1 ==================================================
2 memaslap - Load testing and benchmarking a server
3 ==================================================
8 |client_prefix|\aslap [options]
15 .. envvar:: MEMCACHED_SERVERS
20 :program:`memaslap` is a load generation and benchmark tool for memcached
21 servers. It generates configurable workload such as threads, concurrency,
22 connections, run time, overwrite, miss rate, key size, value size, get/set
23 proportion, expected throughput, and so on. Furthermore, it also tests data
24 verification, expire-time verification, UDP, binary protocol, facebook test,
25 replication test, multi-get and reconnection, etc.
27 Memaslap manages network connections like memcached with
28 libevent. Each thread of memaslap is bound with a CPU core, all
29 the threads don't communicate with each other, and there are several socket
30 connections in each thread. Each connection keeps key size distribution,
31 value size distribution, and command distribution by itself.
33 You can specify servers via the :option:`memaslap --servers` option or via the
34 environment variable :envvar:`MEMCACHED_SERVERS`.
39 Memslap is developed to for the following purposes:
41 Manages network connections with libevent asynchronously.
43 Set both TCP and UDP up to use non-blocking IO.
45 Improves parallelism: higher performance in multi-threads environments.
47 Improves time efficiency: faster processing speed.
49 Generates key and value more efficiently; key size distribution and value size distribution are configurable.
51 Supports get, multi-get, and set commands; command distribution is configurable.
53 Supports controllable miss rate and overwrite rate.
55 Supports data and expire-time verification.
57 Supports dumping statistic information periodically.
59 Supports thousands of TCP connections.
61 Supports binary protocol.
63 Supports facebook test (set with TCP and multi-get with UDP) and replication test.
68 Effective implementation of network.
69 ____________________________________
71 For memaslap, both TCP and UDP use non-blocking network IO. All
72 the network events are managed by libevent as memcached. The network module
73 of memaslap is similar to memcached. Libevent can ensure
74 memaslap can handle network very efficiently.
76 Effective implementation of multi-threads and concurrency
77 _________________________________________________________
79 Memslap has the similar implementation of multi-threads to
80 memcached. Memslap creates one or more self-governed threads;
81 each thread is bound with one CPU core if the system tests setting CPU
84 In addition, each thread has a libevent to manage the events of the network;
85 each thread has one or more self-governed concurrency; and each
86 concurrency has one or more socket connections. All the concurrent tasks don't
87 communicate with each other even though they are in the same thread.
89 Memslap can create thousands of socket connections, and each
90 concurrency has tens of socket connections. Each concurrency randomly or
91 sequentially selects one socket connection from its socket connection pool
92 to run, so memaslap can ensure each concurrency handles one
93 socket connection at any given time. Users can specify the number of
94 concurrency and socket connections of each concurrency according to their
97 Effective implementation of generating key and value
98 ____________________________________________________
100 In order to improve time efficiency and space efficiency,
101 memaslap creates a random characters table with 10M characters. All the
102 suffixes of keys and values are generated from this random characters table.
104 Memslap uses the offset in the character table and the length
105 of the string to identify a string. It can save much memory.
106 Each key contains two parts, a prefix and a suffix. The prefix is an
107 uint64_t, 8 bytes. In order to verify the data set before,
108 memaslap need to ensure each key is unique, so it uses the prefix to identify
109 a key. The prefix cannot include illegal characters, such as '\r', '\n',
110 '\0' and ' '. And memaslap has an algorithm to ensure that.
112 Memslap doesn't generate all the objects (key-value pairs) at
113 the beginning. It only generates enough objects to fill the task window
114 (default 10K objects) of each concurrency. Each object has the following
115 basic information, key prefix, key suffix offset in the character table, key
116 length, value offset in the character table, and value length.
118 In the work process, each concurrency sequentially or randomly selects an
119 object from the window to do set operation or get operation. At the same
120 time, each concurrency kicks objects out of its window and adds new object
123 Simple but useful task scheduling
124 _________________________________
126 Memslap uses libevent to schedule all concurrent tasks of
127 threads, and each concurrency schedules tasks based on the local task
128 window. Memslap assumes that if each concurrency keeps the same
129 key distribution, value distribution and commands distribution, from
130 outside, memaslap keeps all the distribution as a whole.
131 Each task window includes a lot of objects, each object stores its basic
132 information, such as key, value, expire time, and so on. At any time, all
133 the objects in the window keep the same and fixed key and value
134 distribution. If an object is overwritten, the value of the object will be
135 updated. Memslap verifies the data or expire-time according to
136 the object information stored in the task window.
138 Libevent selects which concurrency to handle based on a specific network
139 event. Then the concurrency selects which command (get or set) to operate
140 based on the command distribution. If it needs to kick out an old object and
141 add a new object, in order to keep the same key and value distribution, the
142 new object must have the same key length and value length.
144 If memcached server has two cache layers (memory and SSD), running
145 memaslap with different window sizes can get different cache
146 miss rates. If memaslap adds enough objects into the windows at
147 the beginning, and the cache of memcached cannot store all the objects
148 initialized, then memaslap will get some objects from the second
149 cache layer. It causes the first cache layer to miss. So the user can
150 specify the window size to get the expected miss rate of the first cache
153 Useful implementation of multi-servers , UDP, TCP, multi-get and binary protocol
154 ________________________________________________________________________________
156 Because each thread is self-governed, memaslap can assign
157 different threads to handle different memcached servers. This is just one of
158 the ways in which memaslap tests multiple servers. The only
159 limitation is that the number of servers cannot be greater than the number
160 of threads. The other way to test multiple servers is for replication
161 test. Each concurrency has one socket connection to each memcached server.
162 For the implementation, memaslap can set some objects to one
163 memcached server, and get these objects from the other servers.
165 By default, Memslap does single get. If the user specifies
166 multi-get option, memaslap will collect enough get commands and
167 pack and send the commands together.
169 Memslap tests both the ASCII protocol and binary protocol,
170 but it runs on the ASCII protocol by default.
171 Memslap by default runs on the TCP protocol, but it also
172 tests UDP. Because UDP is unreliable, dropped packages and out-of-order
173 packages may occur. Memslap creates a memory buffer to handle
174 these problems. Memslap tries to read all the response data of
175 one command from the server and reorders the response data. If some packages
176 get lost, the waiting timeout mechanism can ensure half-baked packages will
177 be discarded and the next command will be sent.
182 Below are some usage samples:
184 memaslap -s 127.0.0.1:11211 -S 5s
186 memaslap -s 127.0.0.1:11211 -t 2m -v 0.2 -e 0.05 -b
188 memaslap -s 127.0.0.1:11211 -F config -t 2m -w 40k -S 20s -o 0.2
190 memaslap -s 127.0.0.1:11211 -F config -t 2m -T 4 -c 128 -d 20 -P 40k
192 memaslap -s 127.0.0.1:11211 -F config -t 2m -d 50 -a -n 40
194 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m
196 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m -p 2
198 The user must specify one server at least to run memaslap. The
199 rest of the parameters have default values, as shown below:
201 Thread number = 1 Concurrency = 16
203 Run time = 600 seconds Configuration file = NULL
205 Key size = 64 Value size = 1024
207 Get/set = 9:1 Window size = 10k
209 Execute number = 0 Single get = true
211 Multi-get = false Number of sockets of each concurrency = 1
213 Reconnect = false Data verification = false
215 Expire-time verification = false ASCII protocol = true
217 Binary protocol = false Dumping statistic information periodically = false
219 Overwrite proportion = 0% UDP = false
221 TCP = true Limit throughput = false
223 Facebook test = false Replication test = false
225 Key size, value size and command distribution.
226 ______________________________________________
228 All the distributions are read from the configuration file specified by user
229 with "—cfg_cmd" option. If the user does not specify a configuration file,
230 memaslap will run with the default distribution (key size = 64,
231 value size = 1024, get/set = 9:1). For information on how to edit the
232 configuration file, refer to the "Configuration File" section.
234 The minimum key size is 16 bytes; the maximum key size is 250 bytes. The
235 precision of proportion is 0.001. The proportion of distribution will be
236 rounded to 3 decimal places.
238 The minimum value size is 1 bytes; the maximum value size is 1M bytes. The
239 precision of proportion is 0.001. The proportion of distribution will be
240 rounded to 3 decimal places.
241 Currently, memaslap only tests set and get commands. And it
242 testss 100% set and 100% get. For 100% get, it will preset some objects to
245 Multi-thread and concurrency
246 ____________________________
248 The high performance of memaslap benefits from the special
249 schedule of thread and concurrency. It's important to specify the proper
250 number of them. The default number of threads is 1; the default number of
251 concurrency is 16. The user can use "—threads" and "--concurrency" to
252 specify these variables.
254 If the system tests setting CPU affinity and the number of threads
255 specified by the user is greater than 1, memaslap will try to
256 bind each thread to a different CPU core. So if you want to get the best
257 performance memaslap, it is better to specify the number of
258 thread equal to the number of CPU cores. The number of threads specified by
259 the user can also be less or greater than the number of CPU cores. Because
260 of the limitation of implementation, the number of concurrencies could be
261 the multiple of the number of threads.
263 1. For 8 CPU cores system
267 --threads=2 --concurrency=128
269 --threads=8 --concurrency=128
271 --threads=8 --concurrency=256
273 --threads=12 --concurrency=144
275 2. For 16 CPU cores system
279 --threads=8 --concurrency=128
281 --threads=16 --concurrency=256
283 --threads=16 --concurrency=512
285 --threads=24 --concurrency=288
287 The memaslap performs very well, when
288 used to test the performance of memcached servers.
289 Most of the time, the bottleneck is the network or
290 the server. If for some reason the user wants to
291 limit the performance of memaslap, there
292 are two ways to do this:
294 Decrease the number of threads and concurrencies.
295 Use the option "--tps" that memaslap
296 provides to limit the throughput. This option allows
297 the user to get the expected throughput. For
298 example, assume that the maximum throughput is 50
299 kops/s for a specific configuration, you can specify
300 the throughput equal to or less than the maximum
301 throughput using "--tps" option.
306 Most of the time, the user does not need to specify the window size. The
307 default window size is 10k. For Schooner Memcached, the user can specify
308 different window sizes to get different cache miss rates based on the test
309 case. Memslap testss cache miss rate between 0% and 100%.
310 If you use this utility to test the performance of Schooner Memcached, you
311 can specify a proper window size to get the expected cache miss rate. The
312 formula for calculating window size is as follows:
314 Assume that the key size is 128 bytes, and the value size is 2048 bytes, and
317 1. Small cache cache_size=1M, 100% cache miss (all data get from SSD).
322 (1). cache miss rate 0%
326 (2). cache miss rate 5%
332 (1). cache miss rate 0%
342 The formula for calculating window size for cache miss rate 0%:
344 cache_size / concurrency / (key_size + value_size) \* 0.5
346 The formula for calculating window size for cache miss rate 5%:
348 cache_size / concurrency / (key_size + value_size) \* 0.7
353 Memslap testss both data verification and expire-time
354 verification. The user can use "--verify=" or "-v" to specify the proportion
355 of data verification. In theory, it testss 100% data verification. The
356 user can use "--exp_verify=" or "-e" to specify the proportion of
357 expire-time verification. In theory, it testss 100% expire-time
358 verification. Specify the "--verbose" options to get more detailed error
361 For example: --exp_verify=0.01 –verify=0.1 , it means that 1% of the objects
362 set with expire-time, 10% of the objects gotten will be verified. If the
363 objects are gotten, memaslap will verify the expire-time and
366 multi-servers and multi-config
367 _______________________________
369 Memslap testss multi-servers based on self-governed thread.
370 There is a limitation that the number of servers cannot be greater than the
371 number of threads. Memslap assigns one thread to handle one
372 server at least. The user can use the "--servers=" or "-s" option to specify
377 --servers=10.1.1.1:11211,10.1.1.2:11212,10.1.1.3:11213 --threads=6 --concurrency=36
379 The above command means that there are 6 threads, with each thread having 6
380 concurrencies and that threads 0 and 3 handle server 0 (10.1.1.1); threads 1
381 and 4 handle server 1 (10.1.1.2); and thread 2 and 5 handle server 2
384 All the threads and concurrencies in memaslap are self-governed.
386 So is memaslap. The user can start up several
387 memaslap instances. The user can run memaslap on different client
388 machines to communicate with the same memcached server at the same. It is
389 recommended that the user start different memaslap on different
390 machines using the same configuration.
392 Run with execute number mode or time mode
393 _________________________________________
395 The default memaslap runs with time mode. The default run time
396 is 10 minutes. If it times out, memaslap will exit. Do not
397 specify both execute number mode and time mode at the same time; just
402 --time=30s (It means the test will run 30 seconds.)
404 --execute_number=100000 (It means that after running 100000 commands, the test will exit.)
406 Dump statistic information periodically.
407 ________________________________________
409 The user can use "--stat_freq=" or "-S" to specify the frequency.
415 Memslap will dump the statistics of the commands (get and set) at the frequency of every 20
418 For more information on the format of dumping statistic information, refer to "Format of Output" section.
423 The user can use "--division=" or "-d" to specify multi-get keys count.
424 Memslap by default does single get with TCP. Memslap also testss data
425 verification and expire-time verification for multi-get.
427 Memslap testss multi-get with both TCP and UDP. Because of
428 the different implementation of the ASCII protocol and binary protocol,
429 there are some differences between the two. For the ASCII protocol,
430 memaslap sends one "multi-get" to the server once. For the
431 binary protocol, memaslap sends several single get commands
432 together as "multi-get" to the server.
437 Memslap testss both UDP and TCP. For TCP,
438 memaslap does not reconnect the memcached server if socket connections are
439 lost. If all the socket connections are lost or memcached server crashes,
440 memaslap will exit. If the user specifies the "--reconnect"
441 option when socket connections are lost, it will reconnect them.
443 User can use "--udp" to enable the UDP feature, but UDP comes with some
446 UDP cannot set data more than 1400 bytes.
448 UDP is not tested by the binary protocol because the binary protocol of
449 memcached does not tests that.
451 UDP doesn't tests reconnection.
456 Set data with TCP and multi-get with UDP. Specify the following options:
458 "--facebook --division=50"
460 If you want to create thousands of TCP connections, specify the
462 "--conn_sock=" option.
464 For example: --facebook --division=50 --conn_sock=200
466 The above command means that memaslap will do facebook test,
467 each concurrency has 200 socket TCP connections and one UDP socket.
469 Memslap sets objects with the TCP socket, and multi-gets 50
470 objects once with the UDP socket.
472 If you specify "--division=50", the key size must be less that 25 bytes
473 because the UDP packet size is 1400 bytes.
478 For replication test, the user must specify at least two memcached servers.
479 The user can use "—rep_write=" option to enable feature.
483 --servers=10.1.1.1:11211,10.1.1.2:11212 –rep_write=2
485 The above command means that there are 2 replication memcached servers,
486 memaslap will set objects to both server 0 and server 1, get
487 objects which are set to server 0 before from server 1, and also get objects
488 which are set to server 1 before from server 0. If server 0 crashes,
489 memaslap will only get objects from server 1. If server 0 comes
490 back to life again, memaslap will reconnect server 0. If both
491 server 0 and server 1 crash, memaslap will exit.
493 Supports thousands of TCP connections
494 _____________________________________
496 Start memaslap with "--conn_sock=" or "-n" to enable this
497 feature. Make sure that your system can tests opening thousands of files
498 and creating thousands of sockets. However, this feature does not tests
499 reconnection if sockets disconnect.
503 --threads=8 --concurrency=128 --conn_sock=128
505 The above command means that memaslap starts up 8 threads, each
506 thread has 16 concurrencies, each concurrency has 128 TCP socket
507 connections, and the total number of TCP socket connections is 128 \* 128 =
510 Supports binary protocol
511 ________________________
513 Start memaslap with "--binary" or "-B" options to enable this
514 feature. It testss all the above features except UDP, because the latest
515 memcached 1.3.3 does not implement binary UDP protocol.
521 Since memcached 1.3.3 doesn't implement binary UDP protocol,
522 memaslap does not tests UDP. In addition, memcached 1.3.3 does not tests
523 multi-get. If you specify "--division=50" option, it just sends 50 get
524 commands together as "multi-get" to the server.
529 This section describes the format of the configuration file. By default
530 when no configuration file is specified memaslap reads the default
531 one located at ~/.memaslap.cnf.
533 Below is a sample configuration file:
537 ---------------------------------------------------------------------------
538 #comments should start with '#'
540 #start_len end_len proportion
542 #key length range from start_len to end_len
543 #start_len must be equal to or greater than 16
544 #end_len must be equal to or less than 250
545 #start_len must be equal to or greater than end_len
546 #memaslap will generate keys according to the key range
547 #proportion: indicates keys generated from one range accounts for the total
550 #example1: key range 16~100 accounts for 80%
551 # key range 101~200 accounts for 10%
552 # key range 201~250 accounts for 10%
553 # total should be 1 (0.8+0.1+0.1 = 1)
559 #example2: all keys length are 128 bytes
565 #start_len end_len proportion
567 #value length range from start_len to end_len
568 #start_len must be equal to or greater than 1
569 #end_len must be equal to or less than 1M
570 #start_len must be equal to or greater than end_len
571 #memaslap will generate values according to the value range
572 #proportion: indicates values generated from one range accounts for the
573 total generated values
575 #example1: value range 1~1000 accounts for 80%
576 # value range 1001~10000 accounts for 10%
577 # value range 10001~100000 accounts for 10%
578 # total should be 1 (0.8+0.1+0.1 = 1)
584 #example2: all value length are 128 bytes
590 #cmd_type cmd_proportion
592 #currently memaslap only testss get and set command.
598 #example: set command accounts for 50%
599 # get command accounts for 50%
600 # total should be 1 (0.5+0.5 = 1)
612 At the beginning, memaslap displays some configuration information as follows:
614 servers : 127.0.0.1:11211
624 set proportion: set_prop=0.10
626 get proportion: get_prop=0.90
633 The servers used by memaslap.
637 The number of threads memaslap runs with.
641 The number of concurrencies memaslap runs with.
645 How long to run memaslap.
649 The task window size of each concurrency.
653 The proportion of set command.
657 The proportion of get command.
659 The output of dynamic statistics is something like this:
663 ---------------------------------------------------------------------------------------------------------------------------------
665 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
666 Avg(us) Std_dev Geo_dist
667 Period 5 345826 69165 65.3 0 27 2198 203
669 Global 20 1257935 62896 71.8 0 26 3791 224
673 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
674 Avg(us) Std_dev Geo_dist
675 Period 5 38425 7685 7.3 0 42 628 240
677 Global 20 139780 6989 8.0 0 37 3790 253
681 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
682 Avg(us) Std_dev Geo_dist
683 Period 5 384252 76850 72.5 0 27 2198 207
685 Global 20 1397720 69886 79.7 0 26 3791 227
687 ---------------------------------------------------------------------------------------------------------------------------------
694 Statistics information of get command
698 Statistics information of set command
702 Statistics information of both get and set command
706 Result within a period
718 Throughput, operations/second
726 How many objects can't be gotten
730 The minimum response time
734 The maximum response time
738 The average response time
742 Standard deviation of response time
746 Geometric distribution based on natural exponential function
748 At the end, memaslap will output something like this:
752 ---------------------------------------------------------------------------------------------------------------------------------
753 Get Statistics (1257956 events)
761 8: 484890 459823 12543 824
764 Set Statistics (139782 events)
772 8: 50784 65574 2064 167
775 Total Statistics (1397738 events)
783 8: 535674 525397 14607 991
793 written_bytes: 242516030
794 read_bytes: 1003702556
795 object_bytes: 152086080
800 Run time: 20.0s Ops: 1397754 TPS: 69817 Net_rate: 59.4M/s
801 ---------------------------------------------------------------------------------------------------------------------------------
808 Get statistics of response time
812 Set statistics of response time
816 Both get and set statistics of response time
820 The accumulated and minimum response time
824 The accumulated and maximum response time
828 The accumulated and average response time
832 Standard deviation of response time
836 Geometric distribution based on logarithm 2
840 Total get commands done
844 Total set commands done
848 How many objects can't be gotten from server
852 How many objects need to verify but can't get them
856 How many objects with insistent value
860 How many objects are expired but we get them
864 How many objects are unexpired but we can't get them
880 How many UDP packages are disorder
884 How many UDP packages are lost
888 How many times UDP time out happen
900 Throughput, operations/second
904 The average rate of network
910 List one or more servers to connect. Servers count must be less than
911 threads count. e.g.: --servers=localhost:1234,localhost:11211
914 Number of threads to startup, better equal to CPU numbers. Default 8.
917 Number of concurrency to simulate with load. Default 128.
920 Number of TCP socks per concurrency. Default 1.
922 -x, --execute_number=
923 Number of operations(get and set) to execute for the
924 given test. Default 1000000.
927 How long the test to run, suffix: s-seconds, m-minutes, h-hours,
928 d-days e.g.: --time=2h.
931 Load the configure file to get command,key and value distribution list.
934 Task window size of each concurrency, suffix: K, M e.g.: --win_size=10k.
938 Fixed length of value.
941 The proportion of date verification, e.g.: --verify=0.01
944 Number of keys to multi-get once. Default 1, means single get.
947 Frequency of dumping statistic information. suffix: s-seconds,
948 m-minutes, e.g.: --resp_freq=10s.
951 The proportion of objects with expire time, e.g.: --exp_verify=0.01.
952 Default no object with expire time
955 The proportion of objects need overwrite, e.g.: --overwrite=0.01.
956 Default never overwrite object.
959 Reconnect tests, when connection is closed it will be reconnected.
962 UDP tests, default memaslap uses TCP, TCP port and UDP port of
966 Whether it enables facebook test feature, set with TCP and multi-get with UDP.
969 Whether it enables binary protocol. Default with ASCII protocol.
972 Expected throughput, suffix: K, e.g.: --tps=10k.
975 The first nth servers can write data, e.g.: --rep_write=2.
978 Whether it outputs detailed information when verification fails.
981 Display this message and then exit.
984 Display the version of the application and then exit.
989 memaslap -s 127.0.0.1:11211 -S 5s
991 memaslap -s 127.0.0.1:11211 -t 2m -v 0.2 -e 0.05 -b
993 memaslap -s 127.0.0.1:11211 -F config -t 2m -w 40k -S 20s -o 0.2
995 memaslap -s 127.0.0.1:11211 -F config -t 2m -T 4 -c 128 -d 20 -P 40k
997 memaslap -s 127.0.0.1:11211 -F config -t 2m -d 50 -a -n 40
999 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m
1001 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m -p 2
1006 .. include:: common/note_program_prefix.rst
1007 .. include:: common/note_contrib_options.rst
1014 :manpage:`memcached(1)` :manpage:`libmemcached(3)`