Philipp Gontermann wrote: > Jan Kiszka schrieb an Philipp Gontermann <date=Mon, Apr 03, 2006 at > 02:54:36PM +0200> >> Philipp Gontermann wrote: >>> Hi, >>> >>> we wanted to use rtnet for some hard real-time application, >>> but we are not shure if it will meet our requirements. >>> So here is the scenario: >>> - 2 Computer, directly connected via 100Mbit >>> - first one sends one package of at most 100Bytes >>> - second computer takes this data and does some computation >>> (approx. 300us) and sends back 1000Bytes >>> - then first one again sends data .... >>> >>> We need the responce data is available at the first >>> computer after at most 800us. It is necessary that NO cycle >>> should take make than this time. >>> Is it possible to meet these requirements with rtnet? >>> What ist the minimal roundtrip time which can be reached? >> That's hardware-dependent. With typical mid-range desktops or high-end >> embedded industry PCs (I assume you use x86?) you can go below 1 ms >> cycles, and you will still have time for "real" work. > One side is x86 (1.73Ghz, the other one is a embedded system with > PowerPC 405 (250Mhz) >> Moreover, if you >> have a fixed traffic schedule like this one and you can waive non-rt >> packet tunnelling, you can skip RTmac/TDMA and gain further resources. > Yes, we have fixed schedule. How much time can we save > (approx.) by skipping RTmac/TDMA.
You save the periodic issue or reception of the TDMA sync frame and the overhead for queuing packets and later dequeuing them for timed transmission. I have no numbers at hand (it depends on your targets cycle time anyway), and specifically I have no clue about your PPC board's performance (maybe Wolfgang can comment on this when time permits). > Can you tell anything about the minimal time between receiving data and > sending some other. > Can be as low as a few 10 us on high speed boxes, but can also become several 100 us when transmitting larger packets and when the receiver is low-end. It's best when you test it on your own: set up a "bare" RTnet between your two hosts (no RTmac) and run rtping while putting heavy non-rt load on both boxes. This will give you an idea of the worst-case overhead RTnet, the RTOS, and the hardware introduces. For you final application, you will additionally take another context switch and the application task's work into account. Jan
signature.asc
Description: OpenPGP digital signature
