[Progress Communities] [Progress OpenEdge ABL] Forum Post: RE: Transactions and Undos. Novel

[George Potemkin]

One more question about the jump note technologies: what is the expected ratio between the note reads and the note writes in case of transaction’s undo? V10.2B08 fixed the defect PSC00260992 related to the jump notes: Performance degradation with 100% busy MTX latch knowledgebase.progress.com/.../000042774 Before the fix sometimes (a few times per day) we were getting the following activity in BI file: 06/07/13 Activity: BI Log 12:53:50 06/07/13 12:53 to 06/07/13 12:53 (2 sec) Total Per Min Per Sec Per Tx Total BI writes 1 30 0.50 1.00 BIW BI writes 1 30 0.50 1.00 Records written 30 900 15.00 30.00 Bytes written 3998 119940 1999.00 3998.00 Total BI Reads 287 8610 143.50 287.00 Records read 36483 1094490 18241.50 36483.00 Bytes read 4636911 135847K 2318455.50 4636911.00 Clusters closed 0 0 0.00 0.00 Busy buffer waits 0 0 0.00 0.00 Empty buffer waits 0 0 0.00 0.00 Log force waits 0 0 0.00 0.00 Log force writes 0 0 0.00 0.00 Partial writes 0 0 0.00 0.00 Input buffer hits 11 330 5.50 11.00 Output buffer hits 0 0 0.00 0.00 Mod buffer hits 0 0 0.00 0.00 BO buffer hits 57 1710 28.50 57.00 Note reads were a thousand times larger than note writes if/when transaction's undo was affected by the defect. Example of undo of large transaction with the recent Progress version (OpenEdge Release 11.7.2): Time BI RReads BI RWrites State Flags RReads/sec RWrites/sec 11:54:48 0 0 TBGN FWD 12:01:02 0 1,420,738 Active FWD 0.0 3,798.8 12:31:05 0 8,782,310 Active FWD 0.0 4,087.5 13:01:07 0 13,076,270 Active FWD 0.0 2,384.2 13:31:09 0 21,415,904 Active FWD 0.0 4,630.6 14:01:11 0 35,583,963 Active FWD 0.0 7,866.8 14:29:28 HANGUP signal received. 14:31:13 2,607,605 44,419,628 Active UNDO 1,447.9 4,906.0 15:01:14 11,204,695 60,778,434 Active UNDO 4,773.5 9,083.2 15:31:16 17,365,329 72,351,940 Active UNDO 3,420.7 6,426.2 16:01:17 33,414,993 94,891,840 Active UNDO 8,911.5 12,515.2 16:31:19 46,070,338 123,329,803 Active UNDO 7,026.8 15,790.1 17:01:20 51,805,489 137,095,839 Active UNDO 3,184.4 7,643.6 17:31:22 64,467,997 158,531,609 Active UNDO 7,030.8 11,902.1 18:01:24 71,393,141 170,409,133 Active UNDO 3,845.2 6,595.0 18:31:26 80,826,268 187,387,152 Active UNDO 5,237.7 9,427.0 19:01:27 86,351,644 197,161,683 Active UNDO 3,067.9 5,427.3 19:31:28 97,928,534 213,323,807 Active UNDO 6,428.0 8,974.0 20:01:30 102,544,477 219,782,879 Active UNDO 2,563.0 3,586.4 20:11:07 ? 222,159,114 TEND UNDO ? 4,118.3 During UNDO phase the note writes were 1.5-2 times higher than the reads. Total number of writes during UNDO phase is 4 times higher than the total number of writes during FWD phase. Can it be explained by the jump note technologies? When I run the tests without the concurrent transactions (without the jump notes) the results were quite expected: Note Reads (during UNDO) = Note Writes during UNDO = Note Writes during FWD The formula means that Note Reads = Total Writes / 2. It seems to be true for the example above. Is it just a coincidence? Duration of FWD phase: 9280 sec = 02:34:40 Duration of UNDO phase: 20499 sec = 05:41:39 Is it possible to predict the duration of UNDO phase / the activity during UNDO phase?

Continue reading...