1 XBZRLE (Xor Based Zero Run Length Encoding)
2 ===========================================
4 Using XBZRLE (Xor Based Zero Run Length Encoding) allows for the reduction
5 of VM downtime and the total live-migration time of Virtual machines.
6 It is particularly useful for virtual machines running memory write intensive
7 workloads that are typical of large enterprise applications such as SAP ERP
8 Systems, and generally speaking for any application that uses a sparse memory
11 Instead of sending the changed guest memory page this solution will send a
12 compressed version of the updates, thus reducing the amount of data sent during
14 In order to be able to calculate the update, the previous memory pages need to
15 be stored on the source. Those pages are stored in a dedicated cache
16 (hash table) and are accessed by their address.
17 The larger the cache size the better the chances are that the page has already
18 been stored in the cache.
19 A small cache size will result in high cache miss rate.
20 Cache size can be changed before and during migration.
25 The compression format performs a XOR between the previous and current content
26 of the page, where zero represents an unchanged value.
27 The page data delta is represented by zero and non zero runs.
28 A zero run is represented by its length (in bytes).
29 A non zero run is represented by its length (in bytes) and the new data.
30 The run length is encoded using ULEB128 (http://en.wikipedia.org/wiki/LEB128)
32 There can be more than one valid encoding, the sender may send a longer encoding
33 for the benefit of reducing computation cost.
40 nzrun = length byte...
42 length = uleb128 encoded integer
44 On the sender side XBZRLE is used as a compact delta encoding of page updates,
45 retrieving the old page content from the cache (default size of 64MB). The
46 receiving side uses the existing page's content and XBZRLE to decode the new
49 This work was originally based on research results published
50 VEE 2011: Evaluation of Delta Compression Techniques for Efficient Live
51 Migration of Large Virtual Machines by Benoit, Svard, Tordsson and Elmroth.
52 Additionally the delta encoder XBRLE was improved further using the XBZRLE
55 XBZRLE has a sustained bandwidth of 2-2.5 GB/s for typical workloads making it
56 ideal for in-line, real-time encoding such as is needed for live-migration.
61 05 06 07 08 09 0a 0b 0c 0d 0e 0f 10 11 12 13 68 00 00 6b 00 6d
66 01 02 03 04 05 06 07 08 09 0a 0b 0c 0d 0e 0f 68 00 00 67 00 69
72 e9 07 0f 01 02 03 04 05 06 07 08 09 0a 0b 0c 0d 0e 0f 03 01 67 01 01 69
76 Keeping the hot pages in the cache is effective for decreasing cache
77 misses. XBZRLE uses a counter as the age of each page. The counter will
78 increase after each ram dirty bitmap sync. When a cache conflict is
79 detected, XBZRLE will only evict pages in the cache that are older than
83 ======================
84 1. Verify the destination QEMU version is able to decode the new format.
85 {qemu} info migrate_capabilities
86 {qemu} xbzrle: off , ...
88 2. Activate xbzrle on both source and destination:
89 {qemu} migrate_set_capability xbzrle on
91 3. Set the XBZRLE cache size - the cache size is in MBytes and should be a
92 power of 2. The cache default value is 64MBytes. (on source only)
93 {qemu} migrate_set_cache_size 256m
95 Commit 73af8dd8d7 "migration: Make xbzrle_cache_size a migration parameter"
96 (v2.11.0) deprecated migrate-set-cache-size, therefore, the new parameter
98 {qemu} migrate_set_parameter xbzrle-cache-size 256m
100 4. Start outgoing migration
101 {qemu} migrate -d tcp:destination.host:4444
103 capabilities: xbzrle: on
104 Migration status: active
105 transferred ram: A kbytes
106 remaining ram: B kbytes
108 total time: D milliseconds
111 normal bytes: G kbytes
113 xbzrle transferred: I kbytes
114 xbzrle pages: J pages
118 xbzrle cache-miss: the number of cache misses to date - high cache-miss rate
119 indicates that the cache size is set too low.
120 xbzrle overflow: the number of overflows in the decoding which where the delta
121 could not be compressed. This can happen if the changes in the pages are too
122 large or there are many short changes; for example, changing every second byte
125 Testing: Testing indicated that live migration with XBZRLE was completed in 110
126 seconds, whereas without it would not be able to complete.
128 A simple synthetic memory r/w load generator:
129 .. include <stdlib.h>
133 .. char *buf = (char *) calloc(4096, 4096);
136 .. for (i = 0; i < 4096 * 4; i++) {
137 .. buf[i * 4096 / 4]++;