From e879b80c681a6f93d1e01144b2757f8f898994c0 Mon Sep 17 00:00:00 2001 From: Richard Elling Date: Wed, 8 May 2019 18:09:24 -0700 Subject: [PATCH] fixed typos in zfs_per_txg_dirty_frees_percent and added method to observe txg commit interval --- ZFS-on-Linux-Module-Parameters.md | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/ZFS-on-Linux-Module-Parameters.md b/ZFS-on-Linux-Module-Parameters.md index f944be7..4830e2a 100644 --- a/ZFS-on-Linux-Module-Parameters.md +++ b/ZFS-on-Linux-Module-Parameters.md @@ -2716,9 +2716,10 @@ does not throttle other, perhaps more important, writes. A side effect of this throttle can impact `zfs receive` workloads that contain a large number of frees and the [ignore_hole_birth](#ignore_hole_birth) optimization is disabled. The symptom is that the receive workload causes an increase -in the frequency of txg commits when. Since txg commits also flush data from volatile -caches in HDDs to media, HDD performance can be negatively impacted. Also, since -the frees do not consume much bandwith over the pipe, the pipe can appear to stall. +in the frequency of txg commits. The frequency of txg commits is observable via the +`otime` column of `/proc/spl/kstat/zfs/POOLNAME/txgs`. Since txg commits also flush data +from volatile caches in HDDs to media, HDD performance can be negatively impacted. +Also, since the frees do not consume much bandwidth over the pipe, the pipe can appear to stall. Thus the overall progress of receives is slower than expected. A value of zero will disable this throttle.