Defragmentation Performance Testing

 < Day Day Up > 



Surprisingly, until a couple of years ago, defragmentation had never been thoroughly benchmarked, an omission rectified by National Software Testing Lab (NSTL) of Conshohocken, Pennsylvania. Various reports have been done by NSTL on defragmentation performance testing. Since 1999, NSTL has conducted several benchmarks. The first focused on two of the most common NT configurations for that time, determined through a survey of 6000 system managers running Windows NT. A Pentium II 266-MHz workstation with 96 MB of memory and a 2-GB IDE hard drive, running Outlook and Excel, showed a performance leap of 74.5 percent with a defragmented drive. On a Pentium II 400-MHz workstation with 128 MB RAM and a 4.2-GB hard drive, the improvement rose to 80.6 percent.

Two popular server configurations also underwent testing. A dual Pentium PRO 200 with 128 MB of memory and five 4-GB SCSI hard drives, running RAID-5, Exchange Server, and SQL Server 7.0, recorded an increase of 19.6 percent on a defragmented drive. On a Pentium PRO 200 with 64 MB of RAM and two 4-GB SCSI hard drives, running Exchange and SQL 7.0, performance rose by a hefty 56.1 percent.

National Software Testing Lab tested the effects of fragmentation on files of all types and sizes. The Microsoft Outlook tests, for example, included opening 50 messages simultaneously; moving messages from the inbox to a separate folder; opening (and displaying the to:, from:, subject:, and date:) a large subfolder; a full text search of all messages in a folder for a specific string; and a filter that displayed all messages in a folder that contained an attachment. Each of these tests was executed on the system when the personal folder was fragmented and defragmented. The SQL, Exchange, and Excel tests were carried out in a similar manner.

The results of this testing might exceed what some would have believed possible from defragmenting, but even if the real-world improvements are only half as good as NSTL recorded under laboratory conditions, these improvements represent performance numbers to rival many hardware upgrades.

Windows 2000 Tests

During the year of 2000, NSTL ran additional tests on Windows 2000. To compare Windows NT with Windows 2000, the same configurations were used. On a Pentium II 266-MHz workstation, for example, with 96 MB of memory and a 2-GB IDE hard drive running Windows 2000, Outlook, and Excel, an increase in system speed of 85.5 percent was recorded. On a Pentium II 400-MHz workstation with 128 MB RAM and a 4.2-GB hard drive, also running Windows 2000, Outlook, and Excel, defragmentation resulted in a performance jump of 219.6 percent (see Exhibit 2).

Exhibit 2: Windows 2000 and Windows NT Defragmentation Benchmark Comparison

start example

System Configuration

Windows 2000 (%)

Windows NT (%)

Pentium II 266 w/96 MB RAM, 5 × 2-GB IDC hard drive workstation running Excel and Outlook

85.5

74.5

Pentium II 400 w/128 MB RAM, 1 × 4.2-GB hard drive workstation running Excel and Outlook

219.6

80.6

Dual P PRO 200 Server w/128 MB RAM, 5 × 4-GB SCSI hard drives w/software RAID running Exchange and SQL

61.9

19.6

P PRO 200 server w/64 MB RAM, 2 × 4-GB SCSI hard drives running Exchange and SQL

83.5

56.1

end example

Examine the comparisons made in Exhibit 2. In the server category, a dual Pentium PRO 200 with 128 MB of memory and five 4-GB SCSI hard drives, running Windows 2000, RAID 5, Exchange Server, and SQL Server 7.0, showed an increase of 61.9 percent on a defragmented drive. On a Pentium PRO 200 with 64 MB of RAM, two 4-GB SCSI hard drives, running Windows 2000, Exchange, and SQL 7.0, performance rose by 83.5 percent. Windows 2000 scores were even higher than NT in each category.

During testing, care was taken to approximate real-world Windows 2000 operating conditions. The Excel 2000 test, for example, repeatedly opened and saved Excel files to the fragmented (or defragmented) partition. These documents (varying in size from 5 to 20 MB) contained formulas that the spreadsheet auto-calculated when it was opened. For SQL Server 7.0, testing focused on reading queries from a database and displaying the results from a variety of tables.

Upon completion of testing, NSTL published the white paper System Performance and File Fragmentation in Windows NT, in which it concluded: "Theoretical analysis and real-world performance testing demonstrate that fragmentation has an adverse impact on system performance. The best way to avoid these fragmentation problems, and to keep the system running at optimal performance, is to run a defragmentation program on a regularly scheduled basis." (The full text of the white paper is available at http://www.execsoft.com.)

XP Performance Tests

In Defragmentation Performance Testing — Windows XP, NSTL reported the results of tests that revealed the penalties imposed on Windows XP system performance by fragmentation. Their report documents on-average gains of 109 percent after defragmentation. According to Lloyd Holder, chief executive officer of NSTL, his company established significant performance gains through the use of cost-effective network defragmentation tools. In his opinion, in many of the cases tested, gains of this magnitude often surpass improvements realized through substantial hardware upgrades.

National Software Testing Lab also conducted testing on machines running common enterprise applications under Windows XP Home and XP Professional. The primary objective was to design and conduct a series of tests that would demonstrate the effect of hard drive fragmentation on system performance when common business applications are being run. Researchers used an IBM NetVista Type 6825-12U system with a 36-GB SCSI hard drive. It featured a Pentium IV processor running at 1.6 GHz. The system's BIOS was an IBM 20KT21AUS dated 10/10/2001. The system had 512 MB of SDRAM. The hard drive was an SCSI IBM Ultrastar DDYS-T86950N 36GB with an Adaptec 2916ON SCSI controller card.

Testing personnel partitioned the hard drive into two separate 18-GB partitions (drives C: and D:). They installed XP on the C: drive, and Excel and Outlook on the D: drive. A tool named FragmentFreespace.exe was used to create a 50% fragmented hard drive. The hard drive was filled to 20 percent capacity with approximately 3000 files. A hard drive imaging application was utilized to image this configuration, and it was used as the baseline for each test. NSTL then defragmented the drive using Executive Software's Diskeeper 7.0. To approximate real-world conditions, the Excel test repeatedly opened and saved several Excel files to the fragmented (or defragmented) partition. These files varied in size and contained formulas that were auto-calculated when a spreadsheet was opened. Outlook tests focused on its personal folder database, performing such actions as moving messages from the inbox to a separate folder and a full text search of all messages in a folder for a specific string.

Overall, NSTL found that a system defragmented with Diskeeper performed the tests significantly more quickly than a fragmented system. When running Outlook, researchers recorded an increase in performance of 67.90 to 176.10 percent after defragmentation. Excel tests showed gains of 83.67 percent. Overall, test results show an on-average increase of 109 percent after defragmentation using Diskeeper.

Testing the Windows 2000 Disk Defragmenter

National Software Testing Lab recently took its testing a step further by comparing the results of using Diskeeper on Windows 2000 with those achieved by using the scaled-down defragmenter Windows Disk Defragmenter (WDD), which is built into the operating system. This investigation covered two key areas: speed and effectiveness. In all, four common enterprise configurations were used in the testing (see Exhibit 3). NSTL's Holder states that, after extensive testing, NSTL found that Diskeeper was between 300 and 500 percent faster than the WDD. (A copy of the NSTL report, Comparison Testing: Diskeeper Vs. the Windows 2000 Disk Defragmenter, is available at www.nstl.com.) In addition, NSTL discovered that Diskeeper scored three to five times faster than the "lite" defragmenter that is built in to Windows XP and Windows 2000. Holder believes that, of the three operating systems, Windows 2000 was found to be the most severely affected by disk fragmentation.

Exhibit 3: Diskkeeper versus Windows 2000 Disk Defragmenter

start example

System Configurations (GB)

Diskeeper

Windows 2000 Disk Defragmenter

9

32 minutes, 15 seconds

1 hour, 34 minutes, 9 seconds

30

3 hours, 13 minutes, 38 seconds

14 hours, 42 minutes, 58 seconds

150

6 hours, 19 minutes, 48 seconds

21 hours, 49 minutes, 44 seconds

2 × 60

3 hours, 9 minutes, 44 seconds

9 hours, 23 minutes, 30 seconds

end example

Running the Windows 2000 Disk Defragmenter

Although the WDD is not recommended in an enterprise setting, a brief explanation of how to use it is provided here. Depending on the version of XP/2000, WDD can be found in a couple of ways. Click on Start Programs System Tools Disk Defragmenter or open the Disk Management snap-in and click on Disk Defragmenter. When the WDD window appears, you are given a choice of Analyze or Defragment. It is advisable to choose Analyze first to discover exactly how fragmented a disk has become. A visual representation of fragmentation soon appears along with the Analysis Complete dialog box. This visual representation, though, is rather basic, consisting only of fragmented files being represented in orange, contiguous files in blue, system files in green, and free space in white. You can also view a report to see the extent of fragmentation in specific files. When you choose Defragment in the WDD, the utility works its way through the volume, rearranging files to make them contiguous. When the defragmentation run is completed, "before" and "after" images are displayed. In addition to NTFS, WDD supports FAT and encrypted files.

Anyone newly installing XP or Windows 2000 should immediately analyze the state of fragmentation on each volume as soon as installation is complete. While one would reasonably expect the disks to be free of file fragments, this does not turn out to be the case. Most of the time, considerable amounts of fragmentation are to be found because the installation process creates numerous files and directories that are soon deleted. This situation is exacerbated by installing a browser, Office, service packs, and other applications. Most users end up with a severely fragmented system no matter what they do to try to prevent it.

Users of the Windows 2000 version of the WDD were able to defragment only one volume at a time. If they wanted to address multiple volumes, they had to wait for the utility to complete, choose a different volume, and restart the process. For Windows XP, Microsoft added a simple scheduling script to give system administrators some means of scheduling defragmentation for multiple users/volumes. To use the scheduler:

  • Click Control Panel System Maintenance Scheduled Tasks.

  • Double-click Add Scheduled Task, which starts the Scheduled Task Wizard. In the list of applications, click Defrag and Next.

  • Select Task Frequency and click Next.

  • Select the Time/Day when the Task is to run.

  • Enter the name and password of a user with administrative privileges (a user who has control over the computer, can install software, and can change user passwords).

  • Check the box to open Advanced Settings.

  • In the Run box, add the letter of the drive to defragment so, for example, it looks like C:\windows\system32\defrag.exe c:.

The WDD can run only one volume at a time, so multiple partitions call for the addition of a separately scheduled task for each volume. In the Control Panel, open Scheduled Tasks, and in the File menu point to New and click Scheduled Task. Give the new task a name, and complete the necessary information. Be sure to schedule defragmenting sessions far enough apart so they do not overlap.

WDD Cautions

Several issues should be noted with regard to using the WDD.

Speed

According to NSTL, the WDD is three to five times slower than Diskeeper.

Scheduling

When it comes to scheduling many machines at once, the procedure for using the WDD becomes quite complex and involves the use of scripts and batch files. This may prove difficult to learn, manage, execute, and monitor for less experienced systems administration staff. Scheduling itself, for instance, largely depends on guesswork. As parallel defragmentation is not possible, defragmentation of one volume must be completed before another one can begin. The system administrator has to guess how long a specific volume will take to defragment, allow for more than enough time for it to complete, and set the times of additional volumes accordingly. However, as the WDD often hangs up or takes a long time to run on badly defragmented disks, on disks lacking free space, or on large disks, it is quite possible that the first defrag run will not be completed before the next run is due to start. On these occasions, the entire defrag schedule for that machine is aborted. As the scheduler does not allow for exclusion lists, it is impossible to specify files and directories that are to be excluded from the defragmentation process. On machines where a large number of temporary files are created and deleted on an ongoing basis, this feature would be particularly useful. One final point on scheduling bears a mention with regard to mixed Windows environments. The WDD can schedule only Windows XP machines; thus, a system administrator with a mixed Windows environment cannot schedule WDD on Windows 95, 98, Me, NT, and 2000 machines. As these operating systems are likely to form the bulk of user desktops in the foreseeable future, the scheduling capabilities of the utility are therefore minimal.

Priority Settings

Ideally, scheduled defragmentation processes should only run at low priority. This would ensure that all of the user's applications were serviced before the WDD. In practice, however, the utility contains no priority settings. Thus, the defragmentation processes compete with other applications for system resources. As a result, mission-critical applications run more slowly while defragmentation is being done. Obviously, one way around this is to schedule volumes to be run during off-peak hours. During such times, though, the defrag program may interfere with backup schedules and other off-hours maintenance processes. As the utility is slow and must complete volumes sequentially rather than in parallel, it is quite likely that resource conflict will occur during the course of the night.

Administrative Privileges

The WDD requires having administrative privileges in order to operate. This means that individual users, realizing that fragmentation must be slowing down their machines, cannot take action except to bring it to the attention of the system manager. Thus, everything relating to defragmentation must be referred to one person. The administrator then has to work out how to set up the scripts and batch files involved and know the names of all the systems and the volume configuration of the systems. This is no small task in a small network, never mind a major enterprise. If Windows 2000 machines are also present, the administrator has no choice but to walk from machine to machine to run defrag manually, volume by volume.

Free Space

The WDD requires a lot more free space to operate than do third-party defragmenters. A minimum of 15 percent free space must be available on the hard disk for WDD to run. The defrag program uses this space as a sorting area for file fragments. This does not take into account the MFT Zone, which accounts for another 12.5 percent of the disk, so over 25 percent must be free for the utility to function adequately; otherwise, it tends to thrash around for hours and either stalls or fails to complete properly. To increase the free space on a volume, delete unneeded files or move them to another disk.

Thoroughness

National Software Testing Lab tests on the thoroughness of defragmentation revealed that WDD does a far less thorough job than do third-party products. It tends to have trouble with large files, large disks, badly fragmented files, and volumes with a lack of free space. It also fails to address some types of system files (Exhibit 4). On the 30-GB drive, for example, the WDD left more than 10 percent of the files severely fragmented. Why this lack of thoroughness? It is due primarily to the inability of the WDD to take care of fragmentation in some system files such as the Paging File and the lack of a boot-time defrag function. With a consolidated Paging File, WDD has nothing blocking its redistribution of files or free space on a drive, but when fragmentation of the Paging File sets in, the amount of free space becomes restricted, eventually reaching the point where the utility finds it difficult to function. Users will notice a steady diminishing of performance over time.

Exhibit 4: Thoroughness of Defragmentation

start example

 

After Defragmentation

Before Defragmentation

Initial Total Fragmented Files/Total Excess Fragment

System Configurations (GB)

Initial Total Fragmented Files/Total Excess Fragments

Diskeeper

Windows 2000 Disk Defragmenter

9

805/3,934

0/0

76/1,603

30

15,387/56,785

5/4,289

1,718/9,274

150

49,847/325,608

0/0

24/318

2×60

7,361/18,417

0/0

26/998

end example

The WDD does not address Paging File fragmentation for safety reasons. The defrag APIs built into the operating system are not designed to operate when the Paging File is in operation (i.e., whenever the machine is running). Third-party defragmenters take care of Paging File consolidation offline during boot-time defragmentation, but the WDD has no ability to conduct boot-time runs. For users of Windows 2000, the situation is even worse. The APIs used in Windows XP have been adjusted to allow defrag of the MFT and directories online, but the WDD in Windows 2000 safely defrags only the directories, not the MFT.

I/O Overload

The purpose of eliminating fragmentation is to handle the I/O overload at the disk level caused by files that are shattered into many pieces, but those attempting to deploy the WDD in an enterprise setting will end up costing the company far more in terms of administration expenses and productivity losses than the costs to license a reliable third-party defragmenter. WDD is probably adequate for most home users, but in the enterprise it fails to aggressively eliminate fragmentation, thus resulting in curtailed system performance. System administrators are advised to rely only on proven third-party, network-capable defragmentation utilities that are certified for use on Windows XP/2000 and that can function across all versions of the Windows operating spectrum.



 < Day Day Up > 



Server Disk Management in a Windows Enviornment
Server Disk Management in a Windows Enviornment
ISBN: N/A
EAN: N/A
Year: 2003
Pages: 197

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net