Age | Commit message (Collapse) | Author |
|
|
|
|
|
|
|
|
|
limited than this--at least tag node.def and multiple
configuration on commit.
tested fix with limited regression test.
also slightly modified break on priority values where embedded multinodes with priority values are treated as discrete transactions (which
isolates failures within the multinode and not the parent of the multinode).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
By calling fsync, we can be more assured that file actually makes
it to disk.
|
|
|
|
This fixes several problems:
1) Only try mounting once which avoids several I/O errors
2) Mount floppy with safer options for security
3) Use more optimum methods to reduce boot time
This should really be done by better udev support. But the root of the
problem lies in the Debian upstream package maintainer.
He is hostile to later advancements, and still maintains a static
/dev infrastructure as well as devfs compatiablity!
Ubuntu, SUSE, and Redhat have all migrated to creating devices via udev;
this would allow floppy device to only exist if floppy is present.
And then inserting a floppy could cause udev add event to mount
/media/floppy...
|
|
On normal boot, no changes to config.boot or modes are
needed.
|
|
|
|
Rest of Bug 4428
|
|
|
|
|
|
|
|
Previously the script determined the driver that a NIC was using, then
used that to parse the queue names from /proc/interrupts. Now, it
determines the naming convention just by looking at the queue names.
Now the script should work for all NICs that support one of the two
standard naming conventions.
Also added some support for single-queue NICs.
|
|
references to old values.
|
|
|
|
and should be considered depricated.
|
|
diverges from the node.defs
commit behavior has not changed yet, but can be completely switched over to node.def priority system with simple change.
|
|
will now display both priorities:
interfaces (t: 0, p: 850,850)
the first value is from the old priority file and is the active value, the second value is the value retrieved from the node.def and is not used (yet).
|
|
|
|
Bug 5014
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
vyatta config token at end of file. this will prevent the
accidental overwrite of non vyatta config files. check on prevent existing files that don't contain ' === vyatta-config-version:'
from being saved to.
|
|
-b
Breaks on each priority group before execution and waits for user keystroke to proceed. this allows for examination of system before and after each priority group has been applied
|
|
|
|
|
|
|
|
deleted leafs. Partial regression run on changes, results map to 10/10 build.
|
|
|
|
This change implements a mechanism for setting the IRQ affinity of NIC
interrupts in a way that is optimally balanced. The affinity settings
are static, and are set up at configuration time. I've added "auto"
as a new legal value to the "interfaces ethernet ethX smp_affinity"
configuration parameter. If this value is set, then the system
attempts to perform automatic IRQ affinity assignment.
At this time, automatic IRQ affinity is supported only for multiqueue
NICs that use the igb or bnx2 drivers. Even with these two,
differences in the organization of queues and their naming convention
necessitate some NIC-specific code. The assignment strategy attempts
to assign all queues of all NICs that share the same queue number to
the same CPU. This has been shown to be optimal for IP forwarding
workloads in 2.6.31 kernels.
Setting the value to "auto" for NICs that use other drivers will be a
no-op.
|
|
|
|
will improve performance by avoiding the piecewise priority based copy and just
copy the complete configuration once. On specific test configuration performance improvement brought commit boot time down from >15 minutes to less than 10
minutes.
|
|
|
|
|