IAM ==== PSARC 2008/769: Multiple disk sector size support. Submitter: Larry Liu Owner: Mark Carlson SUMMARY ======= ISSUES ======= Issues for commitment (03/11/2009): jdc-0 What has changed since inception? 1. Corrected the Imported Interface table in questionnaire. 2. Added block diagram in the design doc. 3. Added RMW design. 4. Added emulation mode support with reporting the physical block size. 5. Changed the Exported Interface table according to the design change. jdc-7 If I have an old application issuing DKIOCGMEDIAINFO on an upgraded system, won't the kernel's copyout write past the end of the user's older dk_minfo structure and corrupt memory? (Likely need to renumber ioctl.) The team will fix this with a new dkio(7I) ioctl and leave DKIOCGMEDIAINFO unchanged. See the file: NewIOCTL.txt in the case directory. jdc-8 Since RMW logically occurs only when the application is ignorant of alignment *and* issues an unaligned access, and thus all modified applications will "do the right thing" to avoid any need for performance-harmful RMW activity, why is this an option? Why not have RMW enabled all the time? RMW comes with a performance penalty. The team does not want to leave a hidden performance issue lurking and would prefer to let customers make an explicit choice to either change their applications or run under low performance. ram-1 Is there any hard limit at a 4k sector size? There doesn't appear to be one. At some point, the 4k size barrier may be broken in the future. On x86 platform, the PAGE_SIZE is 4K. So any sector size bigger than 4K could cause problems in page io, and would require a more extensive project. ram-2 You should inform the SAM-QFS team of your proposed changes. They are likely to be interested in larger sector sizes. Yes. We already talked to engineers in SAM-QFS about our large sector size project. And we will work with them to make sure large sector size disk works fine with SAM-QFS. ram-3 Is there any impact on use of a raw device partition as swap, due to increased sector size? No. A partion on a large sector size disk can be used as swap just like the traditional 512 byte disk. gcs-1 It looks like you've changed various structures to add new fields in the middle (e.g., struct disk_type has a newly added lbasize field preceding the existing capacity field). What are the implications for binary compatibility? The struct disk_type is part of the format(1M) internal structure. There should be no issue for binary compatibility as a result. Instead of the change to struct dk_minfo in the original proposal, we will be adding a new data structure dk_minfo_ext. Other than that, there are no changes to data structures proposed by this project. VOTE ==== Approve - Deny - Abstain - Not Participating (NP) - THE NEXT STEP ============= Vote will be held next week.