Hi Greig and Jeff,
> The FTS moves files between site storage elements (specifically SRMs),
> so that is a dependency.
Indeed, I now fixed that one, see below!
Also, as I understand it, the LFC service does not depend on the BDII,
(even though some of the information in the IS refers to the LFC)
I'm also going to follow Jeff's advice and dig this a bit further...
btw. (and food for thought)
I don't know if anyone is involved in the area, but it is interesting to
note that the BDII service as we know it, provides loose integrity,
and therefor it can be a generous source of errors due to caching effects.
In fact on the BASE vs ACID axis (in accordance with Jim Gray's terminology),
BDII is currently too much "BASE", perhaps unsuitable for low-latency
and low-fault-tolerant applications like in urgent computing (tsunamis etc).
btw2.
I believe the SPOF network card for the LCG grid,
could very well lie somewhere among CERN, GOCDB and Tier-1s :-P
I thank you both for your comments,
Fotis
O/H Fotis Georgatos έγραψε:
>
>
> Dear all,
>
> I am trying to find a way to describe the different aspects
> of interconnection among various grid components of the gLite stack.
>
> For this reason I have made an attempt to create a Dependability Matrix,
> which albeit preliminary includes some factual aspects of the system,
> or at least tries to present a probable draft picture of it,
> assuming four base services (IP=network service, DNS, NTP, BDII):
>
> IP DNS NTP BDII GSI RB LFC UI SE WN CE FTS/FPS
> Depends on:
> IP - X X X X X X X X X X X
> DNS - X X X X X X X X X X
> NTP - X
> BDII - X X X X X
>
> GSI - X X X X X X X
> RB - X
> LFC - X X
> UI -
> SE X - X
> WN - X
> CE X X -
> FTS/FPS -
>
> First, let me readily recognize that the matrix is perhaps incomplete,
> if not at cases misleading :) (eg. WN can be as complex as UI)
> Also, I have deliberately omitted services like VOMS, CA, MYPROXY,
> although I am interested in them and intend to include them later on.
>
> Here is my question now:
>
> Is there any definite guide which could help us in finding out
> the exact dependencies of all the middleware components, reliably?
> Maybe there is something wrong up in that matrix, but who is to tell?
>
> I know the ETICS project is doing a mapping of software components
> and their dependencies, as seen in a single system installation,
> but in this case I'm looking for the whole distributed system's relations!
>
> btw.
> If anybody is doing some related work, or any following keyword rings a
> bell:
> PRA, FTA, FME(C)A, ETA, ESD, HRA, MLD, CCF and friends.... please
> contact me!
>
> ps. - ref.
> Probabilistic Risk Assessment Procedures Guide for
> NASA managers and Practitioners, Dr. Michael Stamatelatos, August 2002
>
> thank you in advance,
> and enjoy your vacation for the ones still on the go :),
>
> Fotis
>
|