OmCheeLin commented on PR #688:
URL:
https://github.com/apache/skywalking-banyandb/pull/688#issuecomment-3003471468
| Pattern | Type | ns/op | Compression Ratio | Encode
Time (ns/op) | Decode Time (ns/op) | Encode+Decode (ns/op) |
| ------------------- | ---------- | ------ | ----------------- |
------------------- | ------------------- | --------------------- |
| const | old\_str | 386288 | 0.0004375 | 193984
| 409673 | 603657 |
| const | new\_int64 | 423046 | 0.0001125 | 145797
| 486149 | 631946 |
| deltaConst | old\_str | 379194 | 0.4867 | 206682
| 381112 | 587794 |
| deltaConst | new\_int64 | 417510 | 0.0001250 | 148899
| 469816 | 618715 |
| delta | old\_str | 416599 | 0.6001 | 235115
| 409261 | 644376 |
| delta | new\_int64 | 518178 | 0.1251 | 222822
| 516450 | 739273 |
| incremental | old\_str | 426709 | 0.6008 | 246630
| 425872 | 672502 |
| incremental | new\_int64 | 516221 | 0.1251 | 213302
| 517664 | 730966 |
| small\_fluctuations | old\_str | 888737 | 0.1301 | 620380
| 500583 | 1120963 |
| small\_fluctuations | new\_int64 | 501009 | 0.1251 | 200673
| 506057 | 706730 |
| random | old\_str | 570259 | 0.2552 | 349712
| 483792 | 833504 |
| random | new\_int64 | 591644 | 0.1407 | 267363
| 587271 | 854634 |
This is a benchmark for the int64 type (traditional encoding vs. new
encoding). It shows that the compression ratio (encodedSize / rawSize) has a
significant advantage, while the encode + decode time (ns/op) difference is
minimal.
| Pattern | Type | ns/op | Compression Ratio | Decode
Time (ns/op) | Encode+Decode (ns/op) | Encode Time (ns/op) |
|-------------------|----------------|----------|-------------------|---------------------|-----------------------|---------------------|
| const | old_str | 417,934 | 0.0004375 |
446,205 | 647,333 | 201,128 |
| const | new_float64 | 564,110 | 0.01572 |
584,195 | 785,398 | 201,203 |
| incrementing | old_str | 416,769 | 0.4868 |
423,943 | 655,714 | 231,770 |
| incrementing | new_float64 | 2,779,417| 1.031 |
1,627,414 | 3,048,532 | 1,421,118 |
| small_fluctuations| old_str | 1,215,406| 1.752 |
572,129 | 1,472,848 | 900,719 |
| small_fluctuations| new_float64 | 2,758,764| 1.031 |
1,567,165 | 3,012,954 | 1,445,789 |
| random | old_str | 718,923 | 2.180 |
455,081 | 1,006,748 | 551,668 |
| random | new_float64 | 2,649,813| 1.031 |
1,545,288 | 2,935,953 | 1,390,665 |
This is float64 (traditional encoding vs XOR encoding). The XOR encoding
seems to perform worse than expected. Except for the random pattern case where
the compression ratio is slightly better, in other cases the compression ratio
and encode+decode time (ns/op) are much worse than the traditional method.
The benchmarks for tag and column are similar. Since the test data remains
unchanged, I have already tested them, so I only wrote the benchmark tests
under the measure package.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: notifications-unsubscr...@skywalking.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
