6 memslap - Load testing and benchmarking tool for memcached
26 \ **memslap**\ is a load generation and benchmark tool for memcached(1)
27 servers. It generates configurable workload such as threads, concurrencies, connections,
28 run time, overwrite, miss rate, key size, value size, get/set proportion,
29 expected throughput, and so on. Furthermore, it also supports data
30 verification, expire-time verification, UDP, binary protocol, facebook test,
31 replication test, multi-get and reconnection, etc.
33 Memslap manages network connections like memcached with
34 libevent. Each thread of memslap is bound with a CPU core, all
35 the threads don't communicate with each other, and there are several socket
36 connections in each thread. Each connection keeps key size distribution,
37 value size distribution, and command distribution by itself.
39 You can specify servers via the \ **--servers**\ option or via the
40 environment variable \ ``MEMCACHED_SERVERS``\ .
48 Memslap is developed to for the following purposes:
51 Manages network connections with libevent asynchronously.
55 Set both TCP and UDP up to use non-blocking IO.
59 Improves parallelism: higher performance in multi-threads environments.
63 Improves time efficiency: faster processing speed.
67 Generates key and value more efficiently; key size distribution and value size distribution are configurable.
71 Supports get, multi-get, and set commands; command distribution is configurable.
75 Supports controllable miss rate and overwrite rate.
79 Supports data and expire-time verification.
83 Supports dumping statistic information periodically.
87 Supports thousands of TCP connections.
91 Supports binary protocol.
95 Supports facebook test (set with TCP and multi-get with UDP) and replication test.
105 Effective implementation of network.
106 ====================================
109 For memslap, both TCP and UDP use non-blocking network IO. All
110 the network events are managed by libevent as memcached. The network module
111 of memslap is similar to memcached. Libevent can ensure
112 memslap can handle network very efficiently.
115 Effective implementation of multi-threads and concurrency
116 =========================================================
119 Memslap has the similar implementation of multi-threads to
120 memcached. Memslap creates one or more self-governed threads;
121 each thread is bound with one CPU core if the system supports setting CPU
124 In addition, each thread has a libevent to manage the events of the network;
125 each thread has one or more self-governed concurrencies; and each
126 concurrency has one or more socket connections. All the concurrencies don’t
127 communicate with each other even though they are in the same thread.
129 Memslap can create thousands of socket connections, and each
130 concurrency has tens of socket connections. Each concurrency randomly or
131 sequentially selects one socket connection from its socket connection pool
132 to run, so memslap can ensure each concurrency handles one
133 socket connection at any given time. Users can specify the number of
134 concurrency and socket connections of each concurrency according to their
138 Effective implementation of generating key and value
139 ====================================================
142 In order to improve time efficiency and space efficiency,
143 memslap creates a random characters table with 10M characters. All the
144 suffixes of keys and values are generated from this random characters table.
146 Memslap uses the offset in the character table and the length
147 of the string to identify a string. It can save much memory.
148 Each key contains two parts, a prefix and a suffix. The prefix is an
149 uint64_t, 8 bytes. In order to verify the data set before,
150 memslap need to ensure each key is unique, so it uses the prefix to identify
151 a key. The prefix cannot include illegal characters, such as ‘\r’, ‘\n’,
152 ‘\0’ and ‘ ‘. And memslap has an algorithm to ensure that.
154 Memslap doesn’t generate all the objects (key-value pairs) at
155 the beginning. It only generates enough objects to fill the task window
156 (default 10K objects) of each concurrency. Each object has the following
157 basic information, key prefix, key suffix offset in the character table, key
158 length, value offset in the character table, and value length.
160 In the work process, each concurrency sequentially or randomly selects an
161 object from the window to do set operation or get operation. At the same
162 time, each concurrency kicks objects out of its window and adds new object
166 Simple but useful task scheduling
167 =================================
170 Memslap uses libevent to schedule all the concurrencies of
171 threads, and each concurrency schedules tasks based on the local task
172 window. Memslap assumes that if each concurrency keeps the same
173 key distribution, value distribution and commands distribution, from
174 outside, memslap keeps all the distribution as a whole.
175 Each task window includes a lot of objects, each object stores its basic
176 information, such as key, value, expire time, and so on. At any time, all
177 the objects in the window keep the same and fixed key and value
178 distribution. If an object is overwritten, the value of the object will be
179 updated. Memslap verifies the data or expire-time according to
180 the object information stored in the task window.
182 Libevent selects which concurrency to handle based on a specific network
183 event. Then the concurrency selects which command (get or set) to operate
184 based on the command distribution. If it needs to kick out an old object and
185 add a new object, in order to keep the same key and value distribution, the
186 new object must have the same key length and value length.
188 If memcached server has two cache layers (memory and SSD), running
189 memslap with different window sizes can get different cache
190 miss rates. If memslap adds enough objects into the windows at
191 the beginning, and the cache of memcached cannot store all the objects
192 initialized, then memslap will get some objects from the second
193 cache layer. It causes the first cache layer to miss. So the user can
194 specify the window size to get the expected miss rate of the first cache
198 Useful implementation of multi-servers , UDP, TCP, multi-get and binary protocol
199 ================================================================================
202 Because each thread is self-governed, memslap can assign
203 different threads to handle different memcached servers. This is just one of
204 the ways in which memslap supports multiple servers. The only
205 limitation is that the number of servers cannot be greater than the number
206 of threads. The other way to support multiple servers is for replication
207 test. Each concurrency has one socket connection to each memcached server.
208 For the implementation, memslap can set some objects to one
209 memcached server, and get these objects from the other servers.
211 By default, Memslap does single get. If the user specifies
212 multi-get option, memslap will collect enough get commands and
213 pack and send the commands together.
215 Memslap supports both the ASCII protocol and binary protocol,
216 but it runs on the ASCII protocol by default.
217 Memslap by default runs on the TCP protocol, but it also
218 supports UDP. Because UDP is unreliable, dropped packages and out-of-order
219 packages may occur. Memslap creates a memory buffer to handle
220 these problems. Memslap tries to read all the response data of
221 one command from the server and reorders the response data. If some packages
222 get lost, the waiting timeout mechanism can ensure half-baked packages will
223 be discarded and the next command will be sent.
232 Below are some usage samples:
235 memslap -s 127.0.0.1:11211 -S 5s
239 memslap -s 127.0.0.1:11211 -t 2m -v 0.2 -e 0.05 -b
243 memslap -s 127.0.0.1:11211 -F config -t 2m -w 40k -S 20s -o 0.2
247 memslap -s 127.0.0.1:11211 -F config -t 2m -T 4 -c 128 -d 20 -P 40k
251 memslap -s 127.0.0.1:11211 -F config -t 2m -d 50 -a -n 40
255 memslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m
259 memslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m -p 2
263 The user must specify one server at least to run memslap. The
264 rest of the parameters have default values, as shown below:
266 Thread number = 1 Concurrency = 16
268 Run time = 600 seconds Configuration file = NULL
270 Key size = 64 Value size = 1024
272 Get/set = 9:1 Window size = 10k
274 Execute number = 0 Single get = true
276 Multi-get = false Number of sockets of each concurrency = 1
278 Reconnect = false Data verification = false
280 Expire-time verification = false ASCII protocol = true
282 Binary protocol = false Dumping statistic information
286 Overwrite proportion = 0% UDP = false
288 TCP = true Limit throughput = false
290 Facebook test = false Replication test = false
292 Key size, value size and command distribution.
293 ==============================================
296 All the distributions are read from the configuration file specified by user
297 with “—cfg_cmd” option. If the user does not specify a configuration file,
298 memslap will run with the default distribution (key size = 64,
299 value size = 1024, get/set = 9:1). For information on how to edit the
300 configuration file, refer to the “Configuration File” section.
302 The minimum key size is 16 bytes; the maximum key size is 250 bytes. The
303 precision of proportion is 0.001. The proportion of distribution will be
304 rounded to 3 decimal places.
306 The minimum value size is 1 bytes; the maximum value size is 1M bytes. The
307 precision of proportion is 0.001. The proportion of distribution will be
308 rounded to 3 decimal places.
309 Currently, memslap only supports set and get commands. And it
310 supports 100% set and 100% get. For 100% get, it will preset some objects to
314 Multi-thread and concurrency
315 ============================
318 The high performance of memslap benefits from the special
319 schedule of thread and concurrency. It’s important to specify the proper
320 number of them. The default number of threads is 1; the default number of
321 concurrency is 16. The user can use “—threads” and “--concurrency” to
322 specify these variables.
324 If the system supports setting CPU affinity and the number of threads
325 specified by the user is greater than 1, memslap will try to
326 bind each thread to a different CPU core. So if you want to get the best
327 performance memslap, it is better to specify the number of
328 thread equal to the number of CPU cores. The number of threads specified by
329 the user can also be less or greater than the number of CPU cores. Because
330 of the limitation of implementation, the number of concurrencies could be
331 the multiple of the number of threads.
333 1. For 8 CPU cores system
337 --threads=2 --concurrency=128
339 --threads=8 --concurrency=128
341 --threads=8 --concurrency=256
343 --threads=12 --concurrency=144
345 2. For 16 CPU cores system
349 --threads=8 --concurrency=128
351 --threads=16 --concurrency=256
353 --threads=16 --concurrency=512
355 --threads=24 --concurrency=288
357 The memslap performs very well, when
358 used to test the performance of memcached servers.
359 Most of the time, the bottleneck is the network or
360 the server. If for some reason the user wants to
361 limit the performance of memslap, there
362 are two ways to do this:
364 Decrease the number of threads and concurrencies.
365 Use the option “--tps” that memslap
366 provides to limit the throughput. This option allows
367 the user to get the expected throughput. For
368 example, assume that the maximum throughput is 50
369 kops/s for a specific configuration, you can specify
370 the throughput equal to or less than the maximum
371 throughput using “--tps” option.
378 Most of the time, the user does not need to specify the window size. The
379 default window size is 10k. For Schooner Memcached, the user can specify
380 different window sizes to get different cache miss rates based on the test
381 case. Memslap supports cache miss rate between 0% and 100%.
382 If you use this utility to test the performance of Schooner Memcached, you
383 can specify a proper window size to get the expected cache miss rate. The
384 formula for calculating window size is as follows:
386 Assume that the key size is 128 bytes, and the value size is 2048 bytes, and
389 1. Small cache cache_size=1M, 100% cache miss (all data get from SSD).
394 (1). cache miss rate 0%
398 (2). cache miss rate 5%
404 (1). cache miss rate 0%
414 The formula for calculating window size for cache miss rate 0%:
416 cache_size / concurrency / (key_size + value_size) \* 0.5
418 The formula for calculating window size for cache miss rate 5%:
420 cache_size / concurrency / (key_size + value_size) \* 0.7
427 Memslap supports both data verification and expire-time
428 verification. The user can use "--verify=" or "-v" to specify the proportion
429 of data verification. In theory, it supports 100% data verification. The
430 user can use "--exp_verify=" or "-e" to specify the proportion of
431 expire-time verification. In theory, it supports 100% expire-time
432 verification. Specify the "--verbose" options to get more detailed error
435 For example: --exp_verify=0.01 –verify=0.1 , it means that 1% of the objects
436 set with expire-time, 10% of the objects gotten will be verified. If the
437 objects are gotten, memslap will verify the expire-time and
441 multi-servers and multi-clients
442 ===============================
445 Memslap supports multi-servers based on self-governed thread.
446 There is a limitation that the number of servers cannot be greater than the
447 number of threads. Memslap assigns one thread to handle one
448 server at least. The user can use the "--servers=" or "-s" option to specify
453 --servers=10.1.1.1:11211,10.1.1.2:11212,10.1.1.3:11213 --threads=6 --concurrency=36
455 The above command means that there are 6 threads, with each thread having 6
456 concurrencies and that threads 0 and 3 handle server 0 (10.1.1.1); threads 1
457 and 4 handle server 1 (10.1.1.2); and thread 2 and 5 handle server 2
460 All the threads and concurrencies in memslap are self-governed.
462 So is memslap. The user can start up several
463 memslap instances. The user can run memslap on different client
464 machines to communicate with the same memcached server at the same. It is
465 recommended that the user start different memslap on different
466 machines using the same configuration.
469 Run with execute number mode or time mode
470 =========================================
473 The default memslap runs with time mode. The default run time
474 is 10 minutes. If it times out, memslap will exit. Do not
475 specify both execute number mode and time mode at the same time; just
480 --time=30s (It means the test will run 30 seconds.)
482 --execute_number=100000 (It means that after running 100000 commands, the test will exit.)
485 Dump statistic information periodically.
486 ========================================
489 The user can use "--stat_freq=" or "-S" to specify the frequency.
495 Memslap will dump the statistics of the commands (get and set) at the frequency of every 20
498 For more information on the format of dumping statistic information, refer to “Format of Output” section.
505 The user can use "--division=" or "-d" to specify multi-get keys count.
506 Memslap by default does single get with TCP. Memslap also supports data
507 verification and expire-time verification for multi-get.
509 Memslap supports multi-get with both TCP and UDP. Because of
510 the different implementation of the ASCII protocol and binary protocol,
511 there are some differences between the two. For the ASCII protocol,
512 memslap sends one “multi-get” to the server once. For the
513 binary protocol, memslap sends several single get commands
514 together as “multi-get” to the server.
521 Memslap supports both UDP and TCP. For TCP,
522 memslap does not reconnect the memcached server if socket connections are
523 lost. If all the socket connections are lost or memcached server crashes,
524 memslap will exit. If the user specifies the “--reconnect”
525 option when socket connections are lost, it will reconnect them.
527 User can use “--udp” to enable the UDP feature, but UDP comes with some
530 UDP cannot set data more than 1400 bytes.
532 UDP is not supported by the binary protocol because the binary protocol of
533 memcached does not support that.
535 UDP doesn’t support reconnection.
542 Set data with TCP and multi-get with UDP. Specify the following options:
544 "--facebook --division=50"
546 If you want to create thousands of TCP connections, specify the
548 "--conn_sock=" option.
550 For example: --facebook --division=50 --conn_sock=200
552 The above command means that memslap will do facebook test,
553 each concurrency has 200 socket TCP connections and one UDP socket.
555 Memslap sets objects with the TCP socket, and multi-gets 50
556 objects once with the UDP socket.
558 If you specify "--division=50", the key size must be less that 25 bytes
559 because the UDP packet size is 1400 bytes.
566 For replication test, the user must specify at least two memcached servers.
567 The user can use “—rep_write=” option to enable feature.
571 --servers=10.1.1.1:11211,10.1.1.2:11212 –rep_write=2
573 The above command means that there are 2 replication memcached servers,
574 memslap will set objects to both server 0 and server 1, get
575 objects which are set to server 0 before from server 1, and also get objects
576 which are set to server 1 before from server 0. If server 0 crashes,
577 memslap will only get objects from server 1. If server 0 comes
578 back to life again, memslap will reconnect server 0. If both
579 server 0 and server 1 crash, memslap will exit.
582 Supports thousands of TCP connections
583 =====================================
586 Start memslap with "--conn_sock=" or "-n" to enable this
587 feature. Make sure that your system can support opening thousands of files
588 and creating thousands of sockets. However, this feature does not support
589 reconnection if sockets disconnect.
593 --threads=8 --concurrency=128 --conn_sock=128
595 The above command means that memslap starts up 8 threads, each
596 thread has 16 concurrencies, each concurrency has 128 TCP socket
597 connections, and the total number of TCP socket connections is 128 \* 128 =
601 Supports binary protocol
602 ========================
605 Start memslap with "--binary" or "-B" options to enable this
606 feature. It supports all the above features except UDP, because the latest
607 memcached 1.3.3 does not implement binary UDP protocol.
613 Since memcached 1.3.3 doesn't implement binary UDP protocol,
614 memslap does not support UDP. In addition, memcached 1.3.3 does not support
615 multi-get. If you specify "--division=50" option, it just sends 50 get
616 commands together as “mulit-get” to the server.
625 This section describes the format of the configuration file. By default
626 when no configuration file is specified memslap reads the default
627 one located at ~/.memslap.cnf.
629 Below is a sample configuration file:
634 ***************************************************************************
635 #comments should start with '#'
637 #start_len end_len proportion
639 #key length range from start_len to end_len
640 #start_len must be equal to or greater than 16
641 #end_len must be equal to or less than 250
642 #start_len must be equal to or greater than end_len
643 #memslap will generate keys according to the key range
644 #proportion: indicates keys generated from one range accounts for the total
647 #example1: key range 16~100 accounts for 80%
648 # key range 101~200 accounts for 10%
649 # key range 201~250 accounts for 10%
650 # total should be 1 (0.8+0.1+0.1 = 1)
656 #example2: all keys length are 128 bytes
662 #start_len end_len proportion
664 #value length range from start_len to end_len
665 #start_len must be equal to or greater than 1
666 #end_len must be equal to or less than 1M
667 #start_len must be equal to or greater than end_len
668 #memslap will generate values according to the value range
669 #proportion: indicates values generated from one range accounts for the
670 total generated values
672 #example1: value range 1~1000 accounts for 80%
673 # value range 1001~10000 accounts for 10%
674 # value range 10001~100000 accounts for 10%
675 # total should be 1 (0.8+0.1+0.1 = 1)
681 #example2: all value length are 128 bytes
687 #cmd_type cmd_proportion
689 #currently memslap only supports get and set command.
695 #example: set command accounts for 50%
696 # get command accounts for 50%
697 # total should be 1 (0.5+0.5 = 1)
713 At the beginning, memslap displays some configuration information as follows:
716 servers : 127.0.0.1:11211
736 set proportion: set_prop=0.10
740 get proportion: get_prop=0.90
751 The servers used by memslap.
757 The number of threads memslap runs with.
763 The number of concurrencies memslap runs with.
769 How long to run memslap.
775 The task window size of each concurrency.
781 The proportion of set command.
787 The proportion of get command.
791 The output of dynamic statistics is something like this:
796 ---------------------------------------------------------------------------------------------------------------------------------
798 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
799 Avg(us) Std_dev Geo_dist
800 Period 5 345826 69165 65.3 0 27 2198 203
802 Global 20 1257935 62896 71.8 0 26 3791 224
807 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
808 Avg(us) Std_dev Geo_dist
809 Period 5 38425 7685 7.3 0 42 628 240
811 Global 20 139780 6989 8.0 0 37 3790 253
816 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
817 Avg(us) Std_dev Geo_dist
818 Period 5 384252 76850 72.5 0 27 2198 207
820 Global 20 1397720 69886 79.7 0 26 3791 227
822 ---------------------------------------------------------------------------------------------------------------------------------
833 Statistics information of get command
839 Statistics information of set command
845 Statistics information of both get and set command
851 Result within a period
869 Throughput, operations/second
881 How many objects can’t be gotten
887 The minimum response time
893 The maximum response time
899 The average response time
905 Standard deviation of response time
911 Geometric distribution based on natural exponential function
915 At the end, memslap will output something like this:
920 ---------------------------------------------------------------------------------------------------------------------------------
921 Get Statistics (1257956 events)
929 8: 484890 459823 12543 824
932 Set Statistics (139782 events)
940 8: 50784 65574 2064 167
943 Total Statistics (1397738 events)
951 8: 535674 525397 14607 991
961 written_bytes: 242516030
962 read_bytes: 1003702556
963 object_bytes: 152086080
968 Run time: 20.0s Ops: 1397754 TPS: 69817 Net_rate: 59.4M/s
969 ---------------------------------------------------------------------------------------------------------------------------------
980 Get statistics of response time
986 Set statistics of response time
992 Both get and set statistics of response time
998 The accumulated and minimum response time
1004 The accumulated and maximum response time
1010 The accumulated and average response time
1016 Standard deviation of response time
1022 Geometric distribution based on logarithm 2
1028 Total get commands done
1034 Total set commands done
1040 How many objects can’t be gotten from server
1046 How many objects need to verify but can’t get them
1052 How many objects with insistent value
1058 How many objects are expired but we get them
1064 How many objects are unexpired but we can’t get them
1088 How many UDP packages are disorder
1094 How many UDP packages are lost
1100 How many times UDP time out happen
1118 Throughput, operations/second
1124 The average rate of network
1136 List one or more servers to connect. Servers count must be less than
1137 threads count. e.g.: --servers=localhost:1234,localhost:11211
1140 Number of threads to startup, better equal to CPU numbers. Default 8.
1143 Number of concurrency to simulate with load. Default 128.
1146 Number of TCP socks per concurrency. Default 1.
1148 -x, --execute_number=
1149 Number of operations(get and set) to execute for the
1150 given test. Default 1000000.
1153 How long the test to run, suffix: s-seconds, m-minutes, h-hours,
1154 d-days e.g.: --time=2h.
1157 Load the configure file to get command,key and value distribution list.
1160 Task window size of each concurrency, suffix: K, M e.g.: --win_size=10k.
1164 Fixed length of value.
1167 The proportion of date verification, e.g.: --verify=0.01
1170 Number of keys to multi-get once. Default 1, means single get.
1173 Frequency of dumping statistic information. suffix: s-seconds,
1174 m-minutes, e.g.: --resp_freq=10s.
1177 The proportion of objects with expire time, e.g.: --exp_verify=0.01.
1178 Default no object with expire time
1181 The proportion of objects need overwrite, e.g.: --overwrite=0.01.
1182 Default never overwrite object.
1185 Reconnect support, when connection is closed it will be reconnected.
1188 UDP support, default memslap uses TCP, TCP port and UDP port of
1189 server must be same.
1192 Whether it enables facebook test feature, set with TCP and multi-get with UDP.
1195 Whether it enables binary protocol. Default with ASCII protocol.
1198 Expected throughput, suffix: K, e.g.: --tps=10k.
1201 The first nth servers can write data, e.g.: --rep_write=2.
1204 Whether it outputs detailed information when verification fails.
1207 Display this message and then exit.
1210 Display the version of the application and then exit.
1218 memslap -s 127.0.0.1:11211 -S 5s
1220 memslap -s 127.0.0.1:11211 -t 2m -v 0.2 -e 0.05 -b
1222 memslap -s 127.0.0.1:11211 -F config -t 2m -w 40k -S 20s -o 0.2
1224 memslap -s 127.0.0.1:11211 -F config -t 2m -T 4 -c 128 -d 20 -P 40k
1226 memslap -s 127.0.0.1:11211 -F config -t 2m -d 50 -a -n 40
1228 memslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m
1230 memslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m -p 2
1238 To find out more information please check:
1239 `http://launchpad.org/libmemcached <http://launchpad.org/libmemcached>`_
1247 Mingqiang Zhuang <mingqiangzhuang@hengtiansoft.com> (Schooner Technolgy)
1248 Brian Aker, <brian@tangent.org>
1256 memcached(1) libmemcached(3)