[Sis-csi] VxWorks vs. Linux

[Jim Benson]

For whatever it is worth, and as the pioneer of the MPC750 and VxWorks in space, we have abandoned both and are now going with much more advanced commercial off-the-shelf processors and are running Linux (with Ethernet and USB connections for all plug and play subsystems) on our three formation-flying, networked  microsats for the Missile Defense Agency.

Onward and upward,

Jim Benson
 

-----Original Message-----
From: sis-csi-bounces at mailman.ccsds.org [mailto:sis-csi-bounces at mailman.ccsds.org] On Behalf Of Leigh Torgerson
Sent: Thursday, January 25, 2007 3:49 PM
To: Scott, Keith L.; Marc Blanchet; Lloyd Wood
Cc: sis-csi at mailman.ccsds.org
Subject: RE: [Sis-csi] 2:30 Eastern it is

Just as an aside, this discussion thread is fundamentally why we ported CFDP/TM/TC into a VirtexII Pro FPGA.

With the Deep Impact telemetry stack implemented in FSW (VxWorks) on a PPC750, lab tests showed a couple of megabits/sec max throughput with a full software implementation of the stack (with something like 67% CPU utilization if I remember
correctly..)

With our TRIGA FPGA implementation, we've run CFDP/TM/TC at 600 Mbps in reliable mode. This implementation is being looked at for porting to an Electra slice, using the rad hard version of the FPGA and a SPACR co-processor (used to manage CFDP state).

The flight software does a couple of register writes to the FPGA to set up the metadata and start a transaction, the FPGA interfaces with mass storage over an LVDS interface and conducts the transaction, including any retransmissions, and many of these flight software optimization / OS discussions become moot.

We've got AOS (partial implementation to support
DTN-BP) in the FPGA stack now, and the port of DTN-BP has commenced. Also being worked is the FPGA interface for AMS, and we've started the LTP FPGA design.

regards,
Leigh

At 5:00 PM -0500 1/25/07, Scott, Keith L. wrote:
>I think this is correct.  One step is to ask the question: how fast do 
>we think we could forward packets in space?  The answer depends on the 
>hardware/software available, which will probably be (for the near
>future) VxWorks or some variant.  I suspect that the whole matter is 
>moot if we assume that the packet forwarding isn't sharing resources 
>with the C&DH, as even relatively weak machines can probably outstrip 
>current uplink rates (of TDRSS' 25Mbps, e.g.) by a factor of 2; 
>moderate or powerful machines (in a flight-qualifiable context) by 
>factors of 2-4.
>
>		--keith
>
>
>-----Original Message-----
>From: sis-csi-bounces at mailman.ccsds.org 
>[mailto:sis-csi-bounces at mailman.ccsds.org] On Behalf Of Marc Blanchet
>Sent: Thursday, January 25, 2007 4:56 PM
>To: Lloyd Wood
>Cc: sis-csi at mailman.ccsds.org
>Subject: Re: [Sis-csi] 2:30 Eastern it is
>
>Le 07-01-25 à 16:44, Lloyd Wood a écrit :
>
>>  At Thursday 25/01/2007 15:45 -0500, Marc Blanchet wrote:
>>>>  What rates can VxWorks forward at over the past N years?
>>>>  VxWorks -- how fast can the network stack forward.
>>>
>>>  why bind standards and architecture with specific stack/OS  
>>> implementation?
>>
>>  because you want implementations, and what good is a standard  
>> without implementations?
>
>of course.
>
>>
>>  The 'real world' choices for operating systems in space hardware  
>> tend to be VxWorks or RTEMS. What these two OS architectures have  
>> matters for implementation - e.g. RTEMS lacks processes/memory  
>> management.
>
>ok. but I can run a vxworks stack on a 3Ghz intel quad cpu or on a 
>100Mhz ARM.
>
>>
>>  I think an appreciation of how much performance you can get out of  
>> real-world space hardware is important and has ramifications for  
>> architecture/design decisions.
>
>then, the question should be two-fold:
>a) hardware current and planned. max throughput
>b) penality of stack/OS (preferably on that hardware).
>
>taking b) only to me does not help.
>
>Marc.
>
>>  The effects of Moore's law are brutal in space (exponential  
>> increase seems slower than on Earth, so there's a gap) and thanks  to 
>> power/radhard requirements the processors used are slow. You'll  be 
>> imo lucky to achieve 10Mbps line rate from a stack running on a  
>> power-conserving general-purpose processor without hardware assist  
>> or brutal performance tweaking.
>>
>>
>>  L.
>
>-----
>IPv6 book: Migrating to IPv6, Wiley, 2006, http://www.ipv6book.ca
>
>
>
>_______________________________________________
>Sis-CSI mailing list
>Sis-CSI at mailman.ccsds.org
>http://mailman.ccsds.org/cgi-bin/mailman/listinfo/sis-csi
>
>_______________________________________________
>Sis-CSI mailing list
>Sis-CSI at mailman.ccsds.org
>http://mailman.ccsds.org/cgi-bin/mailman/listinfo/sis-csi


--
-------------------------------------------------------------------
J. Leigh Torgerson                          ltorgerson at jpl.nasa.gov
Jet Propulsion Laboratory - Communications Systems Research Sec.332 DARPA DTN / DSMS Advanced Protocol Implementation Work Area Manager
(818) 393-0695                                   (818) 393-4643 fax
-------------------------------------------------------------------

_______________________________________________
Sis-CSI mailing list
Sis-CSI at mailman.ccsds.org
http://mailman.ccsds.org/cgi-bin/mailman/listinfo/sis-csi