| 1 | Benjamin Poirier |
| 2 | benjamin.poirier@polymtl.ca |
| 3 | 2009, 2010 |
| 4 | |
| 5 | + About time synchronization |
| 6 | This framework performs offline time synchronization. This means that the |
| 7 | synchronization is done after tracing is over. It is not the same as online |
| 8 | synchronization like what is done by NTP. Nor is it directly influenced by it. |
| 9 | |
| 10 | Event timestamps are adjusted according to a clock correction function that |
| 11 | palliates for initial offset and rate offset (ie. clocks that don't start out |
| 12 | at the same value and clocks that don't run at the same speed). It can work on |
| 13 | two or more traces. |
| 14 | |
| 15 | The synchronization is based on relations identified in network traffic |
| 16 | between nodes. So, for it to work, there must be traffic exchanged between the |
| 17 | nodes. At the moment, this must be TCP traffic. Any kind will do (ssh, http, |
| 18 | ...) |
| 19 | |
| 20 | For scientific information about the algorithms used, see: |
| 21 | * Duda, A., Harrus, G., Haddad, Y., and Bernard, G.: Estimating global time in |
| 22 | distributed systems, Proc. 7th Int. Conf. on Distributed Computing Systems, |
| 23 | Berlin, volume 18, 1987 |
| 24 | * Ashton, P.: Algorithms for Off-line Clock Synchronisation, University of |
| 25 | Canterbury, December 1995 |
| 26 | http://www.cosc.canterbury.ac.nz/research/reports/TechReps/1995/tr_9512.pdf |
| 27 | |
| 28 | + Using time synchronization |
| 29 | ++ Recording traces |
| 30 | To use time synchronization you have to record traces on multiple nodes |
| 31 | simultaneously with lttng (the tracer). While recording the traces, you have |
| 32 | to make sure the following markers are enabled: |
| 33 | * dev_receive |
| 34 | * dev_xmit_extended |
| 35 | * tcpv4_rcv_extended |
| 36 | * udpv4_rcv_extended |
| 37 | You can use 'ltt-armall -n' for this. |
| 38 | |
| 39 | You also have to make sure there is some TCP traffic between the traced nodes. |
| 40 | |
| 41 | ++ Viewing traces |
| 42 | Afterwards, you have to make sure all the traces are accessible from a single |
| 43 | machine, where lttv (the viewer) is run. |
| 44 | |
| 45 | Time synchronization is enabled and controlled via the following lttv options, |
| 46 | as seen with "-h": |
| 47 | --sync |
| 48 | synchronize the time between the traces |
| 49 | --sync-stats |
| 50 | print statistics about the time synchronization |
| 51 | See the section "Statistics" for more information. |
| 52 | --sync-null |
| 53 | read the events but do not perform any processing, this |
| 54 | is mostly for performance evaluation |
| 55 | --sync-analysis - argument: chull, linreg, eval |
| 56 | specify the algorithm to use for event analysis. See the |
| 57 | section "Synchronization Alogrithms". |
| 58 | --sync-reduction - argument: accuracy |
| 59 | specify the algorithm to use for factor reduction. See |
| 60 | the section "Reduction Algorithms". |
| 61 | --sync-graphs |
| 62 | output gnuplot graph showing synchronization points |
| 63 | --sync-graphs-dir - argument: DIRECTORY |
| 64 | specify the directory where to store the graphs, by |
| 65 | default in "graphs-<lttv-pid>" |
| 66 | |
| 67 | To enable synchronization, start lttv with the "--sync" option. It can be |
| 68 | used in text mode or in GUI mode. You can add the traces one by one in the GUI |
| 69 | but this will recompute the synchronization after every trace that is added. |
| 70 | Instead, you can save some time by specifying all your traces on the command |
| 71 | line (using -t). |
| 72 | |
| 73 | Example: |
| 74 | lttv-gui -t traces/node1 -t traces/node2 --sync |
| 75 | |
| 76 | ++ Statistics |
| 77 | The --sync-stats option is useful to know how well the synchronization |
| 78 | algorithms worked. Here is an example output (with added comments) from a |
| 79 | successful chull (one of the synchronization algorithms) run of two traces: |
| 80 | LTTV processing stats: |
| 81 | received frames: 452 |
| 82 | received frames that are IP: 452 |
| 83 | received and processed packets that are TCP: 268 |
| 84 | sent packets that are TCP: 275 |
| 85 | TCP matching stats: |
| 86 | total input and output events matched together to form a packet: 240 |
| 87 | Message traffic: |
| 88 | 0 - 1 : sent 60 received 60 |
| 89 | # Note that 60 + 60 < 240, this is because there was loopback traffic, which is |
| 90 | # discarded. |
| 91 | Convex hull analysis stats: |
| 92 | out of order packets dropped from analysis: 0 |
| 93 | Number of points in convex hulls: |
| 94 | 0 - 1 : lower half-hull 7 upper half-hull 9 |
| 95 | Individual synchronization factors: |
| 96 | 0 - 1 : Middle a0= -1.33641e+08 a1= 1 - 4.5276e-08 accuracy 1.35355e-05 |
| 97 | a0: -1.34095e+08 to -1.33187e+08 (delta= 907388) |
| 98 | a1: 1 -6.81298e-06 to +6.72248e-06 (delta= 1.35355e-05) |
| 99 | # "Middle" is the best type of synchronization for chull. See the section |
| 100 | # "Convex Hull" below. |
| 101 | Resulting synchronization factors: |
| 102 | trace 0 drift= 1 offset= 0 (0.000000) start time= 18.799023588 |
| 103 | trace 1 drift= 1 offset= 1.33641e+08 (0.066818) start time= 19.090688494 |
| 104 | Synchronization time: |
| 105 | real time: 0.113308 |
| 106 | user time: 0.112007 |
| 107 | system time: 0.000000 |
| 108 | |
| 109 | ++ Synchronization Algorithms |
| 110 | The synchronization framework is extensible and already includes two |
| 111 | algorithms: chull and linreg. (There is also a special "eval" module |
| 112 | available.) You can choose which analysis algorithm to use with the |
| 113 | --sync-analysis option. |
| 114 | |
| 115 | +++ Convex Hull |
| 116 | chull, the default analysis module, can provide a garantee that there are no |
| 117 | message inversions after synchronization. When printing the statistics, it |
| 118 | will print, for each trace, the type of factors found: |
| 119 | * "Middle", all went according to assumptions and there will be no message |
| 120 | inversions |
| 121 | * "Fallback", it was not possible to garantee no message inversion so |
| 122 | approximate factors were given instead. This may happen during long running |
| 123 | traces where the non-linearity of the clocks was notable. If you can, try to |
| 124 | reduce the duration of the trace. (Sometimes this may happen during a trace |
| 125 | as short as 120s. but sometimes traces 30 mins. or longer are ok, your |
| 126 | milleage may vary). It would also be to improve the algorithms to avoid |
| 127 | this, see the "Todo" section. In any case, you may get better results (but |
| 128 | still no garantee) by choosing the linreg algorithm instead. |
| 129 | * "Absent", the trace pair does not contain common communication events. Are |
| 130 | you sure the nodes exchanged TCP traffic during the trace? |
| 131 | |
| 132 | There are also other, less common, types. See the enum ApproxType in |
| 133 | event_analysis_chull.h. |
| 134 | |
| 135 | +++ Linear Regression |
| 136 | linreg sometimes gives more precise results than chull but it provides no |
| 137 | garantee |
| 138 | |
| 139 | +++ Synchronization evaluation |
| 140 | eval is a special module, it doesn't really perform synchronization, instead |
| 141 | it calculates and prints different metrics about how well traces are |
| 142 | synchronized. Although it can be run like other analysis modules, it is most |
| 143 | useful when run in a postprocessing step, after another synchronization module |
| 144 | has been run. Eval is most commonly run in text mode. To do this, run: |
| 145 | lttv -m sync_chain_batch [usual options, ex: -t traces/node1 -t traces/node2 |
| 146 | --sync ...] |
| 147 | It can also be run from the lttv source tree via runlttv: |
| 148 | ./runlttv -m eval [usual runlttv options, ex: traces/node1 traces/node2] |
| 149 | |
| 150 | eval provides a few more options: |
| 151 | --eval-rtt-file - argument: FILE |
| 152 | specify the file containing RTT information |
| 153 | --eval-graphs - argument: none |
| 154 | output gnuplot graph showing synchronization points |
| 155 | --eval-graphs-dir - argument: eval-graphs-<lttv pid> |
| 156 | specify the directory where to store the graphs |
| 157 | |
| 158 | The RTT file should contain information on the minimum round-trip time between |
| 159 | nodes involved in the trace. This information is used (optionally) in the |
| 160 | evaluation displayed and in the histogram graphs produced. The file should |
| 161 | contain a series of lines of the form: |
| 162 | 192.168.112.56 192.168.112.57 0.100 |
| 163 | The first two fields are the IP addresses of the source and destination hosts. |
| 164 | (hostnames are not supported). The last field is the minimum rtt in ms. The |
| 165 | fields are separated by whitespace. '#' comments a line. |
| 166 | |
| 167 | Many commands can be used to measure the RTT, for example: |
| 168 | ping -s 8 -A -c 8000 -w 10 192.168.112.57 |
| 169 | |
| 170 | Note that this must be repeated in both directions in the file, that is: |
| 171 | 192.168.112.49 192.168.112.50 0.057 |
| 172 | 192.168.112.50 192.168.112.49 0.050 |
| 173 | |
| 174 | ++++ Linear Programming and GLPK |
| 175 | The synchronization evaluation can optionally perform an analysis similar to |
| 176 | chull but by using a linear program in one of the steps. This can be used to |
| 177 | validate a part of the chull algorithm but it can also be used to provide a |
| 178 | measure of the accuracy of the synchronization in any point (this is seen in |
| 179 | the graph output). |
| 180 | |
| 181 | This is enabled by default at configure time (--with-glpk) if the GNU Linear |
| 182 | Programming Kit is available (libglpk). On Debian-like systems (ex. Ubuntu), |
| 183 | install the package "libglpk-dev". |
| 184 | |
| 185 | To see the output of this mode, run: |
| 186 | lttv -m sync_chain_batch --eval-graphs [usual options, ex: -t traces/node1 -t |
| 187 | traces/node2 --sync ...] |
| 188 | |
| 189 | + Reduction Algorithms |
| 190 | Event analysis yields time correction factors between trace pairs. For groups |
| 191 | of more than two traces, an extra step is necessary to identify a reference |
| 192 | trace and calculate correction factors for each trace relative to this |
| 193 | reference. There are usually many possibilities and so this step is called |
| 194 | "factor reduction". |
| 195 | |
| 196 | ++ Accuracy |
| 197 | At the moment, only one algorithm is available to do this, the "accuracy" |
| 198 | algorithm. This algorithm tries to choose the reference and the factors that |
| 199 | yield the best accuracy. See the function header comments in |
| 200 | factor_reduction_accuracy.c for more details. |
| 201 | |
| 202 | + Design |
| 203 | This part describes the design of the synchronization framework. This is to |
| 204 | help programmers interested in: |
| 205 | * adding new synchronization algorithms (analysis part) |
| 206 | There are already two analysis algorithms available: chull and linreg |
| 207 | * using new types of events (processing and matching parts) |
| 208 | There are already two types of events supported: tcp messages and udp |
| 209 | broadcasts |
| 210 | * using time synchronization with another data source/tracer (processing part) |
| 211 | There are already two data sources available: lttng and unittest |
| 212 | |
| 213 | ++ Sync chain |
| 214 | This part is specific to the framework in use: the program doing |
| 215 | synchronization, the executable linking to the event_*.o |
| 216 | eg. LTTV, unittest |
| 217 | |
| 218 | This reads parameters, creates SyncState and calls the init functions of the |
| 219 | modules to be used. The "sync chain" is this set of modules. At the moment |
| 220 | there is only one module at each stage. However, as more modules are added, it |
| 221 | will become relevant to have many modules at the same stage simultaneously. |
| 222 | This will require some modifications. It is already partly supported at the |
| 223 | matching stage through encapsulation of other matching modules. |
| 224 | |
| 225 | sync_chain_unitest:main() provides a fairly simple example of sync chain |
| 226 | implementation. |
| 227 | |
| 228 | ++ Stage 1: Event processing |
| 229 | Specific to the tracing data source. |
| 230 | eg. LTTng, LTT userspace, libpcap |
| 231 | |
| 232 | Read the events from the trace and stuff them in an appropriate Event object. |
| 233 | |
| 234 | ++ Communication between stages 1 and 2: events |
| 235 | Communication is done via objects specialized from Event. At the moment, all |
| 236 | *Event are in data_structures.h. Specific event structures and functions could |
| 237 | be in separate files. This way, adding a new set of modules would require |
| 238 | shipping extra data_structures* files instead of modifying the existing one. |
| 239 | For this to work, Event.type couldn't be an enum, it could be an int and use |
| 240 | #defines or constants defined in the specialized data_structures* files. |
| 241 | Event.event could be a void*. |
| 242 | |
| 243 | ++ Stage 2: Event matching |
| 244 | This stage and its modules are specific to the type of event. Event processing |
| 245 | feeds the events one at a time but event analysis works on groups of events. |
| 246 | Event matching is responsible for forming these groups. Generally speaking, |
| 247 | these can have different types of relation ("one to one", "one to many", or a |
| 248 | mix) and it will influence the overall behavior of the module. |
| 249 | eg. TCP, UDP, MPI |
| 250 | |
| 251 | matchEvent() takes an Event pointer. An actual matching module doesn't have to |
| 252 | be able to process every type of event. It will only be passed events of a |
| 253 | type it can process (according to the .canMatch field of its MatchingModule |
| 254 | struct). |
| 255 | |
| 256 | ++ Communication between stages 2 and 3: event groups |
| 257 | Communication consists of events grouped in Message, Exchange or Broadcast |
| 258 | structs. |
| 259 | |
| 260 | About exchanges: |
| 261 | If one event pair is a packet (more generally, something representable as a |
| 262 | Message), an exchange is composed of at least two packets, one in each |
| 263 | direction. There should be a non-negative minimum "round trip time" (RTT) |
| 264 | between the first and last event of the exchange. This RTT should be as small |
| 265 | as possible so these packets should be closely related in time like a data |
| 266 | packet and an acknowledgement packet. If the events analyzed are such that the |
| 267 | minimum RTT can be zero, there's nothing gained in analyzing exchanges beyond |
| 268 | what can already be figured out by analyzing packets. |
| 269 | |
| 270 | An exchange can also consist of more than two packets, in case one packet |
| 271 | single handedly acknowledges many data packets. In this case, it is best to |
| 272 | use the last data packet. Assuming a linear clock, an acknowledged |
| 273 | packet is as good as any other. However, since the linear clock assumption is |
| 274 | further from reality as the interval grows longer, it is best to keep the |
| 275 | interval between the two packets as short as possible. |
| 276 | |
| 277 | ++ Stage 3: Event analysis |
| 278 | This stage and its modules are specific to the algorithm that analyzes events |
| 279 | to deduce synchronization factors. |
| 280 | eg. convex hull, linear regression, broadcast Maximum Likelihood Estimator |
| 281 | |
| 282 | This module should return a set of synchronization factors for each trace |
| 283 | pair. Some trace pairs may have no factors, their approxType should be set to |
| 284 | ABSENT. |
| 285 | |
| 286 | Instead of having one analyzeEvents() function that can receive any sort of |
| 287 | grouping of events, there are three prototypes: analyzeMessage(), |
| 288 | analyzeExchange() and analyzeBroadcast(). A module implements only the |
| 289 | relevant one(s) and the other function pointers are NULL. |
| 290 | |
| 291 | The approach is different from matchEvent() where there is one point of entry |
| 292 | no mather the type of event. The analyze*() approach has the advantage that |
| 293 | there is no casting or type detection to do. It is also possible to deduce |
| 294 | from the functions pointers which groupings of events a module can analyze. |
| 295 | However, it means each analysis module will have to be modified if there is |
| 296 | ever a new type of event grouping. |
| 297 | |
| 298 | I chose this approach because: |
| 299 | 1) I thought it likely that there will be new types of events but not so |
| 300 | likely that there will be new types of event groups. |
| 301 | 2) all events share some members (time, traceNb, ...) but not event groups |
| 302 | 3) we'll see which one of the two approaches works best and we can adapt |
| 303 | later. |
| 304 | |
| 305 | ++ Stage 4: Factor reduction |
| 306 | This stage reduces the pair-wise synchronization factors obtained in step 3 to |
| 307 | time correction factors for each trace. It is most useful when synchronizing |
| 308 | more than two traces. |
| 309 | |
| 310 | ++ Evolution and adaptation |
| 311 | It is possible to change/add another sync chain and to add other modules. It |
| 312 | has been done. New types of events may need to be added to data_structures.h. |
| 313 | This is only to link between Event-* modules. If the data does not have to be |
| 314 | shared, data_structures.h does not have to be modified. |