[Atlas-anchors-pilot] Q&A, part 1

[Gert Doering]

Hi,

On Tue, Sep 18, 2012 at 01:43:31PM +0200, Daniel Karrenberg wrote:
> > If there's enough interest in building a "RIPE NCC in a box" system,
> > then please stop calling it "Atlas Anchor", because that's not what it
> > would be.
> 
> from where I stand what you are doing is to ask us to operate a
> high quality service on cheap kit. That appears kind of one-sided
> to me, especially in a pilot situation where requirements are fluid
> and operational experience is scarce. Maybe the pilot is not for
> you and you should wait until we move to production. It would be
> sad, since your requirements significantly influenced the Atlas
> Anchor design.

What I'm asking you is "build a reasonable box that does what you need
it to, not 100 times more".  For a box that answers probes, anything you
can buy today will be able to saturate a 100M or even GigE link - and
you can get that (*with* DRAC) for much less money.

I just can't see the need for 4 expensive SAS disks, 32 Gigs of RAM,
and a 6-core high-end CPU for that.

Now, I'm not asking you to run this on 50EUR TL-1043NDs, as I'm not really
sure whether that one's SoC would actually be able to saturate a GigE
link (and the 1043 has no serial console or ilo/drac, which is a no-go 
anyway) - but there's middle ground here, and requirements should be 
reasonable.

If I go to my CEO and tell him "the TTM boxes running fine on 10-year-old
hardware with a 10-year-old-CPU and less than 100W power requirements are
discontinued, instead we have to purchase a machine with lots of disks,
more lots of CPU, plenty of RAM, and 350+W power usage", he'll ask me
"ok, so what are they going to do with it?".  If I then say "ping it",
he'll throw me out of his office.


> Details:
> 
> Yes, one intention of the pilot is to evaluate virtualisation for
> providing multiple services. 

Call me old-fashioned and narrow-minded (and maybe I am), but this is 
fairly uninteresting for us for the target "TTM replacment", and will 
drive the hardware requirements high.  

Worse, virtualization might actually harm your measurements, as precision 
timing and ICMP/UDP burst handling are known issues in virtualized 
environments - we moved our recursive DNS *away* from our VM 
infrastructure because it kept losing packets.

> The pilot hardware cost of slightly
> more than 3000EUR is quite reasonable considering it includes a
> hardware service contract. 

I'm fine with hardware service contracts, and remote management boards,
but about 1000EUR of that is "4x 600G SAS disks, 32G RAM, 6-core", and I 
really do not see the need for that in an Anchor box.

> Reliably operating a large number of
> widely distributed remote systems at a reasonable cost is a goal
> that is directly opposed to using the cheapest available hardware
> and economising on maintenance. We cannot have both and we have
> done our best to strike a balance. It is likely that after evaluation
> of the pilot results we will offer a smaller hardware alternative
> that can only do one service. But do not expect it to be in the
> order of 50EUR. I understand that your motivation for requesting
> an Atlas Anchor is that we take full responsibility to operate it.

True.  I'm not expecting 50 EUR (as explained above), but it serves
as as reality check: if all you want is a reliable box that answers
up to 100Mbit/s of packets, what do you need to achieve that?

> If the cost of joining the pilot turns out to be too high for you
> that would be a shame, especially since your requirements determined
> much of what we aim to achieve with RIPE Atlas Anchors. But maybe
> then the "one service only" version that we will likely offer after
> the pilot in Q2/2013 may be acceptable for you. Please re-consider
> to join the pilot, maybe with only one box.

The "one service only" box sounds what we'd like to have.

Since others have indicated interest in the "give me all of it!" box
already, I think you have the necessary amount of pilot hosts already,
so I'll patiently wait...

regards,

Gert Doering
        -- NetMaster
-- 
have you enabled IPv6 on something today...?

SpaceNet AG                        Vorstand: Sebastian v. Bomhard
Joseph-Dollinger-Bogen 14          Aufsichtsratsvors.: A. Grundner-Culemann
D-80807 Muenchen                   HRB: 136055 (AG Muenchen)
Tel: +49 (89) 32356-444            USt-IdNr.: DE813185279
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 306 bytes
Desc: not available
Url : https://www.ripe.net/ripe/mail/archives/atlas-anchors-pilot/attachments/20120918/a3c79d83/attachment.bin