[dns-operations] compressing DNS traffic data

Jim Reid jim at rfc1035.com
Wed Dec 8 13:55:41 UTC 2010


On 8 Dec 2010, at 11:51, Ondřej Surý wrote:

> I found this logic quite flawed.  Just because the units of something
> are cheap it doesn't mean that:
>
> 1. cost per unit are the same when you start piling them up.  It's not
> just cost of manhours needed to maintain the storage, it's also the
> electricity, cooling, cost per rack space, interconnections.

If you're talking about petabytes of data and deploying monster SANs,  
these things will of course need to be taken into consideration.  
However those costs are largely fixed. Broadly speaking it costs as  
much to look after 1TB of data as it does to look after 0.1TB of data.  
The costs of storing DNS traffic data are not going to be  
significantly lower because of compression. The overheads you mention  
are going to be the biggest factor and they're not going to be  
influenced significantly by the use of compression. So why bother?

> 2. we need to waste it even if it's cheap.  It's the same flawed logic
> why the US cars had such low mpg ratio - because gas was cheap there  
> was
> no "need for efficency".

This is a flawed analogy Ondřej. The price of storage has been going  
in one direction for decades. So has disk capacity. The ability to  
store bits is almost infinite and the unit cost for doing so continues  
to fall. None of these things hold true for oil.

> If the "compression" ratio is 1:5 or 1:10 compared to raw packets,  
> it will pay of now and in the future.

I think the benefits of compression in this case are dubious and  
probably not worth the effort. YMMV.


More information about the dns-operations mailing list