From 7754325932885dea532da94fe08bd5c57feeb6a7 Mon Sep 17 00:00:00 2001 From: Eric Blake Date: Mon, 28 Oct 2019 14:31:03 +0100 Subject: [PATCH] 4300: Polish slide by changing 
 to .

---
 2019-kvm-forum/4300-block-status-impact.html | 65 ++++++++++++++++++++--------
 2019-kvm-forum/notes-04-fast-zero            | 17 ++++----
 2 files changed, 54 insertions(+), 28 deletions(-)

diff --git a/2019-kvm-forum/4300-block-status-impact.html b/2019-kvm-forum/4300-block-status-impact.html
index 948e973..7d8d764 100644
--- a/2019-kvm-forum/4300-block-status-impact.html
+++ b/2019-kvm-forum/4300-block-status-impact.html
@@ -5,23 +5,50 @@
 
Block status: does it change the status quo?

-  Filler text - need a graphic here showing multiple combinations:
-
-  
[WZ] = WRITE_ZEROES
-  
[BS] = BLOCK_STATUS
-
-
-                  O(1) server             O(n) server
-pre-patch         [WZ]                    [WZ...........]
-
-status sees zero  [BS]                    [BS]
-                  (no improvement)        (noticeably faster)
-
-status sees data  [BS][WZ]                [BS][WZ...........]
-		  (noticeably slower)     (marginally slower)
-
-status is slow    [BS............]        [BS............]
-                  (noticeably slower)     (noticeably slower)
-

-
+  Checking whether a block is already zero will eliminate the need to
+  write zeroes to that block, but at what cost?
+
+  

+  block status reports hole
+  block status reports data
+  write zeroes
+
+  

+  

 
 

+    
+      
+      
+      
+    
+    
+      
+      
+      
+    
+    
+      
+      
+      
+    
+    
+      
+      
+      
+    
+    
+      
+      
+      
+    
+  
O(1) serverO(n) server
no block status  
block status sees zero  
block status sees data 
+          
+         
block status is slow 
+          
+         

+
+  

+    If checking once per block is bad, can we instead check once per
+    image?
+  
Or why even check - what if we ensure the image starts
+    with all zeroes?
 

diff --git a/2019-kvm-forum/notes-04-fast-zero b/2019-kvm-forum/notes-04-fast-zero
index 19622d0..c818417 100644
--- a/2019-kvm-forum/notes-04-fast-zero
+++ b/2019-kvm-forum/notes-04-fast-zero
@@ -53,15 +53,14 @@ didn't really help, for a couple of reasons.  If writing zeroes is
 fast, checking the destination first is either a mere tradeoff in
 commands (BLOCK_STATUS replacing WRITE_ZEROES when the destination is
 already zero) or a pessimization (BLOCK_STATUS still has to be
-followed by WRITE_ZEROES).  And if writing zeroes is slow, we have a
-speedup holes when BLOCK_STATUS itself is fast on pre-existing
-destination holes, but we encountered situations such as tmpfs that
-has a linear rather than constant-time lseek(SEEK_HOLE)
-implementation, where we ended up with quadratic behavior all due to
-BLOCK_STATUS calls.  Thus, for now, qemu-img convert does not use
-BLOCK_STATUS, and as mentioned earlier, I used the noextents filter in
-my test case to ensure BLOCK_STATUS is not interfering with timing
-results.
+followed by WRITE_ZEROES).  Conversely, if writing zeroes is slow, the
+penalty of the extra check when the destination is not a hole could be
+in the noise, but whether or not we have a definite speedup for
+avoiding WRITE_ZEROES depends on whether BLOCK_STATUS itself is fast -
+yet we encountered situations such as tmpfs that has a linear rather
+than constant-time lseek(SEEK_HOLE) implementation, where we ended up
+with quadratic behavior all due to BLOCK_STATUS calls.  Thus, for now,
+qemu-img convert does not use BLOCK_STATUS.
 
 * Heading: Pre-zeroing: a tale of two servers
 - 4400- .term
-- 
1.8.3.1