Ian et al,
There will be one gridmon client at each site and two perfmon boxes at the following sites:
ECDF
Glasgow
Imperial
Lancaster
Manchester
Oxford
QMUL
Cheers Pete
--
----------------------------------------------------------------------
Peter Gronbech GridPP Project Manager Tel No. : 01865 273389
Fax No. : 01865 273418
Department of Particle Physics,
University of Oxford,
Keble Road, Oxford OX1 3RH, UK E-mail : [log in to unmask]
----------------------------------------------------------------------
> -----Original Message-----
> From: Testbed Support for GridPP member institutes [mailto:TB-
> [log in to unmask]] On Behalf Of Ian Collier
> Sent: 10 January 2012 15:18
> To: [log in to unmask]
> Subject: Re: Proposed Server for GridMon and PerfSonar
>
> I take your points about reliability and resilience - all of which can be met with
> an R310 - it supports raid and dual PSUs, but also supports single CPUs and
> modest amounts of RAM.
>
> Extra CPUs and extra memory and all round beefier systems do not just cost
> unnecessary money up front but also consume extra power for the entire
> life of the box.
>
> I am very, very much in favour of spending money to make the service
> dependable, but since the one thing you cannot do with these is put other
> services on them and since you really want two at each site, one for
> bandwidth and one for latency then not over specifying them is just prudent.
>
> --Ian
>
>
> ------------------------------------------------
> Ian Collier [log in to unmask]
> RAL Tier1 Fabric Team
> R89 F18, STFC Rutherford Appleton Laboratory
> DIDCOT OX11 0QX UK
> +44 (0)1235 445440 +44 (0)7866 510075
> ------------------------------------------------
>
> On 10 Jan 2012, at 14:54, Ewan MacMahon wrote:
>
> >> -----Original Message-----
> >> From: Testbed Support for GridPP member institutes [mailto:TB-
> >> [log in to unmask]] On Behalf Of Ian Collier
> >>
> >> Just my tuppence, having been running perfsonar for a few months now,
> but
> >> 4GB is ample. Also I CPU is plenty.
> >>
> > Yup - the only reason for the dual CPUs in my version
> > was that the portal won't do single CPU versions as
> > standard, it's a special option that the pre-canned
> > perfsonar box has, but you can't get on the usual
> > R610. We could ask Dell for a single CPU one, but I
> > don't think E5620s cost very much these days.
> >
> >> Dual PSUs are not a bad idea, but far from essential.
> >>
> > Well, I don't know. On one level the availability of
> > the network is the AND of the availability of all the
> > boxes - no-one can test their link to Oxford if the
> > Oxford endpoint is dead - so we should try to keep them
> > up.
> >
> >> Mirrored disks are more or less essential.
> >>
> >> It looks to me as though the spec discussed is much (bordering on grossly)
> >> over what is required.
> >>
> > With the exception of the dual CPUs, it's about the
> > minimum computational spec currently available, but
> > with the reliability features that a long-term monitoring
> > project needs.
> >
> >> It is public money being spent. Specifying boxes close to the actual load
> >> is really a good idea.
> >>
> > It's a value proposition - I think the benefits of having
> > proper reliable, maintainable server type hardware is
> > worth the relatively small additional cost. You don't have
> > to avoid much downtime or save much staff time before these
> > things will have paid for themselves.
> >
> > Ewan
|