From: Lawrence T. Hoff [mailto:[email protected]]
Sent: Tuesday, January 26, 2010 6:09 AM
To: 'Dalesio, Leo'; 'Jeff Hill'
Subject: RE: compact PCI versus VME
I see an interesting theme in
this discussion - although I
am sensitized by "internal" (RHIC) discussions
we had last week....
My thoughts this morning
(for what they are worth):
These days, accelerator controls
seems to include two somewhat
distinct disciplines:
1) "slow controls" - digital I/O, PLC-based logic, analog I/O
at the ~10Hz or slower level,
vacuum, cryo, etc.
2) Data acquisition (DAQ) systems - including global orbit correction
at the many KHz level,
turn-by-turn (and even bunch-by-bunch) beam
diagnostics, LLRF, etc.
The former benefits from low "per-channel" costs, modularity,
scalability, long lifetime,
etc. This has been reasonably served by VME
historically. It is hard to
imagine, e.g. that uTCA would be well suited
for such applications. DAQ
systems requires high bandwidth, large memory,
powerful processors (or
FPGAs).
To
date, we (RHIC) have "made do" with VME for *both* roles -
occasionally bumping into VME
limitations for the most demanding
applications, but for the most
part happy to have a one-size (mostly)
fits all solution.
Recently, we debated investigating non-VME crate solutions (e.g. uTCA) -
Still trying to maintain a
one-size-fits all solution. Our concern was whether
we could find a platform with
the bandwidth needed for even more demanding
DAQ systems moving forward, but
still retain the low per-channel costs
for "slow controls"
(hence my interest in Jeff's observation about low-cost
CompactPCI modules).
One thought was to investigate whether (PMC?) daughter-cards could be used
for "slow controls"
(to control per-channel costs) - only using the (expensive,
switched serial) backplane for
high-bandwidth applications.
Another thought - what Bob seems to be suggesting - is to give up on
the one-size-fits-all solution
- instead select platforms optimized for
DAQ applications and other
platforms optimized for "slow control" applications.
RHIC is de-facto heading in that direction - continuing to use
VME32 (as well as NI Compact
RIO and inexpensive PLCs) for "slow controls",
and relying on a newly
developed crate-less solution for DAQ applications -
based on a XILINX FPGA with PPC
hard core, and 6 XMC daughter card sites
for modularity.
It isn't obvious to me that CompactPCI is significantly better
(or worse) than VME for
"slow controls" - but I am still interested
to hear other's experience. I
suspect that Compact PCI bandwidth is
*not* so significantly
better than VME that it would satisfy our
demanding DAQ needs moving
forward (but I am willing to be corrected).
I strongly suspect that our DAQ
needs can only be met by a switch-serial
backplane standard *or*
a crate-less solution such as our FPGA-based
system - but again, I'm all
ears and ready and willing to be corrected.
-- Larry
From: Dalesio, Leo
[mailto:[email protected]]
Sent: Tuesday, January 26, 2010 6:57 AM
To: Hoff, Lawrence; Jeff Hill
Subject: RE: compact PCI versus VME
Sent to Eric regarding cPCI for timing:
We have very limited use of CompactPCI. We
resist as the life time of this bus seems limited. We are using VME for out
timing modules. Bumpless reboot would be nice, but not necessary. I do not see
us rebooting timing IOCs during operation. Still, it is a nice feature.
When we want a small IOC, the Moxa unit has
an arm processor and is configured to integrate serial and Ethernet
instruments.
=======================================
Further comment:
cPCI seems a computer bus. As such, I think that
VME will be around long after cPCI is no longer available - like. uTCA was
considered as a platform. The fabric seems mostly geared toward replacing
routers. You receive lots of fast serial interfaces. Perhaps this will grow
into the next instrumentation bus. However, the uTCA seems best suited to send
data from 1 place to many places for processing - or mostly - redistribution.
Since our situation is getting fast data from one place to someplace else for
processing, we are developing FPGA based solutions to replace the analog front
ends which convert to digital and then communicate to the IOC. This way, we
program in the capabilities that we want for manipulating high speed data. We
are standardizing on the Digital front end to provide 8 high speed serial
lines. It is used as the fast orbit feedback processor and data
collector, the power supply controller, and the bpm. For each of these
configurations, 1 GBit port is used to communicate to EPICS. This is a
departure from a create based system. Instead of replacing a VME card - we
replace a pizza box that uses the same interfaces. The boxes come in at 2K
each. Also of interest, we can decode the event stream on each pizza box.
Bob
-----Original Message-----
From: Hoff, Lawrence
Sent: Mon 1/25/2010 6:58 PM
To: 'Jeff Hill'; Dalesio, Leo
Subject: RE: compact PCI versus VME
I
have absolutely NO experience with CompactPCI.
RHIC has such a large inventory of VME modules
(both
custom and COTS), that we could only consider
non-VME
options for specific cases. I.e. a wholesale
switch is
out of the question.
Having
said that, we do recognize that VME has
been around a looong time, and we want to know
what
our non-VME options are. Therefore, far from
providing
any negative opinions, I want to be "all
ears". I.e.
I am interested to hear that "many modules
and crates
cost about half as much". That could spur
us to
invest in some CompactPCI - at least to
"kick around
the lab". Can you elaborate on what those
modules are?
I
am also interested in switched-serial-only
standards such as uTCA. We occasionally bump up
against
the VMEbus backplane throughput limits. While
PCI
is faster than VMEbus, both pale in comparison
to
switched serial backplanes. Any experience
there?
-- Larry
> -----Original Message-----
> From: Jeff Hill [mailto:[email protected]]
> Sent: Monday, January 25, 2010 6:46 PM
> To: [email protected]; [email protected]
> Subject: compact PCI versus VME
>
> Hey Bob, Larry,
>
> We are looking at tradeoffs of Compact PCI
versus VME. It seems that
> everything is going to switched serial
fabrics, and that both the new
> compact PCI express and the VME/VXS have
that. However, I notice that
> many
> modules and crates cost about half as much
in compact PCI, and that the
> switched serial stuff is more prevalent in
compact PCI at this time.
>
> I think I understand that BNL is still using
VME for new crates? Did
> you
> guys look at compact PCI at all, have any
negative opinions to offer on
> compact PCI?
>
> Jeff
>
______________________________________________________
> Jeffrey O.
Hill
Email [email protected]
> LANL MS
H820
Voice 505 665 1831
> Los Alamos NM 87545 USA
FAX 505 665 5107
>
> Message content: TSPA
>
>