defragmented
46.7
61.3
sensitive to higher levels of fragmentation on the
server. If users are often found reorganizing the data
In this test each table was opened, a ﬁ eld was se-
in the Exchange mailbox, the impact of fragmentation lected as the primary key, and the change was saved.
can be quite severe.
The table key creation times are directly related to
how much data SQL Server had to touch, and the
SQL Server Bulk Insert
level of fragmentation that had to be dealt with. SQL
We tested SQL Server 2008 with a bulk insert of
Server 2008 does a very good job of managing its
50,000 rows of data. The bulk insert is often the fast-
databases, but defragmentation shows appreciable
est method of getting data into a SQL Server data-
improvement in the performance of tasks such as this
base.
with a performance improvement of over 11 percent
in the most fragmented environments.
SQL Server Bulk Insert Tests (measured in seconds)
With the SQL queries, the two tests differ primar-
Low – Fragmented
22.1
ily in the amount of data that SQL Server returns in
Low – Defragmented
20.9
response to the query. The tests depict the effects of
manipulating the data on a fragmented drive with
Medium – Fragmented
31
peak performance improvements of approximately 18
Medium – Defragmented
25
percent.
High – Fragmented
53.3
SQL Query 1 – Simple (measured in seconds)
High – Defragmented
33.4
Low – Fragmented
23.9
As has been seen with the Exchange tests, a highly
fragmented database structure can have a severe
Low – Defragmented
22.3
negative impact on loading and extracting data from
Medium – Fragmented
28.2
server applications, with our test showing a perfor-
Medium – Defragmented
24.8
mance improvement of 40 percent in the most heav-
ily fragmented environment. Because Microsoft offers
High – Fragmented
43.5
APIs for moving open ﬁ les, defragmentation software
High – Defragmented
33
is able to safely work on database ﬁ les without risk of
data loss or corruption. Loading data into a defrag-
sponsored by Diskeeper
THE IMPACT OF DISK FRAGMENTATION ON SERVERS
7

SQL Query 2 – Complex (measured in seconds)
works better when the files are not fragmented. The
result is improved performance.
Low – Fragmented
35.3
Low – Defragmented
33.3
Throwing more
Medium – Fragmented
41.5
storage resources
Our test
(hardware) at a
Medium – Defragmented
38.5
results showed
problem should
High – Fragmented
61.3
be the last resort,
performance
High – Defragmented
50.8
because it only
improvements
masks the poten-
tial problems that
of upwards of
Conclusion
intelligent disk
The single, consistent result that appears in all of
defragmentation
30 percent.
our tests is that defragmented server drives using
addresses.
Diskeeper deliver better performance.
Quicker response time in databases and mail servers
Every application that touches the hard drive will
means that more time is spent getting work done,
benefit from a good tool that defragments and man-
rather than waiting for information to be delivered.
ages the files on your servers.
Diskeeper is the only true server defragmentation
Almost every role filled by Windows servers in your
software that runs silently in the background, con-
computing environment will benefit from the use of
tinually improving performance.
disk defragmentation software. The simplest file and
print services delivery requires a significant amount
With the current economic and business environ-
of disk I/O and will easily benefit from file defrag-
ment, maximizing ROI becomes even more critical.
mentation. As our simple tests show, even Exchange
Adding Diskeeper to your server toolkit gives you the
and SQL Servers benefit from defragmentation; read-
ability to get the maximum speed from your storage
ing and writing data with either application simply
subsystems of your existing hardware.
David Chernicoff is a technology consultant with
feature and product reviews for more than 20 years
a focus on the mid-market space, Windows IT Pro
and is coauthor of a number of operating system
Senior Contributing Editor, founding Technical
books, ranging from the Windows NT Workstation:
Director for PC Week Labs (now eWeek), former
Professional Reference (New Riders Publishing), to
Lab Director for Windows NT Magazine/Windows
the Microsoft Windows XP Power Toolkit (Microsoft
2000 Magazine (now Windows IT Pro) and formerly
Press), as well as over a dozen eBooks on topics
Chief Technology Officer for a network management
ranging from network switching topologies to pro-
tools ISV. David has been writing computer-related
duction FAX technology.
8
THE IMPACT OF DISK FRAGMENTATION ON SERVERS
sponsored by Diskeeper