Load balancing issue on busy /23 networks (v4.2.5)

[Nicolas C.]

Hello,

I have two DHCP servers (v.4.2.5) in failover mode and on some networks 
I have issue with load-balancing :

Mar 18 13:40:28 primary dhcpd: balancing pool b8e1ea38 172.20.24.0/23 
total 505  free 128  backup 190  lts -31  max-own (+/-)32
Mar 18 13:40:28 primary dhcpd: balanced pool b8e1ea38 172.20.24.0/23 
total 505  free 127  backup 191  lts -32  max-misbal 48

Mar 18 13:40:28 secondary dhcpd: balancing pool b9238a10 172.20.24.0/23 
  total 505  free 302  backup 17  lts -142  max-own (+/-)32  (requesting 
peer rebalance!)
Mar 18 13:40:28 secondary dhcpd: balanced pool b9238a10 172.20.24.0/23 
total 505  free 302  backup 17  lts -142  max-misbal 48

On busy hours, on the secondary, the "backup" reaches "0" and the server 
makes "peer holds all free leases" errors.

On others (less busy) networks, everything works fine :

Mar 18 14:30:07 secondary dhcpd: balancing pool b91847d8 172.21.4.0/23 
total 505  free 142  backup 214  lts 36  max-own (+/-)36
Mar 18 14:30:07 secondary dhcpd: balanced pool b91847d8 172.21.4.0/23 
total 505  free 145  backup 211  lts 33  max-misbal 53

Mar 18 14:30:07 primary dhcpd: balancing pool b8d6a800 172.21.4.0/23 
total 505  free 145  backup 211  lts -33  max-own (+/-)36
Mar 18 14:30:07 primary dhcpd: balanced pool b8d6a800 172.21.4.0/23 
total 505  free 142  backup 214  lts -36  max-misbal 53

This is the failover configuration on the primary :

failover peer "pmf_pth" {
         primary;
         address 172.20.0.2;
         port 647;
         peer address 172.24.0.2;
         peer port 647;
         max-response-delay 30;
         max-unacked-updates 10;
         load balance max seconds 10;
         mclt 1800;
	split 128;
	auto-partner-down 900;
}

What could be the cause of the unbalance?

   * Corruption of the leases database? If yes, should I stop the 
secondary and delete the "dhcpd.leases" file to rebuild it?

   * Clients not willing to switch DHCP server, forcing on server to 
have to many leases? Our Windows clients have Symantec Endpoint 
Protection 12 with the "Smart DHCP" activated.

   * Issue with our switches and some DHCP-related security settings?

   * All of the above?

I tried to increase the "load balance max seconds" value to 10 seconds 
with no success. I have seen lots of messages about the "peer holds all 
free leases" but very few diagnostics and solutions.

Regards,

Nicolas