#include <bits/stdc++.h>
using namespace std;
using ll = long long;
const int LG = 41; // enough for n up to ~1e12 (overkill for usual constraints)
int main(){
ios::sync_with_stdio(false);
cin.tie(nullptr);
int n, q;
if (!(cin >> n >> q)) return 0;
vector<pair<ll,ll>> a(n+1);
for(int i=1;i<=n;i++) cin >> a[i].first >> a[i].second;
vector<vector<int>> up(LG, vector<int>(n+1, 0));
// sm as __int128 to avoid overflow
vector<vector<__int128>> sm(LG, vector<__int128>(n+1, 0));
deque<pair<ll,int>> st;
for(int i=1;i<=n;i++){
ll d = a[i].first;
while(!st.empty() && st.front().first < d){
int id = st.front().second; st.pop_front();
up[0][id] = i;
sm[0][id] = (__int128)a[id].second;
}
st.push_front({d, i});
}
while(!st.empty()){
int id = st.front().second; st.pop_front();
up[0][id] = 0;
sm[0][id] = (__int128)a[id].second;
}
// build binary lifting tables, clamp sums to a very large sentinel to avoid wrap
const __int128 INF = (__int128)9e18; // big enough sentinel
for(int k=1;k<LG;k++){
for(int v=1; v<=n; v++){
int mid = up[k-1][v];
up[k][v] = (mid ? up[k-1][mid] : 0);
__int128 s = sm[k-1][v];
if(mid) s += sm[k-1][mid];
if(s > INF) s = INF;
sm[k][v] = s;
}
}
auto jump_k = [&](int x, long long k){
// jump k single-step moves via up[0] powers
for(int b=0; b<LG && x; ++b){
if(k & (1LL<<b)) x = up[b][x];
}
return x;
};
while(q--){
int cur; ll v_ll; cin >> cur >> v_ll;
__int128 v = (__int128)v_ll;
bool finished = false;
while(cur && !finished){
bool moved = false;
for(int k = LG-1; k >= 0; --k){
if(sm[k][cur] <= v){
if(sm[k][cur] == v){
// exactly fills block of size 2^k starting at cur
if(k == 0){
// last filled is cur itself
// cur remains cur
} else {
cur = jump_k(cur, (1LL<<k) - 1);
}
finished = true;
} else {
v -= sm[k][cur];
cur = up[k][cur];
}
moved = true;
break;
}
}
if(!moved) break;
}
cout << cur << '\n';
}
return 0;
}
