Server Sizing Question

[David A. Evans]

>        t = 12500 (an hour and 20 minutes?  that's an odd choice)
        "t" in your equation winds up being an average of our different 
lease times for our different scopes (ip phones, PC subnets, Wireless... 
plus reboots, chatty clients..  etc) as we are seeing 300,000+ queries per 
day with our current solution in this environment.   I think the math 
still works out to what I was looking for. 
        Thanks for your help.


David A. Evans
Enterprise IP/DNS Management
Global IT Distributed Infrastructure Operations Division
dnsadmin at cat.com



"David W. Hankins" <David_Hankins at isc.org> 
Sent by: dhcp-users-bounce at isc.org
10/10/2006 01:23 PM
Please respond to
dhcp-users at isc.org



To
dhcp-users at isc.org
cc



Subject
Re: Server Sizing Question




Caterpillar: Confidential Green Retain Until: 11/09/2006        Retention 
Category:  G90 - General Matters/Administration


On Tue, Oct 10, 2006 at 11:00:41AM -0500, David A. Evans wrote:
> a lot of differing opinions on it.    What I have not been able to find 
is 
> information on the physical size of  leases file and how much data the 
> DHCP service logging is going to add to the syslog (or wherever its put) 

> for a given address pool and lease time.  Does anyone have a link to 
that 
> information or a good estimate from their experience as to what I might 
> expect in the below environment?  Does going to a failover setup add a 
> noticeable percentage to either of these numbers?

Failover adds some additional timers to each lease.  It's not double,
but it is an increase.

> 60,000 available address
> 350 scopes
> 25,000 clients
> approximately 300,000 queries per day(average 4 queries per second)

Assume approx ~500 bytes per lease, with failover.  It varies, and it
depends on your enviroment and config.  Total size will look something
like:

                 500 * (n + (c * (3600 / (t/2))))

n: number of leases
c: number of active clients.
t: lease-time, t/2 = renewal time.
3600: period at which the lease-file is rewritten (not configurable).

Plug in what values you supplied and we solve for:

                 30000000 + (45000000000 / (t/2)))

                 4 requests/s = 25000 clients / (t/2)

                 t = 12500 (an hour and 20 minutes?  that's an odd choice)

                 Est size: 37,200,000


In failover this gets a little more problematic because the 'initial
lease time' is MCLT.  It's impossible to guess how many of your
clients are going to be in the initial state, how many will be renewing,
and even how many will get 'MCLT optimal lease times' because their
potential-expiry is not yet up (so their initial lease time might be
somewhere between MCLT and lease-time).

It's probably reasonable to assume the worst case and substitute MCLT
for lease-time.


But: this is still an estimate.  If a 'bad' client out there is
renewing like mad (or churning through addresses due to some bug),
you'll use 500 bytes for each transaction with the client.  When
the disk fills and the write fails, you're toast unless removing
the dhcpd.leases~ temporary file frees up enough for a complete,
new file.


As for syslog, I generally just leave that up to log rotators.

-- 
ISC Training!  October 16-20, 2006, in the San Francisco Bay Area,
covering topics from DNS to DDNS & DHCP.  Email training at isc.org.
-- 
David W. Hankins                 "If you don't do it right the first time,
Software Engineer                                you'll just have to do it 
again."
Internet Systems Consortium, Inc.                -- Jack T. Hankins