epa/tr-stat.cc

Go to the documentation of this file.

// Generate statistics from EPA traces 
//   http://ita.ee.lbl.gov/html/contrib/EPA-HTTP.html
//
// All we need to know: 
// 
// (1) client request streams: 
//     <time> <clientID> <serverID> <URL_ID> 
// (2) server page mod stream(s):
//     <serverID> <URL_ID> <PageSize>
//
// Part of the code comes from Steven Gribble's UCB trace parse codes
// 
// $Header: /nfs/jade/vint/CVSROOT/ns-2/indep-utils/webtrace-conv/epa/tr-stat.cc,v 1.3 2005/09/18 23:33:32 tomh Exp $

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <ctype.h>
#include <time.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <tcl.h>

struct URL {
    URL(int i, int sd, int sz) : access(1), id(i), sid(sd), size(sz) {}
    int access; // access counts
    int id;
    int sid, size;
};

struct ReqLog {
    ReqLog() {}
    ReqLog(double t, unsigned int c, unsigned int s, unsigned int u) :
        time(t), cid(c), sid(s), url(u) {}
    double time;
    unsigned int cid, sid, url;
};

FILE *cf, *sf;
double initTime = -1;
double duration = -1;
double startTime = -1;

Tcl_HashTable cidHash;  // Client id (IP, port) hash
static int client = 0;  // client sequence number

static int server = 1;

Tcl_HashTable urlHash;  // URL id hash
static int url = 0; // URL sequence number
static int* umap;   // URL mapping table, used for url sort

ReqLog* rlog = NULL;
unsigned int num_rlog = 0, sz_rlog = 0;

struct Entry {
    char *client;
    unsigned int time;
    char *url;
    int size;
};

static int compare(const void *a1, const void *b1)
{
    const ReqLog *a, *b;
    a = (const ReqLog*)a1, b = (const ReqLog*)b1;
    return (a->time > b->time) ? 1 : 
        (a->time == b->time) ? 0 : -1;
}

void sort_rlog()
{
    qsort((void *)rlog, num_rlog, sizeof(ReqLog), compare);
    double t = rlog[0].time;
    for (unsigned int i = 0; i < num_rlog; i++) {
        rlog[i].time -= t;
        fprintf(cf, "%f %d %d %d\n", rlog[i].time, 
            rlog[i].cid, rlog[i].sid, umap[rlog[i].url]);
    }
    // Record trace duration and # of unique urls
    fprintf(cf, "i %f %u\n", rlog[num_rlog-1].time, url);

    fprintf(stderr, 
        "%d unique clients, %d unique servers, %d unique urls.\n", 
        client, server, url);
}

static int compare_url(const void* a1, const void* b1)
{
    const URL **a, **b;
    a = (const URL**)a1, b = (const URL**)b1;
    return ((*a)->access > (*b)->access) ? -1:
        ((*a)->access == (*b)->access) ? 0 : 1;
}

void sort_url()
{
    // XXX use an interval member of Tcl_HashTable
    URL** tbl = new URL*[urlHash.numEntries];
    Tcl_HashEntry *he;
    Tcl_HashSearch hs;
    int i = 0, sz = urlHash.numEntries;
    for (he = Tcl_FirstHashEntry(&urlHash, &hs);
         he != NULL;
         he = Tcl_NextHashEntry(&hs))
        tbl[i++] = (URL*)Tcl_GetHashValue(he);
    Tcl_DeleteHashTable(&urlHash);

    // sort using access frequencies
    qsort((void *)tbl, sz, sizeof(URL*), compare_url);
    umap = new int[url];
    // write sorted url to page table
    for (i = 0; i < sz; i++) {
        umap[tbl[i]->id] = i;
        fprintf(sf, "%d %d %d %u\n", tbl[i]->sid, i,
            tbl[i]->size, tbl[i]->access);
        delete tbl[i];
    }
    delete []tbl;
}

double lf_analyze(Entry& lfe)
{
    double time;
    int ne, cid, sid, uid;
    Tcl_HashEntry *he;

    time = (double)lfe.time;

    if (initTime < 0) {
        initTime = time;
        time = 0;
    } else 
        time -= initTime;

    // If a trace start time is required, don't do anything
    if ((startTime > 0) && (time < startTime)) 
        return -1;

    // If page size is 0, ignore it
    if (lfe.size == 0) 
        return -1;

    // check client id
    if (!(he = Tcl_FindHashEntry(&cidHash, (const char *)lfe.client))) {
        // new client, allocate a client id
        he = Tcl_CreateHashEntry(&cidHash, (const char *)lfe.client, &ne);
        client++;
        long clientValue = client;
        Tcl_SetHashValue(he, clientValue);
        cid = client;
    } else {
        // existing entry, find its client seqno
        cid = (long)Tcl_GetHashValue(he);
    }

    // only a single server for EPA trace
    sid = 0;

    // check url id
    if (!(he = Tcl_FindHashEntry(&urlHash, (const char*)lfe.url))) {
        // new client, allocate a client id
        he = Tcl_CreateHashEntry(&urlHash, (const char*)lfe.url, &ne);
        URL* u = new URL(++url, sid, lfe.size);
        Tcl_SetHashValue(he, (const char*)u);
        uid = u->id;
    } else {
        // existing entry, find its client seqno
        URL* u = (URL*)Tcl_GetHashValue(he);
        u->access++;
        uid = u->id;
    }

    rlog[num_rlog++] = ReqLog(time, cid, sid, uid);
    //fprintf(cf, "%f %d %d %d\n", time, cid, sid, uid);

    if (startTime > 0) 
        return time - startTime;
    else 
        return time;
}

int get_next_entry(Entry& lfe) 
{
    char buf[1024];

    if (feof(stdin)) 
        return 0;

    fgets(buf, 1024, stdin);
    if (feof(stdin) || ferror(stdin))
        return 0;

    char *tmp = buf, *code, *method, *date;
    lfe.client = strtok(tmp, " ");
    date = strtok(NULL, " "); 
    method = strtok(NULL, " ");     // GET/POST
    *(method++) = 0;
    if (strcmp(method, "GET") != 0) 
        // Only take GET requests
        return -1;

    lfe.url = strtok(NULL, " "); 
    if (strchr(lfe.url, '?') != NULL) 
        // Do not take any url that contains '?'
        return -1;
    strtok(NULL, " ");      // HTTP/1.0
    code = strtok(NULL, " ");   // return code
    if ((atoi(code) != 200) && (atoi(code) != 304)) 
        return -1;
    // last element: size
    tmp = strtok(NULL, " ");
    lfe.size = atoi(tmp);

    // parse date
    // date is from internal string of strtok(), we have to copy it. 
    // What a stupid strtok()!!!!
    tmp = new char[strlen(date)+1];
    strcpy(tmp, date);  
    date = tmp + 1;
    lfe.time = 0;
    date = strtok(date, ":"); // day
    lfe.time = atoi(date);
    date = strtok(NULL, ":"); // hour
    lfe.time = lfe.time*24 + atoi(date);
    date = strtok(NULL, ":"); // minute
    lfe.time = lfe.time*60 + atoi(date);
    date = strtok(NULL, "]");
    lfe.time = lfe.time*60 + atoi(date);
    delete []tmp;

    return 1;
}

int main(int argc, char**argv)
{
    Entry lfntree;
    int      ret;
    double   ctime;

    // Init tcl
    Tcl_Interp *interp = Tcl_CreateInterp();
    if (Tcl_Init(interp) == TCL_ERROR) {
        printf("%s\n", interp->result);
        abort();
    }
    Tcl_InitHashTable(&cidHash, TCL_STRING_KEYS);
    Tcl_InitHashTable(&urlHash, TCL_STRING_KEYS);

    if ((cf = fopen("reqlog", "w")) == NULL) {
        printf("cannot open request log.\n");
        exit(1);
    }
    if ((sf = fopen("pglog", "w")) == NULL) {
        printf("cannot open page log.\n");
        exit(1);
    }

    if ((argc > 4) || (argc < 2)) {
        printf("Usage: %s <trace size> [<time duration>] [<start_time>]\n", argv[0]);
        return 1;
    }
    if (argc >= 3) {
        duration = strtod(argv[2], NULL);
        if (argc == 4) {
            startTime = strtod(argv[3], NULL);
            printf("start time = %f\n", startTime);
        }
    }

    sz_rlog = strtoul(argv[1], NULL, 10);
    rlog = new ReqLog[sz_rlog];

    while((ret = get_next_entry(lfntree)) != 0) {
        // Analyse one log entry
        if (ret < 0)
            continue;
        ctime = lf_analyze(lfntree);
        if ((duration > 0) && (ctime > duration))
            break;
    }
    Tcl_DeleteHashTable(&cidHash);

    fprintf(stderr, "sort url\n");
    sort_url();
    fclose(sf);

    fprintf(stderr, "sort requests\n");
    sort_rlog();
    fclose(cf);

    fprintf(stderr, 
        "%d unique clients, %d unique servers, %d unique urls.\n", 
        client, server, url);
    return 0;
}

---