Sentry Spelling Checker Engine

Home Site index Contact us Catalog Shopping Cart Products Support Search

You are here: Home > Products > Developer tools > Sentry Spelling Checker Engine > How we measured performance

How we measured the performance of Sentry Spelling Checker Engine

• Sentry Spelling Checker Engine Windows SDK
  

• Sentry Spelling Checker Engine Source SDK
  
• Sentry Spelling Checker Engine Java SDK

Sentry Spelling Checker Engine Windows SDK

To measure the performance of the Sentry DLL, we used the "Performance" demo in the Sentry Example Program, written in C++/MFC, included with the Sentry Windows SDK. We ran the tests on a 2.66 GHz computer running Windows XP. The machine has 1GB RAM and 100GB of disk space.

To perform the performance measurements, we obtained the English text of War and Peace from Project Gutenberg. The novel contains 564,224 English words and about 3.2 million characters.

The option settings can affect performance. The following option settings were used:

• Ignore capitalized words: Disabled
  

• Ignore all-caps words: Disabled
  
• Ignore words with numbers: Enabled
  
• Ignore words with mixed case: Disabled
  
• Ignore domain names: Enabled
  
• Report doubled words: Enabled
  
• Case sensitive: Enabled
  
• Phonetic suggestions: Disabled
  
• Typographical suggestions: Enabled
  
• Suggest split words: Enabled
  
• Auto correct: Disabled

The document was then checked 10 times, and the time required to check the spelling of the document was recorded by the software. The following times were recorded for each run:

1. 1547 ms
   

2. 1562 ms
   
3. 1562 ms
   
4. 1547 ms
   
5. 1547 ms
   
6. 1547 ms
   
7. 1547 ms
   
8. 1547 ms
   
9. 1562 ms
   
10. 1547 ms

Average time: 1551.5 ms, or 1.5515 seconds. Sentry checked War and Peace in less than 1.6 seconds, at an average rate of 363,663 words per second (564224 / 1.5515), or about 2.7 microseconds per word.

Sentry Spelling Checker Engine Source SDK

All tests were conducted using the Perform Demo program. Source code for the Perform demo is included with the Sentry Source SDK. The Perform demo uses memory-based dictionaries. The demo was built as a 32-bit Windows console-mode application. To run Perform, enter the name of an ASCII text file on the command line:

perform file.txt

We ran the tests on a 700MHz Pentium III computer running Windows 2000. The machine has 256Mb RAM and 20Gb of disk space.

To perform the performance measurements, we obtained a large document (part of a novel, actually) from the Web. The document contains 37,732 English words and 238,087 characters. The document contains 4,831 different words, ranging from "the" (1643 occurrences) to words like "confused" and "sentient" (one occurrence).

The document was checked using Perform 10 times. Perform displayed the following times for each run:

1. 0.19s
   

2. 0.14s
   
3. 0.16s
   
4. 0.15s
   
5. 0.14s
   
6. 0.14s
   
7. 0.14s
   
8. 0.15s
   
9. 0.12s
   
10. 0.13s

Average time: 0.146s.

Sentry checked the document at an average rate of 181,719 words per second (26531 / .146).

Sentry Spelling Checker Engine Java SDK

All tests were conducted using the PerformanceTest demo which comes with the Sentry Java SDK. We ran the tests on a 700MHz Pentium III computer running Win 2000. The machine has 256Mb RAM and 20Gb of disk space.

To perform the performance measurements, we obtained a large document (part of a novel, actually) from the Web. The document contains 41,164 English words and 238,087 characters. The document contains 4,831 different words, ranging from "the" (1643 occurrences) to words like "confused" and "sentient" (one occurrence).

The document was checked using PerformanceTest 5 times using Java 1.4.2. The following results were displayed for each run:

1. 1963ms (20969 words/second)
   

2. 1963ms (20969 words/second)
   
3. 1963ms (20969 words/second)
   
4. 1953ms (21077 words/second)
   
5. 1943ms (21185 words/second)

Average checking rate in words per second: 21034.