Performance comparisons » History » Version 23
Greg Burri, 06/04/2012 01:26 AM
| 1 | 1 | Greg Burri | h1. Performance comparisons |
|---|---|---|---|
| 2 | |||
| 3 | h2. D-LAN vs ApexDC++ |
||
| 4 | |||
| 5 | We measured the transfer time and the number of CPU cycles for *one* transfer at a time of a *3.2 GiB file*. This is a particular case, the performance of both systems may varies in a concurrent downloading and uploading case. |
||
| 6 | 14 | Greg Burri | |
| 7 | h3. Setup |
||
| 8 | 5 | Greg Burri | |
| 9 | 1 | Greg Burri | * We used two computers on a gigabit LAN. |
| 10 | * There is a local DC++ hub: ADCH++ 2.8.0. |
||
| 11 | 4 | Greg Burri | * We didn't do any other operations during a transfer. |
| 12 | 1 | Greg Burri | * We download/upload only one file at the same time. We cleared queued and file before each measurement (the file is deleted from the trash). |
| 13 | |||
| 14 | * OS: Windows 7 64bits |
||
| 15 | 11 | Greg Burri | * Download computer CPU: Intel Core 2 Duo E8400, 4 GiB of RAM |
| 16 | * Upload computer CPU: Intel Core i7 860, 6 GiB of RAM |
||
| 17 | 1 | Greg Burri | * Apex version: v1.5.2.1276 (64bits) |
| 18 | 2 | Greg Burri | * D-LAN version: 1.1.0 Beta6 |
| 19 | 1 | Greg Burri | |
| 20 | 9 | Greg Burri | * Software used to count cycles: "Process Hacker 2":http://processhacker.sourceforge.net/ |
| 21 | |||
| 22 | 1 | Greg Burri | h3. Download speed |
| 23 | |||
| 24 | We repeated the transfer three times. |
||
| 25 | |||
| 26 | 13 | Greg Burri | | |_. # 1 |_. # 2 |_. # 3 |_. Average |_. Average download speed | |
| 27 | 7 | Greg Burri | |_. ApexDC++ | 160 s | 167 s | 163 s | 163.3 s | _20.1 MiB/s_ | |
| 28 | |_. D-LAN | 78 s | 87 s | 86 s | 83.7 s | _39.2 MiB/s_ | |
||
| 29 | 3 | Greg Burri | |
| 30 | 1 | Greg Burri | h3. CPU Cycles |
| 31 | 2 | Greg Burri | |
| 32 | 12 | Greg Burri | These values are measured in one download/upload of a 3.2 GiB file. |
| 33 | |||
| 34 | 8 | Greg Burri | | [Million of cycles] |_. Download |_. Upload |_. Total | |
| 35 | 7 | Greg Burri | |_. ApexDC++ | 199'613 | 272'655 | _472'268_ | |
| 36 | |_. D-LAN (Core + GUI) | 103'914 (88'800 + 15'114) | 24'086 (20'969 + 3'117) | _128'000_ | |
||
| 37 | 15 | Greg Burri | |
| 38 | h2. D-LAN vs DC++ - Downloading a lot of tiny files |
||
| 39 | |||
| 40 | 22 | Greg Burri | We measure the time to transfer a directory containing a lot of tiny files. The size of the directory is 2.74 GiB, it contains 3918 directories and 4684 files. Actually this is the directory of the game "League Of Legends":http://euw.leagueoflegends.com/. |
| 41 | 15 | Greg Burri | |
| 42 | h3. Setup |
||
| 43 | |||
| 44 | * We used two computers on a gigabit LAN. |
||
| 45 | * There is a local DC++ hub: ADCH++ 2.8.1. |
||
| 46 | * We didn't do any other operations during a transfer. |
||
| 47 | * We did only one transfer at the same time. We cleared queued and the directory before each measurement (the files are deleted from the trash). |
||
| 48 | 19 | Greg Burri | * Both output directories are located on the same hard drive. |
| 49 | 15 | Greg Burri | |
| 50 | * OS: Windows 7 64bits |
||
| 51 | * Download computer CPU: Intel Core i7 860, 6 GiB of RAM |
||
| 52 | 1 | Greg Burri | * Upload computer CPU: Intel Core 2 Duo E8400, 4 GiB of RAM |
| 53 | 18 | Greg Burri | * DC++ version: 0.799 with download completed notifications disable |
| 54 | 15 | Greg Burri | * D-LAN version: 1.1.0 Beta8 |
| 55 | |||
| 56 | h3. Download time |
||
| 57 | 1 | Greg Burri | |
| 58 | 21 | Greg Burri | We repeated the transfer two times. |
| 59 | |||
| 60 | 19 | Greg Burri | | |_. # 1 |_. # 2 |_. Average |_. Factor | |
| 61 | 20 | Greg Burri | |_. DC++ | 9min 39s | 9min 46s | 9min 43s | 7.2 | |
| 62 | 19 | Greg Burri | |_. D-LAN | 1min 20s | 1min 22s | 1min 21s | 1 | |
| 63 | 16 | Greg Burri | |
| 64 | It seems DC++ has some troubles when there is a lot of little file to transfer. |
||
| 65 | 23 | Greg Burri | |
| 66 | h3. Remarks |
||
| 67 | |||
| 68 | DC++ copied the empty files but didn't create the empty directories. |