summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorRichard W.M. Jones <rjones@redhat.com>2012-07-04 16:44:26 +0100
committerRichard W.M. Jones <rjones@redhat.com>2012-07-04 16:44:35 +0100
commit1b87f89b5c61f7b840663ddebaa58beb716bb1e9 (patch)
tree19f0072cf290e1e547cb23b7920b1259d1c7047d
parenta95fc923880018e5e533d69ba2bef702e0f6dd3c (diff)
downloadlibguestfs-1b87f89b5c61f7b840663ddebaa58beb716bb1e9.tar.gz
libguestfs-1b87f89b5c61f7b840663ddebaa58beb716bb1e9.tar.xz
libguestfs-1b87f89b5c61f7b840663ddebaa58beb716bb1e9.zip
todo: Investigate qemu caching modes.
-rw-r--r--TODO16
1 files changed, 16 insertions, 0 deletions
diff --git a/TODO b/TODO
index a89d52e7..247f98a1 100644
--- a/TODO
+++ b/TODO
@@ -498,3 +498,19 @@ should. There is a command line tool designed for scripts called
this approach works, it doesn't solve the MBR problem, so likely we'd
have to write a library for that (or perhaps go back to sfdisk but
using a very abstracted interface over sfdisk).
+
+qemu caching
+------------
+
+(Suggested by Paolo Bonzini and Kevin Wolf)
+
+Measure the effect of cache=none, cache=directsync,
+cache=writethrough, cache=writeback.
+
+It's doubtful that using cache=none is useful, since it disables the
+host cache making read-heavy workloads slower (they rely entirely on
+the smaller appliance kernel's cache). And in libguestfs we don't
+necessarily care about ongoing data integrity while writing, as long
+as data is reliably written out when g.sync or g.close return. Also
+in libguestfs we effectively control the whole stack, so we can ensure
+write barriers happen when we want.