| Bug #29352 | Falcon full table scan is slow | ||
|---|---|---|---|
| Submitted: | 26 Jun 2007 6:50 | Modified: | 26 May 2010 17:49 |
| Reporter: | Yoshinori Matsunobu | Email Updates: | |
| Status: | Unsupported | Impact on me: | |
| Category: | MySQL Server: Falcon storage engine | Severity: | S5 (Performance) |
| Version: | 6.0.1-alpha | OS: | Any |
| Assigned to: | Kelly Long | CPU Architecture: | Any |
| Tags: | F_PERFORMANCE | ||
[26 Jun 2007 6:50]
Yoshinori Matsunobu
[26 Jun 2007 9:27]
Hakan Küçükyılmaz
Verified as described. InnoDB mysql> select * from t1 where c2 = 1000; +-----+------+----------------------+ | c1 | c2 | c3 | +-----+------+----------------------+ | 999 | 1000 | ABCDEFGHIJabcdefghij | +-----+------+----------------------+ 1 row in set (6.31 sec) Falcon mysql> select * from t2 where c2 = 1000; +-----+------+----------------------+ | c1 | c2 | c3 | +-----+------+----------------------+ | 999 | 1000 | ABCDEFGHIJabcdefghij | +-----+------+----------------------+ 1 row in set (52.32 sec) Best regards, Hakan
[26 Oct 2007 9:24]
Kevin Lewis
Jim, Can you address this performance issue?
[30 Jan 2008 21:47]
Kevin Lewis
Theses triage values assume that the solution is a read-ahead capability
[31 Jan 2008 15:44]
Hakan Küçükyılmaz
Latest numbers with mysql-6.0-falcon-team tree on a 8-CPU/8GB machine
InnoDB:
- load 10 million rows
02:04 min, 01:39 min, 01:38 min
- select count(*)
04.57 sec, 04.57 sec, 04.57 sec
- select * from t1 where c2 = 1000
05.84 sec, 05.85 sec, 05.84 sec
Falcon, page size 4k:
- load 10 million rows
04:22 min, 04:21 min, 04:22 min
- select count(*)
49.57 sec, 49.90 sec, 49.44 sec
- select * from t1 where c2 = 1000
51.36 sec, 50.41 sec, 46.19 sec
Falcon, page size 32k:
- load 10 million rows
09:42 min, 09:41 min, 09:42 min
- select count(*)
01:13 min, 45.33 sec, 51.87 sec
- select * from t1 where c2 = 1000
46.90 sec, 51.40 sec, 50.67 sec
Result:
- page size has no effect on table scan
- table scan is a magnitude slower on Falcon than on InnoDB
[29 Aug 2008 20:08]
Kevin Lewis
Kelly, Here is a performance bug we have had in the list for a long time. We need to understand this difference that Hakan have verified twice over the last year.
