1 ==================================================
2 memaslap - Load testing and benchmarking a server
3 ==================================================
16 .. envvar:: MEMCACHED_SERVERS
23 :program:`memaslap` is a load generation and benchmark tool for memcached
24 servers. It generates configurable workload such as threads, concurrencies,
25 connections, run time, overwrite, miss rate, key size, value size, get/set
26 proportion, expected throughput, and so on. Furthermore, it also testss data
27 verification, expire-time verification, UDP, binary protocol, facebook test,
28 replication test, multi-get and reconnection, etc.
30 Memaslap manages network connections like memcached with
31 libevent. Each thread of memaslap is bound with a CPU core, all
32 the threads don't communicate with each other, and there are several socket
33 connections in each thread. Each connection keeps key size distribution,
34 value size distribution, and command distribution by itself.
36 You can specify servers via the :option:`memslap --servers` option or via the
37 environment variable :envvar:`MEMCACHED_SERVERS`.
45 Memslap is developed to for the following purposes:
48 Manages network connections with libevent asynchronously.
52 Set both TCP and UDP up to use non-blocking IO.
56 Improves parallelism: higher performance in multi-threads environments.
60 Improves time efficiency: faster processing speed.
64 Generates key and value more efficiently; key size distribution and value size distribution are configurable.
68 Supports get, multi-get, and set commands; command distribution is configurable.
72 Supports controllable miss rate and overwrite rate.
76 Supports data and expire-time verification.
80 Supports dumping statistic information periodically.
84 Supports thousands of TCP connections.
88 Supports binary protocol.
92 Supports facebook test (set with TCP and multi-get with UDP) and replication test.
102 Effective implementation of network.
103 ____________________________________
106 For memaslap, both TCP and UDP use non-blocking network IO. All
107 the network events are managed by libevent as memcached. The network module
108 of memaslap is similar to memcached. Libevent can ensure
109 memaslap can handle network very efficiently.
112 Effective implementation of multi-threads and concurrency
113 _________________________________________________________
116 Memslap has the similar implementation of multi-threads to
117 memcached. Memslap creates one or more self-governed threads;
118 each thread is bound with one CPU core if the system testss setting CPU
121 In addition, each thread has a libevent to manage the events of the network;
122 each thread has one or more self-governed concurrencies; and each
123 concurrency has one or more socket connections. All the concurrencies don’t
124 communicate with each other even though they are in the same thread.
126 Memslap can create thousands of socket connections, and each
127 concurrency has tens of socket connections. Each concurrency randomly or
128 sequentially selects one socket connection from its socket connection pool
129 to run, so memaslap can ensure each concurrency handles one
130 socket connection at any given time. Users can specify the number of
131 concurrency and socket connections of each concurrency according to their
135 Effective implementation of generating key and value
136 ____________________________________________________
139 In order to improve time efficiency and space efficiency,
140 memaslap creates a random characters table with 10M characters. All the
141 suffixes of keys and values are generated from this random characters table.
143 Memslap uses the offset in the character table and the length
144 of the string to identify a string. It can save much memory.
145 Each key contains two parts, a prefix and a suffix. The prefix is an
146 uint64_t, 8 bytes. In order to verify the data set before,
147 memaslap need to ensure each key is unique, so it uses the prefix to identify
148 a key. The prefix cannot include illegal characters, such as ‘\r’, ‘\n’,
149 ‘\0’ and ‘ ‘. And memaslap has an algorithm to ensure that.
151 Memslap doesn’t generate all the objects (key-value pairs) at
152 the beginning. It only generates enough objects to fill the task window
153 (default 10K objects) of each concurrency. Each object has the following
154 basic information, key prefix, key suffix offset in the character table, key
155 length, value offset in the character table, and value length.
157 In the work process, each concurrency sequentially or randomly selects an
158 object from the window to do set operation or get operation. At the same
159 time, each concurrency kicks objects out of its window and adds new object
163 Simple but useful task scheduling
164 _________________________________
167 Memslap uses libevent to schedule all the concurrencies of
168 threads, and each concurrency schedules tasks based on the local task
169 window. Memslap assumes that if each concurrency keeps the same
170 key distribution, value distribution and commands distribution, from
171 outside, memaslap keeps all the distribution as a whole.
172 Each task window includes a lot of objects, each object stores its basic
173 information, such as key, value, expire time, and so on. At any time, all
174 the objects in the window keep the same and fixed key and value
175 distribution. If an object is overwritten, the value of the object will be
176 updated. Memslap verifies the data or expire-time according to
177 the object information stored in the task window.
179 Libevent selects which concurrency to handle based on a specific network
180 event. Then the concurrency selects which command (get or set) to operate
181 based on the command distribution. If it needs to kick out an old object and
182 add a new object, in order to keep the same key and value distribution, the
183 new object must have the same key length and value length.
185 If memcached server has two cache layers (memory and SSD), running
186 memaslap with different window sizes can get different cache
187 miss rates. If memaslap adds enough objects into the windows at
188 the beginning, and the cache of memcached cannot store all the objects
189 initialized, then memaslap will get some objects from the second
190 cache layer. It causes the first cache layer to miss. So the user can
191 specify the window size to get the expected miss rate of the first cache
195 Useful implementation of multi-servers , UDP, TCP, multi-get and binary protocol
196 ________________________________________________________________________________
199 Because each thread is self-governed, memaslap can assign
200 different threads to handle different memcached servers. This is just one of
201 the ways in which memaslap tests multiple servers. The only
202 limitation is that the number of servers cannot be greater than the number
203 of threads. The other way to test multiple servers is for replication
204 test. Each concurrency has one socket connection to each memcached server.
205 For the implementation, memaslap can set some objects to one
206 memcached server, and get these objects from the other servers.
208 By default, Memslap does single get. If the user specifies
209 multi-get option, memaslap will collect enough get commands and
210 pack and send the commands together.
212 Memslap testss both the ASCII protocol and binary protocol,
213 but it runs on the ASCII protocol by default.
214 Memslap by default runs on the TCP protocol, but it also
215 tests UDP. Because UDP is unreliable, dropped packages and out-of-order
216 packages may occur. Memslap creates a memory buffer to handle
217 these problems. Memslap tries to read all the response data of
218 one command from the server and reorders the response data. If some packages
219 get lost, the waiting timeout mechanism can ensure half-baked packages will
220 be discarded and the next command will be sent.
229 Below are some usage samples:
232 memaslap -s 127.0.0.1:11211 -S 5s
236 memaslap -s 127.0.0.1:11211 -t 2m -v 0.2 -e 0.05 -b
240 memaslap -s 127.0.0.1:11211 -F config -t 2m -w 40k -S 20s -o 0.2
244 memaslap -s 127.0.0.1:11211 -F config -t 2m -T 4 -c 128 -d 20 -P 40k
248 memaslap -s 127.0.0.1:11211 -F config -t 2m -d 50 -a -n 40
252 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m
256 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m -p 2
260 The user must specify one server at least to run memaslap. The
261 rest of the parameters have default values, as shown below:
263 Thread number = 1 Concurrency = 16
265 Run time = 600 seconds Configuration file = NULL
267 Key size = 64 Value size = 1024
269 Get/set = 9:1 Window size = 10k
271 Execute number = 0 Single get = true
273 Multi-get = false Number of sockets of each concurrency = 1
275 Reconnect = false Data verification = false
277 Expire-time verification = false ASCII protocol = true
279 Binary protocol = false Dumping statistic information
283 Overwrite proportion = 0% UDP = false
285 TCP = true Limit throughput = false
287 Facebook test = false Replication test = false
289 Key size, value size and command distribution.
290 ______________________________________________
293 All the distributions are read from the configuration file specified by user
294 with “—cfg_cmd” option. If the user does not specify a configuration file,
295 memaslap will run with the default distribution (key size = 64,
296 value size = 1024, get/set = 9:1). For information on how to edit the
297 configuration file, refer to the “Configuration File” section.
299 The minimum key size is 16 bytes; the maximum key size is 250 bytes. The
300 precision of proportion is 0.001. The proportion of distribution will be
301 rounded to 3 decimal places.
303 The minimum value size is 1 bytes; the maximum value size is 1M bytes. The
304 precision of proportion is 0.001. The proportion of distribution will be
305 rounded to 3 decimal places.
306 Currently, memaslap only testss set and get commands. And it
307 testss 100% set and 100% get. For 100% get, it will preset some objects to
311 Multi-thread and concurrency
312 ____________________________
315 The high performance of memaslap benefits from the special
316 schedule of thread and concurrency. It’s important to specify the proper
317 number of them. The default number of threads is 1; the default number of
318 concurrency is 16. The user can use “—threads” and “--concurrency” to
319 specify these variables.
321 If the system tests setting CPU affinity and the number of threads
322 specified by the user is greater than 1, memaslap will try to
323 bind each thread to a different CPU core. So if you want to get the best
324 performance memaslap, it is better to specify the number of
325 thread equal to the number of CPU cores. The number of threads specified by
326 the user can also be less or greater than the number of CPU cores. Because
327 of the limitation of implementation, the number of concurrencies could be
328 the multiple of the number of threads.
330 1. For 8 CPU cores system
334 --threads=2 --concurrency=128
336 --threads=8 --concurrency=128
338 --threads=8 --concurrency=256
340 --threads=12 --concurrency=144
342 2. For 16 CPU cores system
346 --threads=8 --concurrency=128
348 --threads=16 --concurrency=256
350 --threads=16 --concurrency=512
352 --threads=24 --concurrency=288
354 The memaslap performs very well, when
355 used to test the performance of memcached servers.
356 Most of the time, the bottleneck is the network or
357 the server. If for some reason the user wants to
358 limit the performance of memaslap, there
359 are two ways to do this:
361 Decrease the number of threads and concurrencies.
362 Use the option “--tps” that memaslap
363 provides to limit the throughput. This option allows
364 the user to get the expected throughput. For
365 example, assume that the maximum throughput is 50
366 kops/s for a specific configuration, you can specify
367 the throughput equal to or less than the maximum
368 throughput using “--tps” option.
375 Most of the time, the user does not need to specify the window size. The
376 default window size is 10k. For Schooner Memcached, the user can specify
377 different window sizes to get different cache miss rates based on the test
378 case. Memslap testss cache miss rate between 0% and 100%.
379 If you use this utility to test the performance of Schooner Memcached, you
380 can specify a proper window size to get the expected cache miss rate. The
381 formula for calculating window size is as follows:
383 Assume that the key size is 128 bytes, and the value size is 2048 bytes, and
386 1. Small cache cache_size=1M, 100% cache miss (all data get from SSD).
391 (1). cache miss rate 0%
395 (2). cache miss rate 5%
401 (1). cache miss rate 0%
411 The formula for calculating window size for cache miss rate 0%:
413 cache_size / concurrency / (key_size + value_size) \* 0.5
415 The formula for calculating window size for cache miss rate 5%:
417 cache_size / concurrency / (key_size + value_size) \* 0.7
424 Memslap testss both data verification and expire-time
425 verification. The user can use "--verify=" or "-v" to specify the proportion
426 of data verification. In theory, it testss 100% data verification. The
427 user can use "--exp_verify=" or "-e" to specify the proportion of
428 expire-time verification. In theory, it testss 100% expire-time
429 verification. Specify the "--verbose" options to get more detailed error
432 For example: --exp_verify=0.01 –verify=0.1 , it means that 1% of the objects
433 set with expire-time, 10% of the objects gotten will be verified. If the
434 objects are gotten, memaslap will verify the expire-time and
438 multi-servers and multi-config
439 _______________________________
442 Memslap testss multi-servers based on self-governed thread.
443 There is a limitation that the number of servers cannot be greater than the
444 number of threads. Memslap assigns one thread to handle one
445 server at least. The user can use the "--servers=" or "-s" option to specify
450 --servers=10.1.1.1:11211,10.1.1.2:11212,10.1.1.3:11213 --threads=6 --concurrency=36
452 The above command means that there are 6 threads, with each thread having 6
453 concurrencies and that threads 0 and 3 handle server 0 (10.1.1.1); threads 1
454 and 4 handle server 1 (10.1.1.2); and thread 2 and 5 handle server 2
457 All the threads and concurrencies in memaslap are self-governed.
459 So is memaslap. The user can start up several
460 memaslap instances. The user can run memaslap on different client
461 machines to communicate with the same memcached server at the same. It is
462 recommended that the user start different memaslap on different
463 machines using the same configuration.
466 Run with execute number mode or time mode
467 _________________________________________
470 The default memaslap runs with time mode. The default run time
471 is 10 minutes. If it times out, memaslap will exit. Do not
472 specify both execute number mode and time mode at the same time; just
477 --time=30s (It means the test will run 30 seconds.)
479 --execute_number=100000 (It means that after running 100000 commands, the test will exit.)
482 Dump statistic information periodically.
483 ________________________________________
486 The user can use "--stat_freq=" or "-S" to specify the frequency.
492 Memslap will dump the statistics of the commands (get and set) at the frequency of every 20
495 For more information on the format of dumping statistic information, refer to “Format of Output” section.
502 The user can use "--division=" or "-d" to specify multi-get keys count.
503 Memslap by default does single get with TCP. Memslap also testss data
504 verification and expire-time verification for multi-get.
506 Memslap testss multi-get with both TCP and UDP. Because of
507 the different implementation of the ASCII protocol and binary protocol,
508 there are some differences between the two. For the ASCII protocol,
509 memaslap sends one “multi-get” to the server once. For the
510 binary protocol, memaslap sends several single get commands
511 together as “multi-get” to the server.
518 Memslap testss both UDP and TCP. For TCP,
519 memaslap does not reconnect the memcached server if socket connections are
520 lost. If all the socket connections are lost or memcached server crashes,
521 memaslap will exit. If the user specifies the “--reconnect”
522 option when socket connections are lost, it will reconnect them.
524 User can use “--udp” to enable the UDP feature, but UDP comes with some
527 UDP cannot set data more than 1400 bytes.
529 UDP is not testsed by the binary protocol because the binary protocol of
530 memcached does not tests that.
532 UDP doesn’t tests reconnection.
539 Set data with TCP and multi-get with UDP. Specify the following options:
541 "--facebook --division=50"
543 If you want to create thousands of TCP connections, specify the
545 "--conn_sock=" option.
547 For example: --facebook --division=50 --conn_sock=200
549 The above command means that memaslap will do facebook test,
550 each concurrency has 200 socket TCP connections and one UDP socket.
552 Memslap sets objects with the TCP socket, and multi-gets 50
553 objects once with the UDP socket.
555 If you specify "--division=50", the key size must be less that 25 bytes
556 because the UDP packet size is 1400 bytes.
563 For replication test, the user must specify at least two memcached servers.
564 The user can use “—rep_write=” option to enable feature.
568 --servers=10.1.1.1:11211,10.1.1.2:11212 –rep_write=2
570 The above command means that there are 2 replication memcached servers,
571 memaslap will set objects to both server 0 and server 1, get
572 objects which are set to server 0 before from server 1, and also get objects
573 which are set to server 1 before from server 0. If server 0 crashes,
574 memaslap will only get objects from server 1. If server 0 comes
575 back to life again, memaslap will reconnect server 0. If both
576 server 0 and server 1 crash, memaslap will exit.
579 Supports thousands of TCP connections
580 _____________________________________
583 Start memaslap with "--conn_sock=" or "-n" to enable this
584 feature. Make sure that your system can tests opening thousands of files
585 and creating thousands of sockets. However, this feature does not tests
586 reconnection if sockets disconnect.
590 --threads=8 --concurrency=128 --conn_sock=128
592 The above command means that memaslap starts up 8 threads, each
593 thread has 16 concurrencies, each concurrency has 128 TCP socket
594 connections, and the total number of TCP socket connections is 128 \* 128 =
598 Supports binary protocol
599 ________________________
602 Start memaslap with "--binary" or "-B" options to enable this
603 feature. It testss all the above features except UDP, because the latest
604 memcached 1.3.3 does not implement binary UDP protocol.
610 Since memcached 1.3.3 doesn't implement binary UDP protocol,
611 memaslap does not tests UDP. In addition, memcached 1.3.3 does not tests
612 multi-get. If you specify "--division=50" option, it just sends 50 get
613 commands together as “mulit-get” to the server.
622 This section describes the format of the configuration file. By default
623 when no configuration file is specified memaslap reads the default
624 one located at ~/.memaslap.cnf.
626 Below is a sample configuration file:
631 ---------------------------------------------------------------------------
632 #comments should start with '#'
634 #start_len end_len proportion
636 #key length range from start_len to end_len
637 #start_len must be equal to or greater than 16
638 #end_len must be equal to or less than 250
639 #start_len must be equal to or greater than end_len
640 #memaslap will generate keys according to the key range
641 #proportion: indicates keys generated from one range accounts for the total
644 #example1: key range 16~100 accounts for 80%
645 # key range 101~200 accounts for 10%
646 # key range 201~250 accounts for 10%
647 # total should be 1 (0.8+0.1+0.1 = 1)
653 #example2: all keys length are 128 bytes
659 #start_len end_len proportion
661 #value length range from start_len to end_len
662 #start_len must be equal to or greater than 1
663 #end_len must be equal to or less than 1M
664 #start_len must be equal to or greater than end_len
665 #memaslap will generate values according to the value range
666 #proportion: indicates values generated from one range accounts for the
667 total generated values
669 #example1: value range 1~1000 accounts for 80%
670 # value range 1001~10000 accounts for 10%
671 # value range 10001~100000 accounts for 10%
672 # total should be 1 (0.8+0.1+0.1 = 1)
678 #example2: all value length are 128 bytes
684 #cmd_type cmd_proportion
686 #currently memaslap only testss get and set command.
692 #example: set command accounts for 50%
693 # get command accounts for 50%
694 # total should be 1 (0.5+0.5 = 1)
710 At the beginning, memaslap displays some configuration information as follows:
713 servers : 127.0.0.1:11211
733 set proportion: set_prop=0.10
737 get proportion: get_prop=0.90
748 The servers used by memaslap.
754 The number of threads memaslap runs with.
760 The number of concurrencies memaslap runs with.
766 How long to run memaslap.
772 The task window size of each concurrency.
778 The proportion of set command.
784 The proportion of get command.
788 The output of dynamic statistics is something like this:
793 ---------------------------------------------------------------------------------------------------------------------------------
795 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
796 Avg(us) Std_dev Geo_dist
797 Period 5 345826 69165 65.3 0 27 2198 203
799 Global 20 1257935 62896 71.8 0 26 3791 224
804 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
805 Avg(us) Std_dev Geo_dist
806 Period 5 38425 7685 7.3 0 42 628 240
808 Global 20 139780 6989 8.0 0 37 3790 253
813 Type Time(s) Ops TPS(ops/s) Net(M/s) Get_miss Min(us) Max(us)
814 Avg(us) Std_dev Geo_dist
815 Period 5 384252 76850 72.5 0 27 2198 207
817 Global 20 1397720 69886 79.7 0 26 3791 227
819 ---------------------------------------------------------------------------------------------------------------------------------
830 Statistics information of get command
836 Statistics information of set command
842 Statistics information of both get and set command
848 Result within a period
866 Throughput, operations/second
878 How many objects can’t be gotten
884 The minimum response time
890 The maximum response time
896 The average response time
902 Standard deviation of response time
908 Geometric distribution based on natural exponential function
912 At the end, memaslap will output something like this:
917 ---------------------------------------------------------------------------------------------------------------------------------
918 Get Statistics (1257956 events)
926 8: 484890 459823 12543 824
929 Set Statistics (139782 events)
937 8: 50784 65574 2064 167
940 Total Statistics (1397738 events)
948 8: 535674 525397 14607 991
958 written_bytes: 242516030
959 read_bytes: 1003702556
960 object_bytes: 152086080
965 Run time: 20.0s Ops: 1397754 TPS: 69817 Net_rate: 59.4M/s
966 ---------------------------------------------------------------------------------------------------------------------------------
977 Get statistics of response time
983 Set statistics of response time
989 Both get and set statistics of response time
995 The accumulated and minimum response time
1001 The accumulated and maximum response time
1007 The accumulated and average response time
1013 Standard deviation of response time
1019 Geometric distribution based on logarithm 2
1025 Total get commands done
1031 Total set commands done
1037 How many objects can’t be gotten from server
1043 How many objects need to verify but can’t get them
1049 How many objects with insistent value
1055 How many objects are expired but we get them
1061 How many objects are unexpired but we can’t get them
1085 How many UDP packages are disorder
1091 How many UDP packages are lost
1097 How many times UDP time out happen
1115 Throughput, operations/second
1121 The average rate of network
1133 List one or more servers to connect. Servers count must be less than
1134 threads count. e.g.: --servers=localhost:1234,localhost:11211
1137 Number of threads to startup, better equal to CPU numbers. Default 8.
1140 Number of concurrency to simulate with load. Default 128.
1143 Number of TCP socks per concurrency. Default 1.
1145 -x, --execute_number=
1146 Number of operations(get and set) to execute for the
1147 given test. Default 1000000.
1150 How long the test to run, suffix: s-seconds, m-minutes, h-hours,
1151 d-days e.g.: --time=2h.
1154 Load the configure file to get command,key and value distribution list.
1157 Task window size of each concurrency, suffix: K, M e.g.: --win_size=10k.
1161 Fixed length of value.
1164 The proportion of date verification, e.g.: --verify=0.01
1167 Number of keys to multi-get once. Default 1, means single get.
1170 Frequency of dumping statistic information. suffix: s-seconds,
1171 m-minutes, e.g.: --resp_freq=10s.
1174 The proportion of objects with expire time, e.g.: --exp_verify=0.01.
1175 Default no object with expire time
1178 The proportion of objects need overwrite, e.g.: --overwrite=0.01.
1179 Default never overwrite object.
1182 Reconnect tests, when connection is closed it will be reconnected.
1185 UDP tests, default memaslap uses TCP, TCP port and UDP port of
1186 server must be same.
1189 Whether it enables facebook test feature, set with TCP and multi-get with UDP.
1192 Whether it enables binary protocol. Default with ASCII protocol.
1195 Expected throughput, suffix: K, e.g.: --tps=10k.
1198 The first nth servers can write data, e.g.: --rep_write=2.
1201 Whether it outputs detailed information when verification fails.
1204 Display this message and then exit.
1207 Display the version of the application and then exit.
1215 memaslap -s 127.0.0.1:11211 -S 5s
1217 memaslap -s 127.0.0.1:11211 -t 2m -v 0.2 -e 0.05 -b
1219 memaslap -s 127.0.0.1:11211 -F config -t 2m -w 40k -S 20s -o 0.2
1221 memaslap -s 127.0.0.1:11211 -F config -t 2m -T 4 -c 128 -d 20 -P 40k
1223 memaslap -s 127.0.0.1:11211 -F config -t 2m -d 50 -a -n 40
1225 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m
1227 memaslap -s 127.0.0.1:11211,127.0.0.1:11212 -F config -t 2m -p 2
1235 To find out more information please check:
1236 `http://libmemcached.org/ <http://libmemcached.org/>`_
1244 Mingqiang Zhuang <mingqiangzhuang@hengtiansoft.com> (Schooner Technolgy)
1245 Brian Aker, <brian@tangent.org>
1252 :manpage:`memcached(1)` :manpage:`libmemcached(3)`