forked from openzfs/zfs
-
Notifications
You must be signed in to change notification settings - Fork 7
/
TODO
51 lines (39 loc) · 2.08 KB
/
TODO
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
SUMMARY OF MAJOR KNOWN PROBLEMS IN v0.4.6 (Development Release)
* Fault Management (FM) and sysevent support / analog.
bugzilla 14866, 15645
This is probably the biggest remaining chunk of work. Linux has no
direct equivalent of the Solaris Fault Management Architecture (FMA)
and we need one. All fault information is currently ignored and no
disk errors are even logged. We need to settle on a design for this
but minimally it needs to log the events to syslog.
* Implement the ZVOL.
bugzilla xxxxx
This should be pretty staight forward now that the DMU is fully
implemented and solid. It just needs to be done.
* Implement the ZPL.
bugzilla xxxxx
Getting basic ZPL support should be pretty straight forward. Moving
beyond that to fully integrate with the VFS for things like mmap and
file locking will be trickier.
* Integrate the ZFS-FUSE port in to this code base.
bugzilla xxxxx
Merging the zfs-fuse code base in with this project would be nice from a
code maintence standpoint. This code base is quite a bit newer than
zfs-fuse and it already provides a libzpool library for zfs-fuse to link
against. This should be a pretty straight forward addition.
* Emulate kthreads with pthreads in userspace.
bugzilla xxxxx
There is a patch available for this but each time I've integrated it
I've observed SIGSEGVs in ztest. Once this patch is in place ztest
can be used to use the kthread API which brings us one step closer
to being able to run it in the kernel as an additional sanity check.
* DMU Performance
bugzilla 13566
While performance is currently not bad it is not where it needs to be
for production use. The latest test results which can be found in the
docs directly show that on hardware which is capable of 8GB/s we only
see a few GB/s when running through the DMU. To address this we need
to finish getting the code working with the kernel lock profiler and
look for some hot locks. Additionally, it would be interesting to run
the same tests on Solaris (once we have a ZVOL/ZPL) and compare the
performance. It's not at all clear to me Solaris currently does better.