Telecoms Forensics Equipment Ltd

                         Rid your space of all internet vermin, get a Subpico CAT! 


Faster Results

  • Today the demands on volume processing and their HPC/Server Farm environments are enormous and it is not uncommon for some to have issues. For mission critical processing Cloud Based services may not always be the answer. Some may run their own HPCs while many CSPs acknowledge this and provide dedicated servers for this purpose which is a step in the right direction. However they are still restrained by the current generic transport technologies, top heavy XML/SOAP etc. flows, with bloated OS's.
  • With our technology your parallel applications will execute and communicate faster, on die, in package, on processor, or remote. Our Technology is purpose built from the ground up for MPP we can deliver faster and cost effectively.
  • Our many years experience building mission critical systems for Government and Carrier, underline our ability to deliver an industrial grade results.

Better Performance

  • Today 5-10 microseconds latency is not uncommon at the signalling layer. While 10+ us application end to end (ping-pong) for some is considered pretty good. We ping-pong in ~1+/- nano seconds, and as mentioned above, it makes no difference if the nodes are on different dies, packages, processors or chassis cards data to and from your application is direct.
UNIT[80] xPTS[8], CPU 16/CARD, ACTIVE 10
 0 RF[0].mmem = 0x7f70f00008c0, 13088 B
> 'xlib/6b8b4567_' lib_handle @ 0x7f70e8000910, sz 13088 B
clock ndelta 92 ns (cost)
 CORES = 160 0 :: 1_____done RTT: 621 (ns), 2 (ns/core), intfs 20,intf_loss 100 ns
clock ndelta 112 ns (cost)
 CORES = 160 0 :: 1_____done RTT: 553 (ns), 3 (ns/core), intfs 20,intf_loss 100 ns
clock ndelta 114 ns (cost)
 CORES = 160 0 :: 1_____done RTT: 625 (ns), 2 (ns/core), intfs 20,intf_loss 100 ns
0 - $0 0x7f70f00008c0 sz 13088 B
As an example, above are round-trip-time metrics taken from a head node on our old TEST RIG. The application  'xlib/6b8b4567_' is your standard ring application. It simply parses a shared variable around a ring of 160 process nodes, each node increments the variable and parses it on. The TEST RIG uses DDR2/Southbridge-IO and combines the up/down transport. Our production systems use up-to-date triple/quad channel memory + integrated pcie/io, seperate up/down transports and interface losses ~ 2 ns (measured as time between cntrl-fifoA send and cntrlr-fifoB receive). Providing you blistering picosec's node/node and RTT ~1+/- ns ping-pong performance.

Fine-Grained Execution Control

Having massive processing power at your finger tips is one thing, getting actual compute time is another.  We provide the technology for you to run immediate and/or scheduled job loads. In addition, our unique architecture gives you fine-grained execution control allowing ad hoc runtime design and testing, shortening your project design cycle, costs and time to market. 


NOTE: the clock ndelta is a measure of the timer library call, because we are working at very high speeds even looking at your watch incurs a significant time cost. The intf_loss ~5ns inter-card.

  • Whether your average message size is a few bytes or Gigs (our framing is 8/32K) our system fabric is cache and pipeline packed/aligned for optimum message processing
performance. We operate with zero wait/copy and locking offering your code maximum CPU utilization and efficiency.