Addenda and Errata for Oracle Core Chapter 6 Writing and Recovery
Back to Index
||In the Note on this page I mention the option for setting the redo log block size in 11.2; Charles Hooper has written a related note that you might want to read.
||In the side-bar with the title “Messages” I mention the log file sync timeout, and how this has changed from being fixed at one second up to 11.1, but being configurable from 18.104.22.168 with a new default value of 0.1 second (10 centi-seconds). This point came up in a discussion I had recently about a RAC system that was showing a lot of “Block Lost” errors on just one of its two nodes; the link being that the other node was showing a very large number of log file sync timeouts. When a block is sent from one RAC node to another the sending node first has to flush the log buffer to disc, and if this takes too long the received will time out at 0.5 seconds and report a lost block.
||In the Note on this page, I make some comments about the absolute and relative file numbers for a file. I failed to point out that the relative file number is “relative” to the tablespace. The numbering system is a bit strange, however, and this is for reasons of backward compatibility with Oracle version 7 which only allowed 1023 (or possibly 1022) datafiles in the database compared to the limit of 65,533 that appeared in version 8.In fact the limit for version 6 was only 63 (plus or minus 1) which is why it’s a good idea to use the functions supplied in the dbms_rowid package to translate data block addresses into (file number, block number) pairs rather than try to work out the convoluted way that Oracle mangled the bits to make version 7 backwards compatible with versin 6.
||In the side-bar with the title “Messages” last para: “… where the start of current free space is …” should be “… where the end of current free space is …”.
||The Note on this page references a couple of new instance statistics, but calls one of them “redo synch write time (usec)” – this should “redo synch time (usec)” without the “write”.
||Figure 6-5, right hand portion: there are four arrows pointing at the top buffer header. The two shorter arrows (i.e. the ones attached to the lower part of the buffer header) should be pointing away from the buffer header.
||Second paragraph last three lines (just above the heading “Checkpoints and Queues” the text reads “…it can unlink them from WRITE_AUX, relink them to REPL_MAIN, and…”; the relink goes to REPL_AUX, not REPL_MAIN
Back to Index
RSS feed for comments on this post.