OpenSolaris

Printable Version Enter a New Search
Bug ID 6832481
Synopsis ZFS separate intent log bypass property
State 10-Fix Delivered (Fix available in build)
Category:Subcategory kernel:zfs
Keywords spbc_s10uX
Responsible Engineer Neil Perrin
Reported Against
Duplicate Of
Introduced In solaris_nevada
Commit to Fix snv_122
Fixed In snv_122
Release Fixed solaris_nevada(snv_122)
Related Bugs 6872047 , 6881114
Submit Date 21-April-2009
Last Update Date 25-January-2010
Description
Oracle manages 2 major types of files, the data files and the redo log files.
Writes to redo log writes are in the path of all transactions and low latency is a requirement. Thus we invented the separate intent log.

Data files are also the subject of writes from DB writers as a form of scrubbing dirty blocks to insure DB buffers availability. Write latency is much less and issue here and we are more concerned with acheiving a level of throughput. There is a high degree of concurrency of request which can also be tuned. 

Those writes are thus O_DSYNC write and today will target separate intent log blocks when present, possibly saturaring our devices and degrading the log writer latency.

This RFE suggest that a dataset such as one holding Oracle datafile could be tagged as avoiding SLOG blocks for it's synchronous write path. This will help increase oracle performance and ease deployment (today we have to create 2 pools one for log files and one for DB files which is more cumbersome than necessary). We expect that this dataset property would become a best practice for most Oracle deployments. 

Amberroad is already on board with the suggestion and would support it in the UI once available in zfs.
Work Around
N/A
Comments
N/A