summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorJakob Unterwurzacher2017-03-19 20:23:49 +0100
committerJakob Unterwurzacher2017-03-19 20:23:49 +0100
commitb78c3bd5168327a378d5a444aaafd712a4221933 (patch)
treefe839f62a62dc01a4bde887b4edbea01653d65b1
parentd2e9e8e2f8483cdaf3e15619ff957e65bbc9c67c (diff)
MANPAGE: document "-serialize_reads"
-rw-r--r--Documentation/MANPAGE.md18
1 files changed, 18 insertions, 0 deletions
diff --git a/Documentation/MANPAGE.md b/Documentation/MANPAGE.md
index 66956d6..7387424 100644
--- a/Documentation/MANPAGE.md
+++ b/Documentation/MANPAGE.md
@@ -181,6 +181,24 @@ Options:
**-ro**
: Mount the filesystem read-only
+**-serialize_reads**
+: The kernel usually submits multiple concurrent reads to service
+ userspace requests and kernel readahead. gocryptfs serves them
+ concurrently and in arbitrary order. On backing storage that performs
+ poorly for concurrent or out-of-order reads (like Amazon Cloud Drive),
+ this behavoir can cause very slow read speeds.
+
+ The `-serialize_reads`
+ option does two things: (1) reads will be submitted one-by-one (no
+ concurrency) and (2) gocryptfs tries to order the reads by file
+ offset order.
+
+ The ordering requires gocryptfs to wait a certain time before
+ submitting a read. The serialization introduces extra locking.
+ These factors will limit throughput to below 70MB/s.
+
+ For more details visit https://github.com/rfjakob/gocryptfs/issues/92 .
+
**-scryptn int**
: scrypt cost parameter logN. Setting this to a lower value speeds up
mounting but makes the password susceptible to brute-force attacks