February 24th, 2012, 10:37 am
Am wrong, or is this completely wrongheaded?QuoteIntilop debuts TCP Offload Engine offering low latency using Altera Stratix IV FPGASANTA CLARA, Calif., 20 Feb. 2012. Intilop Inc., maker of complex ultra-low-latency networking Mega IP building blocks and systems, is introducing its 4th Gen. SX-Series 10G ultra-low-latency TCP Offload Engine (TOE) + EMAC + Altera PHY Mega IP cores running on a Stratix IV field-programmable gate array (FPGA). The TOE?s architecture is scalable, customizable, and adaptable. Intilop's product-line solutions are available in flexible FPGA/ASIC/SoC technologies, which can accommodate a diverse set of technical design specifications.Doesn't TCP just do a CRC check or something, and discard the entire packet and do a resend if it doesn't match up? You could have one bad byte and throw the whole thing away. Or worse, you could be using some crap hardware signal processing to resolve the bytes, when a more ambitious software algorithm could extract them accurately from the line signal. I know software is not as fast as hardware. But anything is faster than a resend. At the very least it could pass on the raw signal to the software, in hopes it could flip the ambiguous byte and get a correct CRC matchup in less time than it takes to resend.Again, I don't really know what I am talking about, so I post it to see if anyone else does.