You are here: Home > Publications > RIPE Labs > Daniel Karrenberg > A RIPE Atlas Probe for Every RIPE NCC Member

A RIPE Atlas Probe for Every RIPE NCC Member

Daniel Karrenberg — 02 Nov 2011
RIPE Atlas has made steady progress in its first year. But we have more ambitious plans. Please read below how we are suggesting to achieve them and why we need your support.

RIPE Atlas has made steady progress in its first year: we have more than 850 active probes right now, we have collected a lot of static measurements and deployed the first 'maps' visible to everyone. More maps are coming soon. These maps will be available to you and the RIPE community. They will provide an unprecedented level of situational awareness about the state of the Internet in our region and beyond. For example, here is a map showing the DNS root name sever with the lowest ping time from each probe:

Figure 1: 'Closest' DNS Root Name Server
Click here for a live version!

 
And here is an example of what is coming: a map of the anycast instances of K-root used by all probes based on DNS test queries:
 
Figure 2: Which probes see which instance (colour coded) of k-root?
(purple: ams-ix, green: denic, red: linx, yellow: nap, white: tokyo, blue: other)

Note that these are mere first examples. We are about to release the first beta of the 'user defined measurements' feature. This will allow you, as a probe host or as a RIPE Atlas sponsor, to run measurements that are of specific interest to you. For instance, you could monitor the reachability of your network or of a specific server from dozens of probes all over the world. This will augment analysis of incoming traffic with actual user views: the outside perspective onto your network. User-defined measurements will make it even more attractive to sponsor and host RIPE Atlas probes.
 
The results of the user defined measurements are of course not only available to the user that requested them, but also be added to the vast pool of all measurements. This way the results of user defined measurements are useful for the community at large and not just the individual RIPE Atlas hosts or sponsors.
 
 
 
 
The data we are already collecting now is a gold mine for understanding what is happening throughout the entire network. We have demonstrated that we are on the right track: the idea works, the architecture is sound and the potential is enormous. We are at the threshold of bringing you a totally new awareness about the state of the network in our region. For instance you will be able to know with certainty that it is not just you who has performance issues with a certain destination but everyone, or everyone from your geographic area. Since the information comes from the RIPE NCC, an independent source, it will also be very convincing.
 

Current Plans

We will reach 1,024 active probes by the end of the year. We have budgeted for twice this amount by mid-2012 and to double it again to 4,096 active probes by the end of 2012. This is achievable by steady development with the currently budgeted level of resources and investment in new probe hardware. Both RIPE Atlas probe hosts and RIPE Atlas sponsors will enjoy an increasingly powerful platform to run their own measurements. The community will benefit from the results of all of these measurements via a new and interesting 'map' product every 1-2 months. We will also provide more and better information about RIPE Atlas for the community and we will help specific communities form around RIPE Atlas (such as probe hosts and sponsors). All this is based not only upon our widely recognised expertise in the field, but also on the unique, neutral and impartial position of the RIPE NCC. At the same time it reinforces that very position.
 

Is it Enough ?

 

It is quite tempting to be content with our achievements and the useful things coming out of the pipeline. However, we have to realise that this still leaves us far from the original vision which is to provide a comprehensive real-time picture of the RIPE NCC service region Internet. There are currently more than 15,000 Autonomous Systems (ASes) active in the RIPE NCC service region originating more than 67,000 IPv4 prefixes. 4,096 probes cannot provide a comprehensive picture for such a big part of the network. Of course, we can do our best to distribute probes cleverly, but we will not reach a comprehensive picture for a long time.
 
 
Figure 3: Simulation of 5,000 randomly distributed probes in the RIPE NCC service region.
Colouring is random everywhere but in Estonia.

 
There clearly is room for a more ambitious approach. But exactly how much more ambitious should we be? The number of ASes gives a first indication that it should be 15,000 or more. Clearly there are some very large ASes that would need more than one probe in order to make the picture complete. Another indication is buying power for probe hardware. In discussions with several suppliers for probe hardware, it has become clear that there is a significant threshold for pricing at around 20,000 probes. More importantly, this threshold also exists for customisation of hardware; at this level of commitment it becomes possible to negotiate hardware designs specific for our application.
 
 
 
 
Figure 4: Simulation of 20,000 randomly distributed probes in the RIPE NCC service region.
Colouring is random everywhere but in Estonia.
 

A More Ambitious Approach

Based on these considerations, I propose a more ambitious plan: let us deploy 20,000 RIPE Atlas probes within the next 18 months. This way we can give each RIPE NCC member at least one probe. Members in larger charging categories will receive additional probes in order to provide additional coverage in big ASes, as will RIPE Atlas sponsors. Let us also increase our efforts to make the raw measurement results more easily accessible via APIs, such that it becomes easier to integrate them into member's systems and presentations. There are numerous new benefits to the community that will be possible with this number of RIPE Atlas probes: fine-grained SSL 'perspectives', much improved address hijacking detection and many more. We will also be able to accelerate the integration of other active measurement services such as RIPE TTM and DNSMON and the correlation of active measurements with the routing measurements from RIS. In this ambitious approach we will also have resources to improve cooperation with other active measurement efforts while maintaining and extending our lead in this area. We will strengthen the position of the RIPE NCC as an impartial and trusted source of data about the Internet in the RIPE NCC service region.
Figure 5: Simulation of 10,000 randomly distributed probes in the RIPE NCC service region.
Colouring is random everywhere, but probes routed via a very large AS.

 
The additional cost for this more ambitious plan is EUR 1700k in 2012 and approximately EUR 800k in the years thereafter. EUR 1000k of the initial cost is an investment in 20,000 additional probes and their deployment. While this may sound large in absolute terms, it represents less than 10% of the current budget and also less than 10% of the current reserves of the RIPE NCC association. This is an engineer's view of the cost: it is the money that we need to spend. On the balance sheet this will look better because we are buying and producing assets while we spend the money. So, as an association we can afford to realise this ambitious plan. All we need is the courage to do so and the vision of this unique tool we are building and the benefits that it will bring.
 

The Way Forward

I understand that right now you can only see the contours of the services and the usefulness of RIPE Atlas. Maybe we want to delay the decision to proceed for a few months until we have cashed in on the current developments, until more probes are online, user defined measurements are deployed and a larger number of map products are available to all.
 
 
 
Delaying another full year however, would be too much. RIPE NCC members who agree that we should be more ambitious developing RIPE Atlas in 2012 should speak up about it. I am not asking for carte blanche, just for more resources than currently planned to become available as we show results.
[The wording of the preceding paragraph has been changed after initial publication.]

8 Comments

Hank Nussbacher says:
02 Nov, 2011 03:30 PM
I am totally against spending another 2.5MEuro on this. Ask yourself what problem are you trying to solve in the Internet that needs 2.5MEuro. Things like this are in the category of "nice to have" but not worth that level of spending.
Alexander Isavnin says:
02 Nov, 2011 07:55 PM
Could you explain mo detailed, how it's planned to spend mentionrd E1700k and E800k?
Daniel Karrenberg says:
03 Nov, 2011 03:06 PM
Alexander,

there is no 'plan' to spend any money yet. I hope that such a
plan will be made once this discussion shows sufficient support.
So much for the formal answer.

Then here is what believe you want to know: For the purpose of starting
the discussion I made a back-of-the-envelope calculation of the
additional resources needed. I am happy to share this. But please
consider that it is a very rough calculation that aims to be
conservative. Also I repeat that these are not directly translatable
into budget figures for a number of reasons like the possibility to
write off the costs for assets over several years.


Per probe:

                                        FTEs Expenditure per Probe
Probe Hardware 400000 20
Probe Distribution 200000 10
First Line Support 2 100000 5
Controller & Database Infrastructure 150000 7.5
Operators 2 150000 7.5
                                                ------- ---
                                                1000000 50
                                                =======

General:

Infrastructure Development 2 150000
Map Products etc. 2 150000
Liaison & Dev. w Academics & others 1 90000
Product Support 2 150000
Management 1 120000
                                       -- -------
                                       12 660000
                                       == =======


The 800k/y figure for maintenance assumes that most of the people
remain necessary after the 20k deployment. That is very rough of course.
The error margin here is high, say 20%. My point is that it will cost
a non-trivial amount of money, but it does not make sense to
invest energy in calculating it in detail before we know that
there is enough support to make a detailed plan.

Once we have reached that stage, the staff will draft a more detailed
budget and ask for approval by the executive board. This could be as
part of the annual activity plan and budget cycle or in between if the
urgency is agreed. How the board involves the membership it up to them.
But the decision is with the board.

Let me know if I have not answered the whole question.

Daniel
Daniel Karrenberg says:
03 Nov, 2011 04:55 PM
A better formatted version of this response is at
https://labs.ripe.net/[…]/view

Daniel
Nikolay Todorov says:
19 Nov, 2011 04:47 PM
Hi there,
i registered for a probe from here: http://atlas.ripe.net/pre-register/ a couple of days ago. Is this the right place and if not, could you please tell me where i should register.
Mirjam Kühne says:
20 Nov, 2011 10:53 AM
Hi Nikolay. Yes, this is the correct place. Please note that it can take a while until you receive your probe depending on your location and network details.
Nikolay Todorov says:
24 Jan, 2012 01:22 PM
Hi, it's me again. I was wondering if there is any way to know when i'm going to receive a probe. And one more thing - am i supposed to receive an email when i register for a probe, because i didn't.
Ronen Preisler says:
26 Jan, 2012 02:29 PM
Hi Nikolay,
I would like to confirm that we have received your probe application in good order. We have now contacted you via email regarding your application. Apologies for the waiting time, there is currently a small backlog of applications and we try to respond to all applicants as soon as we can.
Regards, Ronen
Add comment

You can add a comment by filling out the form below. Comments are moderated so they won't appear immediately. If you have a RIPE NCC Access account, we would like you to log in.