9 memslap - Load testing and benchmarking tool for memcached
29 \ **memslap**\ is a load generation and benchmark tool for memcached(1)
30 servers. It generates configurable workload such as threads, concurrencies, connections,
31 run time, overwrite, miss rate, key size, value size, get/set proportion,
32 expected throughput, and so on. Furthermore, it also supports data
33 verification, expire-time verification, UDP, binary protocol, facebook test,
34 replication test, multi-get and reconnection, etc.
36 Memslap manages network connections like memcached with
37 libevent. Each thread of memslap is bound with a CPU core, all
38 the threads don't communicate with each other, and there are several socket
39 connections in each thread. Each connection keeps key size distribution,
40 value size distribution, and command distribution by itself.
42 You can specify servers via the \ **--servers**\ option or via the
43 environment variable \ ``MEMCACHED_SERVERS``\ .
51 Memslap is developed to for the following purposes:
54 Manages network connections with libevent asynchronously.
58 Set both TCP and UDP up to use non-blocking IO.
62 Improves parallelism: higher performance in multi-threads environments.
66 Improves time efficiency: faster processing speed.
70 Generates key and value more efficiently; key size distribution and value size distribution are configurable.
74 Supports get, multi-get, and set commands; command distribution is configurable.
78 Supports controllable miss rate and overwrite rate.
82 Supports data and expire-time verification.
86 Supports dumping statistic information periodically.
90 Supports thousands of TCP connections.
94 Supports binary protocol.
98 Supports facebook test (set with TCP and multi-get with UDP) and replication test.
108 Effective implementation of network.
109 ====================================
112 For memslap, both TCP and UDP use non-blocking network IO. All
113 the network events are managed by libevent as memcached. The network module
114 of memslap is similar to memcached. Libevent can ensure
115 memslap can handle network very efficiently.
118 Effective implementation of multi-threads and concurrency
119 =========================================================
122 Memslap has the similar implementation of multi-threads to
123 memcached. Memslap creates one or more self-governed threads;
124 each thread is bound with one CPU core if the system supports setting CPU
127 In addition, each thread has a libevent to manage the events of the network;
128 each thread has one or more self-governed concurrencies; and each
129 concurrency has one or more socket connections. All the concurrencies don’t
130 communicate with each other even though they are in the same thread.
132 Memslap can create thousands of socket connections, and each
133 concurrency has tens of socket connections. Each concurrency randomly or
134 sequentially selects one socket connection from its socket connection pool
135 to run, so memslap can ensure each concurrency handles one
136 socket connection at any given time. Users can specify the number of
137 concurrency and socket connections of each concurrency according to their
141 Effective implementation of generating key and value
142 ====================================================
145 In order to improve time efficiency and space efficiency,
146 memslap creates a random characters table with 10M characters. All the
147 suffixes of keys and values are generated from this random characters table.
149 Memslap uses the offset in the character table and the length
150 of the string to identify a string. It can save much memory.
151 Each key contains two parts, a prefix and a suffix. The prefix is an
152 uint64_t, 8 bytes. In order to verify the data set before,
153 memslap need to ensure each key is unique, so it uses the prefix to identify
154 a key. The prefix cannot include illegal characters, such as ‘\r’, ‘\n’,
155 ‘\0’ and ‘ ‘. And memslap has an algorithm to ensure that.
157 Memslap doesn’t generate all the objects (key-value pairs) at
158 the beginning. It only generates enough objects to fill the task window
159 (default 10K objects) of each concurrency. Each object has the following
160 basic information, key prefix, key suffix offset in the character table, key
161 length, value offset in the character table, and value length.
163 In the work process, each concurrency sequentially or randomly selects an
164 object from the window to do set operation or get operation. At the same
165 time, each concurrency kicks objects out of its window and adds new object
169 Simple but useful task scheduling
170 =================================
173 Memslap uses libevent to schedule all the concurrencies of
174 threads, and each concurrency schedules tasks based on the local task
175 window. Memslap assumes that if each concurrency keeps the same
176 key distribution, value distribution and commands distribution, from
177 outside, memslap keeps all the distribution as a whole.
178 Each task window includes a lot of objects, each object stores its basic
179 information, such as key, value, expire time, and so on. At any time, all
180 the objects in the window keep the same and fixed key and value
181 distribution. If an object is overwritten, the value of the object will be
182 updated. Memslap verifies the data or expire-time according to
183 the object information stored in the task window.
185 Libevent selects which concurrency to handle based on a specific network
186 event. Then the concurrency selects which command (get or set) to operate
187 based on the command distribution. If it needs to kick out an old object and
188 add a new object, in order to keep the same key and value distribution, the
189 new object must have the same key length and value length.
191 If memcached server has two cache layers (memory and SSD), running
192 memslap with different window sizes can get different cache
193 miss rates. If memslap adds enough objects into the windows at
194 the beginning, and the cache of memcached cannot store all the objects
195 initialized, then memslap will get some objects from the second
196 cache layer. It causes the first cache layer to miss. So the user can
197 specify the window size to get the expected miss rate of the first cache
201 Useful implementation of multi-servers , UDP, TCP, multi-get and binary protocol
202 ================================================================================
205 Because each thread is self-governed, memslap can assign
206 different threads to handle different memcached servers. This is just one of
207 the ways in which memslap supports multiple servers. The only
208 limitation is that the number of servers cannot be greater than the number
209 of threads. The other way to support multiple servers is for replication
210 test. Each concurrency has one socket connection to each memcached server.
211 For the implementation, memslap can set some objects to one
212 memcached server, and get these objects from the other servers.
214 By default, Memslap does single get. If the user specifies
215 multi-get option, memslap will collect enough get commands and
216 pack and send the commands together.
218 Memslap supports both the ASCII protocol and binary protocol,
219 but it runs on the ASCII protocol by default.
220 Memslap by default runs on the TCP protocol, but it also
221 supports UDP. Because UDP is unreliable, dropped packages and out-of-order
222 packages may occur. Memslap creates a memory buffer to handle
223 these problems. Memslap tries to read all the response data of
224 one command from the server and reorders the response data. If some packages
225 get lost, the waiting timeout mechanism can ensure half-baked packages will
226 be discarded and the next command will be sent.
235 Below are some usage samples:
238 memslap -s 127.0.0.1:11211 -S 5s
242 memslap -s 127.0.0.1:11211 -t 2m -v 0.2 -e 0.05 -b
246 memslap -s 127.0.0.1:11211 -F config -t 2m -w 40k -S 20s -o 0.2
250 memslap -s 127.0.0.1:11211 -F config -t 2m -T 4 -c 128 -d 20 -P 40k
254 memslap -s 127.0.0.1:11211 -F config -t 2m -d 50 -a -n 40
258 memslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m
262 memslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m -p 2
266 The user must specify one server at least to run memslap. The
267 rest of the parameters have default values, as shown below:
269 Thread number = 1 Concurrency = 16
271 Run time = 600 seconds Configuration file = NULL
273 Key size = 64 Value size = 1024
275 Get/set = 9:1 Window size = 10k
277 Execute number = 0 Single get = true
279 Multi-get = false Number of sockets of each concurrency = 1
281 Reconnect = false Data verification = false
283 Expire-time verification = false ASCII protocol = true
285 Binary protocol = false Dumping statistic information
289 Overwrite proportion = 0% UDP = false
291 TCP = true Limit throughput = false
293 Facebook test = false Replication test = false
295 Key size, value size and command distribution.
296 ==============================================
299 All the distributions are read from the configuration file specified by user
300 with “—cfg_cmd” option. If the user does not specify a configuration file,
301 memslap will run with the default distribution (key size = 64,
302 value size = 1024, get/set = 9:1). For information on how to edit the
303 configuration file, refer to the “Configuration File” section.
305 The minimum key size is 16 bytes; the maximum key size is 250 bytes. The
306 precision of proportion is 0.001. The proportion of distribution will be
307 rounded to 3 decimal places.
309 The minimum value size is 1 bytes; the maximum value size is 1M bytes. The
310 precision of proportion is 0.001. The proportion of distribution will be
311 rounded to 3 decimal places.
312 Currently, memslap only supports set and get commands. And it
313 supports 100% set and 100% get. For 100% get, it will preset some objects to
317 Multi-thread and concurrency
318 ============================
321 The high performance of memslap benefits from the special
322 schedule of thread and concurrency. It’s important to specify the proper
323 number of them. The default number of threads is 1; the default number of
324 concurrency is 16. The user can use “—threads” and “--concurrency” to
325 specify these variables.
327 If the system supports setting CPU affinity and the number of threads
328 specified by the user is greater than 1, memslap will try to
329 bind each thread to a different CPU core. So if you want to get the best
330 performance memslap, it is better to specify the number of
331 thread equal to the number of CPU cores. The number of threads specified by
332 the user can also be less or greater than the number of CPU cores. Because
333 of the limitation of implementation, the number of concurrencies could be
334 the multiple of the number of threads.
336 1. For 8 CPU cores system
340 --threads=2 --concurrency=128
342 --threads=8 --concurrency=128
344 --threads=8 --concurrency=256
346 --threads=12 --concurrency=144
348 2. For 16 CPU cores system
352 --threads=8 --concurrency=128
354 --threads=16 --concurrency=256
356 --threads=16 --concurrency=512
358 --threads=24 --concurrency=288
360 The memslap performs very well, when
361 used to test the performance of memcached servers.
362 Most of the time, the bottleneck is the network or
363 the server. If for some reason the user wants to
364 limit the performance of memslap, there
365 are two ways to do this:
367 Decrease the number of threads and concurrencies.
368 Use the option “--tps” that memslap
369 provides to limit the throughput. This option allows
370 the user to get the expected throughput. For
371 example, assume that the maximum throughput is 50
372 kops/s for a specific configuration, you can specify
373 the throughput equal to or less than the maximum
374 throughput using “--tps” option.
381 Most of the time, the user does not need to specify the window size. The
382 default window size is 10k. For Schooner Memcached, the user can specify
383 different window sizes to get different cache miss rates based on the test
384 case. Memslap supports cache miss rate between 0% and 100%.
385 If you use this utility to test the performance of Schooner Memcached, you
386 can specify a proper window size to get the expected cache miss rate. The
387 formula for calculating window size is as follows:
389 Assume that the key size is 128 bytes, and the value size is 2048 bytes, and
392 1. Small cache cache_size=1M, 100% cache miss (all data get from SSD).
397 (1). cache miss rate 0%
401 (2). cache miss rate 5%
407 (1). cache miss rate 0%
417 The formula for calculating window size for cache miss rate 0%:
419 cache_size / concurrency / (key_size + value_size) \* 0.5
421 The formula for calculating window size for cache miss rate 5%:
423 cache_size / concurrency / (key_size + value_size) \* 0.7
430 Memslap supports both data verification and expire-time
431 verification. The user can use "--verify=" or "-v" to specify the proportion
432 of data verification. In theory, it supports 100% data verification. The
433 user can use "--exp_verify=" or "-e" to specify the proportion of
434 expire-time verification. In theory, it supports 100% expire-time
435 verification. Specify the "--verbose" options to get more detailed error
438 For example: --exp_verify=0.01 –verify=0.1 , it means that 1% of the objects
439 set with expire-time, 10% of the objects gotten will be verified. If the
440 objects are gotten, memslap will verify the expire-time and
444 multi-servers and multi-clients
445 ===============================
448 Memslap supports multi-servers based on self-governed thread.
449 There is a limitation that the number of servers cannot be greater than the
450 number of threads. Memslap assigns one thread to handle one
451 server at least. The user can use the "--servers=" or "-s" option to specify
456 --servers=10.1.1.1:11211,10.1.1.2:11212,10.1.1.3:11213 --threads=6 --concurrency=36
458 The above command means that there are 6 threads, with each thread having 6
459 concurrencies and that threads 0 and 3 handle server 0 (10.1.1.1); threads 1
460 and 4 handle server 1 (10.1.1.2); and thread 2 and 5 handle server 2
463 All the threads and concurrencies in memslap are self-governed.
465 So is memslap. The user can start up several
466 memslap instances. The user can run memslap on different client
467 machines to communicate with the same memcached server at the same. It is
468 recommended that the user start different memslap on different
469 machines using the same configuration.
472 Run with execute number mode or time mode
473 =========================================
476 The default memslap runs with time mode. The default run time
477 is 10 minutes. If it times out, memslap will exit. Do not
478 specify both execute number mode and time mode at the same time; just
483 --time=30s (It means the test will run 30 seconds.)
485 --execute_number=100000 (It means that after running 100000 commands, the test will exit.)
488 Dump statistic information periodically.
489 ========================================
492 The user can use "--stat_freq=" or "-S" to specify the frequency.
498 Memslap will dump the statistics of the commands (get and set) at the frequency of every 20
501 For more information on the format of dumping statistic information, refer to “Format of Output” section.
508 The user can use "--division=" or "-d" to specify multi-get keys count.
509 Memslap by default does single get with TCP. Memslap also supports data
510 verification and expire-time verification for multi-get.
512 Memslap supports multi-get with both TCP and UDP. Because of
513 the different implementation of the ASCII protocol and binary protocol,
514 there are some differences between the two. For the ASCII protocol,
515 memslap sends one “multi-get” to the server once. For the
516 binary protocol, memslap sends several single get commands
517 together as “multi-get” to the server.
524 Memslap supports both UDP and TCP. For TCP,
525 memslap does not reconnect the memcached server if socket connections are
526 lost. If all the socket connections are lost or memcached server crashes,
527 memslap will exit. If the user specifies the “--reconnect”
528 option when socket connections are lost, it will reconnect them.
530 User can use “--udp” to enable the UDP feature, but UDP comes with some
533 UDP cannot set data more than 1400 bytes.
535 UDP is not supported by the binary protocol because the binary protocol of
536 memcached does not support that.
538 UDP doesn’t support reconnection.
545 Set data with TCP and multi-get with UDP. Specify the following options:
547 "--facebook --division=50"
549 If you want to create thousands of TCP connections, specify the
551 "--conn_sock=" option.
553 For example: --facebook --division=50 --conn_sock=200
555 The above command means that memslap will do facebook test,
556 each concurrency has 200 socket TCP connections and one UDP socket.
558 Memslap sets objects with the TCP socket, and multi-gets 50
559 objects once with the UDP socket.
561 If you specify "--division=50", the key size must be less that 25 bytes
562 because the UDP packet size is 1400 bytes.
569 For replication test, the user must specify at least two memcached servers.
570 The user can use “—rep_write=” option to enable feature.
574 --servers=10.1.1.1:11211,10.1.1.2:11212 –rep_write=2
576 The above command means that there are 2 replication memcached servers,
577 memslap will set objects to both server 0 and server 1, get
578 objects which are set to server 0 before from server 1, and also get objects
579 which are set to server 1 before from server 0. If server 0 crashes,
580 memslap will only get objects from server 1. If server 0 comes
581 back to life again, memslap will reconnect server 0. If both
582 server 0 and server 1 crash, memslap will exit.
585 Supports thousands of TCP connections
586 =====================================
589 Start memslap with "--conn_sock=" or "-n" to enable this
590 feature. Make sure that your system can support opening thousands of files
591 and creating thousands of sockets. However, this feature does not support
592 reconnection if sockets disconnect.
596 --threads=8 --concurrency=128 --conn_sock=128
598 The above command means that memslap starts up 8 threads, each
599 thread has 16 concurrencies, each concurrency has 128 TCP socket
600 connections, and the total number of TCP socket connections is 128 \* 128 =
604 Supports binary protocol
605 ========================
608 Start memslap with "--binary" or "-B" options to enable this
609 feature. It supports all the above features except UDP, because the latest
610 memcached 1.3.3 does not implement binary UDP protocol.
616 Since memcached 1.3.3 doesn't implement binary UDP protocol,
617 memslap does not support UDP. In addition, memcached 1.3.3 does not support
618 multi-get. If you specify "--division=50" option, it just sends 50 get
619 commands together as “mulit-get” to the server.
628 This section describes the format of the configuration file. By default
629 when no configuration file is specified memslap reads the default
630 one located at ~/.memslap.cnf.
632 Below is a sample configuration file:
637 ***************************************************************************
638 #comments should start with '#'
640 #start_len end_len proportion
642 #key length range from start_len to end_len
643 #start_len must be equal to or greater than 16
644 #end_len must be equal to or less than 250
645 #start_len must be equal to or greater than end_len
646 #memslap will generate keys according to the key range
647 #proportion: indicates keys generated from one range accounts for the total
650 #example1: key range 16~100 accounts for 80%
651 # key range 101~200 accounts for 10%
652 # key range 201~250 accounts for 10%
653 # total should be 1 (0.8+0.1+0.1 = 1)
659 #example2: all keys length are 128 bytes
665 #start_len end_len proportion
667 #value length range from start_len to end_len
668 #start_len must be equal to or greater than 1
669 #end_len must be equal to or less than 1M
670 #start_len must be equal to or greater than end_len
671 #memslap will generate values according to the value range
672 #proportion: indicates values generated from one range accounts for the
673 total generated values
675 #example1: value range 1~1000 accounts for 80%
676 # value range 1001~10000 accounts for 10%
677 # value range 10001~100000 accounts for 10%
678 # total should be 1 (0.8+0.1+0.1 = 1)
684 #example2: all value length are 128 bytes
690 #cmd_type cmd_proportion
692 #currently memslap only supports get and set command.
698 #example: set command accounts for 50%
699 # get command accounts for 50%
700 # total should be 1 (0.5+0.5 = 1)
716 At the beginning, memslap displays some configuration information as follows:
719 servers : 127.0.0.1:11211
739 set proportion: set_prop=0.10
743 get proportion: get_prop=0.90
754 The servers used by memslap.
760 The number of threads memslap runs with.
766 The number of concurrencies memslap runs with.
772 How long to run memslap.
778 The task window size of each concurrency.
784 The proportion of set command.
790 The proportion of get command.
794 The output of dynamic statistics is something like this:
799 ---------------------------------------------------------------------------------------------------------------------------------
801 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
802 Avg(us) Std_dev Geo_dist
803 Period 5 345826 69165 65.3 0 27 2198 203
805 Global 20 1257935 62896 71.8 0 26 3791 224
810 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
811 Avg(us) Std_dev Geo_dist
812 Period 5 38425 7685 7.3 0 42 628 240
814 Global 20 139780 6989 8.0 0 37 3790 253
819 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
820 Avg(us) Std_dev Geo_dist
821 Period 5 384252 76850 72.5 0 27 2198 207
823 Global 20 1397720 69886 79.7 0 26 3791 227
825 ---------------------------------------------------------------------------------------------------------------------------------
836 Statistics information of get command
842 Statistics information of set command
848 Statistics information of both get and set command
854 Result within a period
872 Throughput, operations/second
884 How many objects can’t be gotten
890 The minimum response time
896 The maximum response time
902 The average response time
908 Standard deviation of response time
914 Geometric distribution based on natural exponential function
918 At the end, memslap will output something like this:
923 ---------------------------------------------------------------------------------------------------------------------------------
924 Get Statistics (1257956 events)
932 8: 484890 459823 12543 824
935 Set Statistics (139782 events)
943 8: 50784 65574 2064 167
946 Total Statistics (1397738 events)
954 8: 535674 525397 14607 991
964 written_bytes: 242516030
965 read_bytes: 1003702556
966 object_bytes: 152086080
971 Run time: 20.0s Ops: 1397754 TPS: 69817 Net_rate: 59.4M/s
972 ---------------------------------------------------------------------------------------------------------------------------------
983 Get statistics of response time
989 Set statistics of response time
995 Both get and set statistics of response time
1001 The accumulated and minimum response time
1007 The accumulated and maximum response time
1013 The accumulated and average response time
1019 Standard deviation of response time
1025 Geometric distribution based on logarithm 2
1031 Total get commands done
1037 Total set commands done
1043 How many objects can’t be gotten from server
1049 How many objects need to verify but can’t get them
1055 How many objects with insistent value
1061 How many objects are expired but we get them
1067 How many objects are unexpired but we can’t get them
1091 How many UDP packages are disorder
1097 How many UDP packages are lost
1103 How many times UDP time out happen
1121 Throughput, operations/second
1127 The average rate of network
1139 List one or more servers to connect. Servers count must be less than
1140 threads count. e.g.: --servers=localhost:1234,localhost:11211
1143 Number of threads to startup, better equal to CPU numbers. Default 8.
1146 Number of concurrency to simulate with load. Default 128.
1149 Number of TCP socks per concurrency. Default 1.
1151 -x, --execute_number=
1152 Number of operations(get and set) to execute for the
1153 given test. Default 1000000.
1156 How long the test to run, suffix: s-seconds, m-minutes, h-hours,
1157 d-days e.g.: --time=2h.
1160 Load the configure file to get command,key and value distribution list.
1163 Task window size of each concurrency, suffix: K, M e.g.: --win_size=10k.
1167 Fixed length of value.
1170 The proportion of date verification, e.g.: --verify=0.01
1173 Number of keys to multi-get once. Default 1, means single get.
1176 Frequency of dumping statistic information. suffix: s-seconds,
1177 m-minutes, e.g.: --resp_freq=10s.
1180 The proportion of objects with expire time, e.g.: --exp_verify=0.01.
1181 Default no object with expire time
1184 The proportion of objects need overwrite, e.g.: --overwrite=0.01.
1185 Default never overwrite object.
1188 Reconnect support, when connection is closed it will be reconnected.
1191 UDP support, default memslap uses TCP, TCP port and UDP port of
1192 server must be same.
1195 Whether it enables facebook test feature, set with TCP and multi-get with UDP.
1198 Whether it enables binary protocol. Default with ASCII protocol.
1201 Expected throughput, suffix: K, e.g.: --tps=10k.
1204 The first nth servers can write data, e.g.: --rep_write=2.
1207 Whether it outputs detailed information when verification fails.
1210 Display this message and then exit.
1213 Display the version of the application and then exit.
1221 memslap -s 127.0.0.1:11211 -S 5s
1223 memslap -s 127.0.0.1:11211 -t 2m -v 0.2 -e 0.05 -b
1225 memslap -s 127.0.0.1:11211 -F config -t 2m -w 40k -S 20s -o 0.2
1227 memslap -s 127.0.0.1:11211 -F config -t 2m -T 4 -c 128 -d 20 -P 40k
1229 memslap -s 127.0.0.1:11211 -F config -t 2m -d 50 -a -n 40
1231 memslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m
1233 memslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m -p 2
1241 To find out more information please check:
1242 `http://launchpad.org/libmemcached <http://launchpad.org/libmemcached>`_
1250 Mingqiang Zhuang <mingqiangzhuang@hengtiansoft.com> (Schooner Technolgy)
1251 Brian Aker, <brian@tangent.org>
1259 memcached(1) libmemcached(3)