Skip to content
Test Plan for 4096 disks
Test Plan
-
Objective
- Demonstrate that a Linux system can address, probe and do some amount of IO on a disk subsystem with 4096 disk
- Test this feature in conjunction with other features, including udev, filesystems, LVM, EVMS
Outputs
- Once code is mature and equipment is available, integration test with multipath IO, Hotplug CPU, Hotplug memory, Hotplug IO Subsystem.
- When equipment is available on new architectures or IO sub-systems, reuse the scripts to quickly repeat the tests
- Reuse the tests to do regression tests at yet to be defined points in Linux 2.6 development
Phases
- Phase One - ISCSI testing at OSDL for script development.
- Mostly DONE - markh patches to iscsi list?
- Phase Two - Physical device testing at IBM. Test with udev filesystems ext3 and reiser, LVM and EVMS.
- Tested 256 physical devices with ext3
- Phase Three - Post request for machine time on on other architectures and io subsystems and utilize as they become available. Integration with the other products mentioned above as equipment and/or code becomes available.
Gating Factors
- Phase One -
- working iSCSI targe code ( unh|| Netapp ) DONE
- Phase two -
- configured system w/ 4096 physical LUNs
- Phase three
- donations of other equipment
- Maturity of Multipath I/O, Hotplug features.
- Test suites for MPIO and Hotplug.
Actions
- Phase one
- Get unh-iscsi target code running on OSDL system - DONE
- Create scripts to parse dmesg and count LUNs - DONE
- Create scripts to configure large number of LUNs on target - DONE
- Create/understand process of discovering LUNs on client - DONE
- Create scripts to install and configure large number of ext3 or reiser filesystems on target. - DONE
- Create scripts to install and configure udev on system. ( NoT NEEDED )
- Create scripts to install and configure LVM or EVMS on system.
- Create IO test harness with iozone and/or disktest ( DONE )
- Run demonstration at OSDL, discover/fix code issues ( DONE )
- Repeat until OSDL rig capacity exhausted.
- Phase Two
- Document IBM setup, ensure it can be torn down and rebuilt ( DONE )
- Run iozone/disktest on small number of spindles to establish baseline IO performance ( DONE )
- Configure full 4k spindles. ( DONE @ 1024 spindles )
- Re-run IO tests to verify raw performance. ( DONE @ 1024 spindles )
- Install and configure udev on 4k spindles. Run raw io tests against udev devices. ( DONE 1024 spindles )
- Install and configure ext3 file systems. Run buffered and unbuffered io tests against the file system. Same for reiser. ( DONE for ext3 @ 1024 spindles )
- Install and configure LVM. Run raw io tests against the LVM devices. Same for EVMS.