[soac-newgtldapsup-wg] Estimating the cost of continuity capacity

[Eric Brunner-Williams <ebw@xxxxxxxxxxxxxxxxxxxx>]

TDG-Legal Colleagues,

I just looked over the January 2011 numbers for .cat.

The sum of all add, modify, or delete operations on all objects for 
the month was 48012, slightly greater than the number of domains.
The definition of "object" in this registry includes domain objects, 
contact object, and host objects.
I propose that we agree that as a estimating tool, that the sum of all 
operations on all objects, is equal to the number of domain objects in 
the registry.
This means that we ignore the (check + info) operations, .cat had 
1,361,398 of these, the (check only) operations, .cat had an 
additional 1,147,927 of these, and the (info only) operations, .cat 
had an additional 213,471 of these. The reason we ignore these 2.7m 
transactions is that they are non-essential to continuity operations.
These are also read-only operations, and on average complete in half 
the time as add, modify and delete operations.
The reason we need not ignore the add operations is two part. First, 
we know registry's growth rate and can factor out the add operations 
to a first order. In the case of .cat the growth rate is 10k/year, or 
1k/month with an 80% renewal rate. Second, the add, modify and delete 
operations are reported as an aggregate, and the contribution of the 
add operations is less than a quarter of all transactions.
I further propose that we agree that the operations are not highly 
clustered in time, that they occur during 2000 hours of a 8760 hour 
year corresponding to a 5 day work week of 8 hour days, and with 
uniform distribution within those 2000 hours.
If you agree to these two proposals, we may define a continuity 
transactional load in units of 10k domain registrations at the point 
of transition to continuity.
(10k domain objects) x (12 months) = 120,000 transactions/year

120,000 / 2000 = 60 transactions/hour

I propose therefore a continuity transactional capacity definition of 
one transaction/minute for a standard "unit" of 10k domain objects.
It is possible that registries with unlimited admission policies have 
radically different transactional properties. Not wanting to apply 
data from the application of a community-based ("sponsored" in the 
2004 contractual form) admission policy to applications lacking that, 
or any restriction on admission, I propose the 1t/min/unit apply to 
applications designated as "community-based" in the pending evaluation.
If this is not an acceptable definition for continuity transactional 
capacity, please let me know. Again "unit" here is 10k domains.
If we can agree to this, and if we assume that the transactions should 
be completed in 3 seconds or less, for nearly all transactions, then 
the continuity transactional capacity cost can be defined by the cost 
for systems capable of completion of all i/o and computation 
associated with a standard database and read/write operation in 3 seconds.
I will offer estimates on the cost of zone file generation, and name 
server capacity following the same methodology, modulo the caveat 
above on acceptable definition for continuity transactional capacity.
Eric