I don't want to sound discouraging butpersonally I'd avoid introducing
something alternative that is not used by the WLCG community, it's the
reason gridmon wasn't a good solution and even though I agree the
perfsonar hardware requirements are a bit excessive and I think 1 box is
enough, they cannot be an issue in the UK because we already have it.
Proposing this as an alternative to WLCG can be done but without
bandwidth measurements it will not go far.
cheers
alessandra
On 29/01/2014 12:04, Jeremy Coles wrote:
> Dear All,
>
> Some institutes may already have such a probe.
>
> We would need to put together a small cost/benefit proposal if seeking to request a collaboration purchase, and then we may be asked whether sites could not have purchased these with the last round of hardware purchases (i.e. the cost is in the noise of money already awarded). perfSONAR boxes were funded already.
>
> GridMON was a bespoke solution and we struggled to maintain it… let alone get others to adopt is as a standard.
>
> I can anticipate some of the replies to this message but best that they come from the sites than from me.
>
> Jeremy
>
>
>
>
> On 29 Jan 2014, at 11:32, Duncan Rand wrote:
>
>> Perhaps we could start with 20 probes, one for each GridPP site - that would be a more reasonable £600.
>>
>> Duncan
>>
>> On 28 Jan 2014, at 18:22, Ewan MacMahon wrote:
>>
>>> Hi all,
>>>
>>> I said in the ops meeting that I'd put together a quick summary
>>> about the RIPE Atlas network probes, so here it is. The documentation
>>> on the RIPE site[1] is pretty good, so I'll keep this relatively
>>> brief[2].
>>>
>>> Firstly, 'RIPE Atlas' is an 'Atlas' in the sense that it's a 'mapping'
>>> project, designed to survey the internet (so nothing to do with our
>>> ATLAS[3]), and 'RIPE' as in the European Internet Protocol Registry,
>>> though the project isn't restricted to Europe and there are probes all
>>> round the planet.
>>>
>>> The individual probes fulfil a function similar to our PerfSonar boxes,
>>> in that they provide endpoints to run tests from. A key difference is
>>> that the tests are much more basic, and all centred around latency and
>>> functionality, not at all around bandwidth. Accordingly the probe
>>> hardware is much cheaper, being a repurposed TP-Link router[4]. There's
>>> a photo of our one sat on top of (and USB powered from) our PerfSonar
>>> boxes here:
>>> http://www-pnp.physics.ox.ac.uk/~macmahon/ripe_atlas/RIPE-Atlas-probe-UKI-SOUTHGRID-OX-HEP.jpg
>>>
>>> As a matter of routine a probe will do various standard measurements,
>>> mostly pings and traceroutes to static targets like the DNS root servers
>>> and dedicated RIPE Atlas Anchor nodes. By hosting a probe, you get to
>>> see the results of its routine measurements which given some idea of
>>> your connectivity to the rest of the internet.
>>>
>>> There is also a credits system - hosting a probe accumulates credits for
>>> as long as it's up and working, and you can then spend them on having
>>> the system run your own measurements, either on a one-off or recurring
>>> basis. For example, we had a slight question the other day whether our
>>> web server was unreachable from some places, so I did a quick one-off
>>> test of having a random 300 probes ping it. There's a screenshot of a
>>> graphical view of the results here:
>>> http://www-pnp.physics.ox.ac.uk/~macmahon/ripe_atlas/ripe-atlas-example.png
>>> (it turned out to be fine)
>>>
>>> Most of the probes are handed out free of charge, just by asking for one
>>> (which is how we and QMUL got ours). That's possible because of some
>>> organisations sponsoring the costs. The key advantages of being a
>>> sponsoring organisation are[5] that you get to see the routine
>>> measurements from all of the probes you've sponsored, that you get credits
>>> from all the probes you've sponsored (which can be spent on measurements
>>> using any of the probes, yours or not), and the RIPE folks add you to
>>> their publicity, which given that they're one of the main top-level
>>> internet co-ordination organisations, has good impact[6]. I don't know
>>> what it actually costs to sponsor a probe, but I can email the RIPE
>>> contact address and ask if we think we're interested, but taking the
>>> ~£30 retail price of the probe as a rough ballpark, spending (say)
>>> £10,000 (which isn't a huge amount in the context of the whole GridPP
>>> project) would get us about three hundred probes to distribute around
>>> our own sites, the rest of the UK academic network, and anywhere else
>>> we wanted to put them. Given that there's a total of just under five
>>> thousand probes currently on the whole system, I suspect we could go
>>> for a smaller batch if we wanted to.
>>>
>>> So, to summarise, the RIPE Atlas system has some similarities to
>>> PerfSonar, but in practice it fills a different niche, and they
>>> complement each other well. I think there is some value in our getting
>>> involved, and I think we should seriously look into the possibility of
>>> becoming a sponsor to do so.
>>>
>>> Ewan
>>>
>>> [1] https://atlas.ripe.net/
>>> [2] I could have been even less brief.
>>> [3] We can only be grateful they didn't call it 'Dirac'
>>> [4] One of these: http://www.amazon.co.uk/TP-Link-TL-MR3020-Portable-Wireless-Router/dp/B00634PLTW
>>> [5] https://atlas.ripe.net/get-involved/become-a-sponsor/
>>> [6] IMPACT!
--
Facts aren't facts if they come from the wrong people. (Paul Krugman)
|