![]() |
lvs-users
|
| To: | Bradley McLean <bradlist@xxxxxxxxx> |
|---|---|
| Subject: | Re: wlc scheduling broken with servers with weight 0? |
| Cc: | lvs-users@xxxxxxxxxxxxxxxxxxxxxx |
| From: | Julian Anastasov <ja@xxxxxx> |
| Date: | Fri, 1 Nov 2002 20:57:01 +0200 (EET) |
Hello,
On Fri, 1 Nov 2002, Bradley McLean wrote:
> I see I misread the lines below (mixed up dest and least), but
> what the heck is the meaning of loh * dest->weight? Shouldn't
> we be using loh with least and doh with dest?
From ip_vs_wlc.c:
The comparison of h1*w2 > h2*w1 is equivalent to that of h1/w1 > h2/w2
> Rewrite:
> if ( ( atomic_read(&dest->weight) > 0 ) &&
> (loh * atomic_read(&least->weight) >
> doh * atomic_read(&dest->weight)) ) {
> least = dest;
> loh = doh;
Let's see example with the rewritten formula:
2 RS, each with 1 conn:
loh=50
least->weight=2
doh=50
dest->weight=1
50 * 2 > 50 * 1 => Yes
Why the next conn should use dest? least has weight 2 (may be
it is 2 times faster CPU).
> -Brad
Regards
--
Julian Anastasov <ja@xxxxxx>
|
| <Prev in Thread] | Current Thread | [Next in Thread> |
|---|---|---|
| ||
| Previous by Date: | IRC channel, Marc-Christian Petersen |
|---|---|
| Next by Date: | lvs doesn't seem to be clearing/Zeroring, Tim Cronin |
| Previous by Thread: | Re: wlc scheduling broken with servers with weight 0?, Bradley McLean |
| Next by Thread: | Re: wlc scheduling broken with servers with weight 0?, Bradley McLean |
| Indexes: | [Date] [Thread] [Top] [All Lists] |