1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524
|
/*
* Copyright (c) 2009, Raphael Manfredi
*
*----------------------------------------------------------------------
* This file is part of gtk-gnutella.
*
* gtk-gnutella is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* gtk-gnutella is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with gtk-gnutella; if not, write to the Free Software
* Foundation, Inc.:
* 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*----------------------------------------------------------------------
*/
/**
* @ingroup dht
* @file
*
* Stable node recording.
*
* The purpose here is to record the lifetime of nodes that we cache as
* "roots", to be able to compute the probability of a published value to
* still be present on the DHT one hour later (which is the current hardwired
* expiration time in the LimeWire servents and chances to have that changed
* are slim, as of 2009-10-19).
*
* The maths:
*
* The probability of presence of nodes on the network is modelled through
* the following probability density function:
*
* f(x) = 1 / (2*x^2) for x >= 1
*
* with "x" being the number of hours a servent is up. We consider that
* there is a probability of 0.5 that a servent will not stay up for more
* than 1 hour. For those which do, we use f(x), and it is trivial to
* see that the primitive of f is:
*
* F(x) = -1/2x
*
* and that the integral of f(x) between 1 and +oo is F(+oo) - F(1) = 0.5.
*
* Therefore, f(x) is a suitable probability density function fox x >= 1,
* given our hypothesis that P(x >= 1) = 0.5.
*
* Given the limit of F(x) is 0 when x |--> +oo, the integral of f(x) between
* ``a'' and ``+oo'' is -F(a).
*
* This is used to easily compute a (theoretical) conditional probability
* of presence of a node for "h" more hours given that it was already
* alive for H hours (with H >= 1):
*
* P(x >= (H + h))|x >= H = -F(H + h) / -F(H) = H / (H + h)
*
* For the [0, 1] interval, we use the following probability density:
*
* f(x) = 1/2 for x in [0, 1]
*
* Its primitive is:
*
* F(x) = x/2
*
* and the integral of f(x) over [0, 1] is F(1) - F(0) = 0.5
*
* Since f(1) = 0.5 in the two definitions we gave above, the function
* defines a continuous probability density whose integral over [0, +oo] is 1:
*
* f(x) = 1/2 for x in [0, 1]
* f(x) = 1 / (2*x^2) for x >= 1
*
* If a node has been alive "H" hours already (H >= 1), what is the time
* increment h (in hours) that satisfies:
*
* p = P(x >= (H + h))|x >= H
*
* where p is a given probability of presence? This satisfies the equation:
*
* p = H / (H + h)
*
* and solving for h we have:
*
* h = H * (1 - p) / p
*
* Applications:
*
* - for publishing operations, given the set of nodes to which the value
* was stored, and knowing the current stability of these nodes, compute
* the probability of having the published value still there before the
* next publishing cycle. If too low, we can then attempt republishing
* sooner than the default.
*
* - for values we store and for which we have an original, we can determine
* the probability that the publisher remains alive for one more publishing
* period and, if high enough, increase the expiration date for the published
* value, so as to reserve the entry slot in the key (given we have a fixed
* limited amount of values per keys) to the most stable publishers.
*
* - for alive node checks: when a node has been up at least one hour, we
* can compute how much time we can wait before pinging the node again,
* knowing that we want to be sure that it is still likely to be up with
* a probability of "p" (for instance p = 95%).
*
* Implementation:
*
* We keep a table mapping a node's KUID with the set { first seen, last seen }.
* Each time we get a STORE reply (not necessarily OK) from the node, we update
* the "time last seen". We don't care about IP:port changes because the node
* remains in the DHT with the same KUID, hence it will be quickly reachable
* again thanks to the routing table updates. We don't consider KUID conflicts
* a problem here given our application.
*
* For practical considerations, we make the above table persistent so that
* immediate restarts can reuse the information collected from a past run.
*
* When no updates are seen on a given node for more than 2 * republish period,
* we consider the node dead and reclaim its entry.
*
* @author Raphael Manfredi
* @date 2009
*/
#include "common.h"
#include "stable.h"
#include "if/dht/kuid.h"
#include "if/dht/knode.h"
#include "if/dht/value.h"
#include "core/gnet_stats.h"
#include "core/settings.h"
#include "if/gnet_property_priv.h"
#include "lib/atoms.h"
#include "lib/cq.h"
#include "lib/dbmw.h"
#include "lib/dbstore.h"
#include "lib/stringify.h"
#include "lib/tm.h"
#include "lib/override.h" /* Must be the last header included */
#define STABLE_DB_CACHE_SIZE 4096 /**< Cached amount of stable nodes */
#define STABLE_MAP_CACHE_SIZE 64 /**< Amount of SDBM pages to cache */
#define STABLE_UPPER_THRESH (3600 * 24 * 30 * 3) /**< ~ 3 months in s */
#define STABLE_EXPIRE (2 * DHT_VALUE_REPUBLISH) /**< 2 republish periods */
#define STABLE_PROBA (0.3333) /**< 33.33% */
#define STABLE_PRUNE_PERIOD (DHT_VALUE_REPUBLISH * 1000)
#define STABLE_SYNC_PERIOD (60 * 1000)
static cperiodic_t *stable_sync_ev;
static cperiodic_t *stable_prune_ev;
/**
* DBM wrapper to associate a target KUID with the set timestamps.
*/
static dbmw_t *db_lifedata;
static char db_stable_base[] = "dht_stable";
static char db_stable_what[] = "DHT stable nodes";
#define LIFEDATA_STRUCT_VERSION 0
/**
* Information about a target KUID that is stored to disk.
* The structure is serialized first, not written as-is.
*/
struct lifedata {
time_t first_seen; /**< Time when we first seen the node */
time_t last_seen; /**< Last time we saw the node */
uint8 version; /**< Structure version information */
};
/**
* Get lifedata from database, returning NULL if not found.
*/
static struct lifedata *
get_lifedata(const kuid_t *id)
{
struct lifedata *ld;
ld = dbmw_read(db_lifedata, id, NULL);
if (NULL == ld && dbmw_has_ioerr(db_lifedata)) {
g_warning("DBMW \"%s\" I/O error, bad things could happen...",
dbmw_name(db_lifedata));
}
return ld;
}
/**
* Given a node who has been alive for t seconds, return the probability
* that it will be alive in d seconds.
*
* @param t elapsed time the node has been alive
* @param d future delta time for which we want the probability
*
* @return the probability that the node will be alive in d seconds.
*/
double
stable_alive_probability(time_delta_t t, time_delta_t d)
{
double th = t / 3600.0;
double dh = d / 3600.0;
if (0 == t + d || t <= 0)
return 0.0;
if (d <= 0)
return 1.0;
/*
* See leading file comment for an explanation.
*
* NB: this is a theoretical probability, implied by our choice for
* the probability density function.
*/
if (t >= 3600) {
return th / (th + dh);
} else {
/*
* The area under f(x) between 1 and +oo is 0.5
* The area under f(x) between x and 1 is 0.5 - x/2
*/
if (t + d >= 3600) {
double a1 = 1.0 / (2.0 * (th + dh)); /* From t+d to +oo */
double a2 = 1.0 - th / 2.0; /* From t to +oo */
return a1 / a2;
} else {
double a1 = 1.0 - (th + dh) / 2.0; /* From t+d to +oo */
double a2 = 1.0 - th / 2.0; /* From t to +oo */
return a1 / a2;
}
}
}
/**
* Given a node which was first seen at ``first_seen'' and last seen at
* ``last_seen'', return probability that node still be alive now.
*
* @param first_seen first time node was seen / created
* @param last_seen last time node was seen
*
* @return the probability that the node be still alive now.
*/
double
stable_still_alive_probability(time_t first_seen, time_t last_seen)
{
time_delta_t life;
time_delta_t elapsed;
life = delta_time(last_seen, first_seen);
if (life <= 0)
return 0.0;
elapsed = delta_time(tm_time(), last_seen);
/*
* Safety precaution: regardless of the past lifetime of the node, if
* we have not heard from it for more than STABLE_UPPER_THRESH, then
* consider it dead.
*/
return elapsed < STABLE_UPPER_THRESH ?
stable_alive_probability(life, elapsed) : 0.0;
}
/**
* Record activity on the node.
*/
void
stable_record_activity(const knode_t *kn)
{
struct lifedata *ld;
struct lifedata new_ld;
knode_check(kn);
g_assert(kn->flags & KNODE_F_ALIVE);
ld = get_lifedata(kn->id);
if (NULL == ld) {
ld = &new_ld;
new_ld.version = LIFEDATA_STRUCT_VERSION;
new_ld.first_seen = kn->first_seen;
new_ld.last_seen = kn->last_seen;
gnet_stats_count_general(GNR_DHT_STABLE_NODES_HELD, +1);
} else {
if (kn->last_seen <= ld->last_seen)
return;
ld->last_seen = kn->last_seen;
}
dbmw_write(db_lifedata, kn->id->v, ld, sizeof *ld);
}
/**
* Estimate probability of presence for a value published to some roots in
* a given time frame.
*
* @param d how many seconds in the future?
* @param rs the STORE lookup path, giving root candidates
* @param status the array of STORE status for each entry in the path
*
* @return an estimated probability of presence of the value in the network.
*/
double
stable_store_presence(time_delta_t d,
const lookup_rs_t *rs, const uint16 *status)
{
double q = 1.0;
size_t i;
size_t count = lookup_result_path_length(rs);
/*
* We may be called by publish callbacks invoked to clean up because
* the operation was cancelled. Maybe the DHT was disabled during the
* operation, meaning our data structures have been cleaned up? In that
* case, abort immediately.
*
* NOTE: this is not an assertion, it can happen in practice and needs to
* be explicitly checked for.
*/
if (NULL == db_lifedata) /* DHT disabled dynamically */
return 0.0;
/*
* The probability of presence is (1 - q) where q is the probability
* that the value be lost by all the nodes, i.e. that all the nodes
* to which the value was published to be gone in "d" seconds.
*/
for (i = 0; i < count; i++) {
if (status[i] == STORE_SC_OK) {
const knode_t *kn = lookup_result_nth_node(rs, i);
struct lifedata *ld = get_lifedata(kn->id);
if (NULL == ld) {
return 0.0; /* Cannot compute a suitable probability */
} else {
time_delta_t alive = delta_time(ld->last_seen, ld->first_seen);
double p = stable_alive_probability(alive, d);
q *= (1.0 - p); /* (1 - p) is proba this node will be gone */
}
}
}
return 1.0 - q;
}
/**
* DBMW foreach iterator to remove old entries.
* @return TRUE if entry must be deleted.
*/
static bool
prune_old(void *key, void *value, size_t u_len, void *u_data)
{
const kuid_t *id = key;
const struct lifedata *ld = value;
time_delta_t d;
bool expired;
double p;
(void) u_len;
(void) u_data;
d = delta_time(tm_time(), ld->last_seen);
if (d <= STABLE_EXPIRE) {
expired = FALSE;
p = 1.0;
} else {
p = stable_still_alive_probability(ld->first_seen, ld->last_seen);
expired = p < STABLE_PROBA;
}
if (GNET_PROPERTY(dht_stable_debug) > 4) {
g_debug("DHT STABLE node %s life=%s last_seen=%s, p=%.2f%%%s",
kuid_to_hex_string(id),
compact_time(delta_time(ld->last_seen, ld->first_seen)),
compact_time2(d), p * 100.0,
expired ? " [EXPIRED]" : "");
}
return expired;
}
/**
* Prune the database, removing old entries not updated since at least
* STABLE_EXPIRE seconds and which have less than STABLE_PROBA chance of
* still being alive, given our probability density function.
*/
static void
stable_prune_old(void)
{
if (GNET_PROPERTY(dht_stable_debug)) {
g_debug("DHT STABLE pruning old stable node records (%zu)",
dbmw_count(db_lifedata));
}
dbmw_foreach_remove(db_lifedata, prune_old, NULL);
gnet_stats_set_general(GNR_DHT_STABLE_NODES_HELD, dbmw_count(db_lifedata));
if (GNET_PROPERTY(dht_stable_debug)) {
g_debug("DHT STABLE pruned old stable node records (%zu remaining)",
dbmw_count(db_lifedata));
}
dbstore_shrink(db_lifedata);
}
/**
* Callout queue periodic event to expire old entries.
*/
static bool
stable_periodic_prune(void *unused_obj)
{
(void) unused_obj;
stable_prune_old();
return TRUE; /* Keep calling */
}
/**
* Callout queue periodic event to synchronize persistent DB.
*/
static bool
stable_sync(void *unused_obj)
{
(void) unused_obj;
dbstore_sync(db_lifedata);
return TRUE; /* Keep calling */
}
/**
* Serialization routine for lifedata.
*/
static void
serialize_lifedata(pmsg_t *mb, const void *data)
{
const struct lifedata *ld = data;
pmsg_write_u8(mb, LIFEDATA_STRUCT_VERSION);
pmsg_write_time(mb, ld->first_seen);
pmsg_write_time(mb, ld->last_seen);
}
/**
* Deserialization routine for lifedata.
*/
static void
deserialize_lifedata(bstr_t *bs, void *valptr, size_t len)
{
struct lifedata *ld = valptr;
g_assert(sizeof *ld == len);
bstr_read_u8(bs, &ld->version);
bstr_read_time(bs, &ld->first_seen);
bstr_read_time(bs, &ld->last_seen);
}
/**
* Initialize node stability caching.
*/
G_GNUC_COLD void
stable_init(void)
{
dbstore_kv_t kv = { KUID_RAW_SIZE, NULL, sizeof(struct lifedata), 0 };
dbstore_packing_t packing =
{ serialize_lifedata, deserialize_lifedata, NULL };
g_assert(NULL == db_lifedata);
g_assert(NULL == stable_sync_ev);
g_assert(NULL == stable_prune_ev);
/* Legacy: remove after 0.97 -- RAM, 2011-05-03 */
dbstore_move(settings_config_dir(), settings_dht_db_dir(), db_stable_base);
db_lifedata = dbstore_open(db_stable_what, settings_dht_db_dir(),
db_stable_base, kv, packing, STABLE_DB_CACHE_SIZE, kuid_hash, kuid_eq,
GNET_PROPERTY(dht_storage_in_memory));
dbmw_set_map_cache(db_lifedata, STABLE_MAP_CACHE_SIZE);
stable_prune_old();
stable_sync_ev = cq_periodic_main_add(STABLE_SYNC_PERIOD,
stable_sync, NULL);
stable_prune_ev = cq_periodic_main_add(STABLE_PRUNE_PERIOD,
stable_periodic_prune, NULL);
}
/**
* Close node stability caching.
*/
G_GNUC_COLD void
stable_close(void)
{
dbstore_close(db_lifedata, settings_dht_db_dir(), db_stable_base);
db_lifedata = NULL;
cq_periodic_remove(&stable_sync_ev);
cq_periodic_remove(&stable_prune_ev);
}
/* vi: set ts=4 sw=4 cindent: */
|