An incomplete PQ answer

[Ian Batten]

>
> we're now sending 1.5E9 text messages/week...  if you need to record
>
>        time, date, [4]
>        source number,  [10]
>        destination number,  [10]
>        geographic location of source and destination [10?]
>        and the byte-length [1]
>        (but not of course the content!)



Probably about 4TB per year for the CDRs, 16TB per year if you wanted =20=

to store the messages as well (although they're not all 160 bytes and =20=

the text will compress).  For voice and mobile, were I in the office =20
and it not 0045 I could get the volumes from our regulatory wonks, but =20=

as a rough guess let's size it for 100m subscribers each making 10 =20
calls per day, so 7E9 CDRs per week each of about the same size, so =20
perhaps 20TB/year.   Can we agree to a total of 40TB/year to store all =20=

CDRs for voice plus the full text of every SMS plus 10% for luck?

The underlying cost of the storage is around $1000/TB (based on an EMC =20=

AX4 SATA array).   None the less, if you gold plate, replicate, gold =20
plate again, plus government mark up, have twenty consultants dance =20
around the rack, you're not going to spend more than =A35000/TB capital, =
=20
and as you can put the a couple of hundred TB your operating costs =20
will be trivial once you've paid for the electricity (about 50W/TB, so =20=

say 1kWH/day/TB).

So you can buy 40TB usable (say 80TB raw if you're replicating) for, =20
at the absolute, ludicrously inflated, maximum of =A3200K (I bought 40TB =
=20
a couple of years ago, based around 500GB SATA spindles, plus some =20
exotic NAS heads and a load of replication software, and it cost less =20=

than half that, so my contention that =A35K/TB includes replication =20
seems reasonable), and it'll cost at the outside =A34K/year to power --- =
=20
that's assuming no spin-down, which is obvious nonsense.    It'll fit =20=

in half a rack.

So why all the extra cash...?

ian=