Concurrent hits limitation

[Nils-Henner Krueger]

Am 18.01.2010 16:23, schrieb sthaug at nethelp.no:
>> We are using dynamic DNS update and running two scsi disk on
>> RAID-1. We are having  IBM X3650 server with 4 gb RAM.
> 
> The 4 GB of RAM won't help you much (dhcpd for 200K IP addresses
> needs considerably less than that). But two SCSI disks with RAID-1 -
> is this software RAID or do you have a hardware RAID controller with
> battery backed cache? The difference can be dramatic.

What about the following idea:

Use a machine with several gigs of ram and put the dhcpd.leases on a
reasonably sized ramdisk. If you DO guarantee that there won't be a
power outage for both ha members at the same time (by putting the
machines in different data center locations, using uninteruptable power
supply and so on, the usual things), the machine that crashed and lost
the whole leases-db will send an update request to the other cluster
member and get the complete content. This might take a few minutes,
depending on the size, but after that it should be up and working again.
Needless to say the speedup by using a ramdisk is quit good. Comments on
that?

>> We have around 2, 00,000 Ip's. Our lease time is 4 days, Target
>> hits (DORA) will be 5000 per second.
> 
> I find it hard to believe that you'll see 5000 DORAs per second,
> unless all of your clients come online at the same time. Do they?
> 
> As a comparison, we have around 100K IP addresses with a 1 day lease,
> in a failover configuration. *Expected* number of Request/Ack from
> this load is around 100.000 * 2 / 86400 = 2.3 Request/Ack per
> second.

Original poster says "Actually we are in to IPTV services so clients
booting their STB's several times in a day."
But nevertheless that will be magnitudes away from 5000 DORAs/sec.


nhk