An FPGA-based Prototyping Platform for Research in High-Speed Interprocessor Communication V. Papaefstathiou, G.Kalokairinos, A.Ioannou, M.Papamichael, G.Mihelogiannakis, S.Kavadias, E.Vlahos, D.Pnevmatikatos and M.Katevenis
Inst. of Computer Sci. (ICS) – FORTH – Crete, Greece
1
Key Features (First System) • • • • • • •
Remote DMA (RDMA) based operation Notifications: departure and/or arrival, interrupt or enqueue Remote Enqueue for short messages, multiple senders Credit-based flow control: lossless communication Per-destination Virtual Output Queues (VOQ’s): flow isolation Extensive event logging, debugging & performance counters Switch: – 8x8 implemention – 32-bit datapath @78.125 MHz – achieved up to 16x16 – 16-bit dpth @156.25MHz to fit in FPGA
• Linux already adapted for this platform (kernel-mode comm.) • MPI port for this platform under way 2
Photograph of First System (8 nodes)
3
Hardware Development Cost (1st System, 2 versions)
RocketIO Links (10pm)
First Version ( plain PCI ) (20pm)
2004
2005
Multipath Routing (10pm)
Second Version PCI-X (12pm) Multiple VOQs (6pm)
Demo Evaluation
Buf. Crossbar Switch (6pm)
Future System
2006
pm = person-months 4
NI Photo, with 4 RocketIO links
5
NI Architecture
PCI-X
100MHz
PCI-X PHY
64-bit PCI-X @100MHz to Host Per-dest. DMA request Q’s Per-destination VOQ’s DMA segementation into packets Link bundling: 4×2.5 = 10 Gb/s DMA Request Queues – inverse multiplexing multipath routing • Out-of-order packet arrivals: – DMA body immediately written into memory – headers wait in resequ. Q’s – nxt sys: just count # bytes – notify completion after resequ. / count complete – resequ. tolerates 1-pck loss
Target PCI-X I/F
Perf. & Debug Counters
Initiator PCI-X I/F
DMA Engine
VOQs
body
• • • • •
Completion Notification headers Reseque.
MultiPath
Link Interface Link PHY
RocketIO 6
Next Generation (2007) Node: Block Diagram
On-chip, @ 266MHz
PowerPC
NI must be simple and small compared to CPU and its local memory
An FPGA-based Prototyping Platform for Research in ...
written into memory. â headers wait in resequ. Q's ... (fast, on-chip, up to. 306 KB). Next Generation (2007) Node: Block Diagram. 10 Gb/s. 10 Gb/s. NI must be ...
cial âStart Flagâ bit is set in the last one of the clus- tered requests; at ..... specific FPGA environment and the specific RocketIO ..... [20] Broadcom Corporation.
vide an aggregate bandwidth of 4Ã2.5 = 10 Gb/s. We ..... Table 1 shows the person-months spent in the .... packet from the NIC, namely 40 bytes, we spend 16.
The real-time nature of the in-memory data platform supports Intel IT's goal of a ..... large volumes of data quickly because of the in-memory computing. Data that ...
considerations attached with the CV of the contractor outlining all the experience obtained and relevant for the project to the email [email protected] by ...
fields, from telecommunications to transportation, to medical applications. Such systems ... Essential measures of reliability for digital communication systems are the bit error rate (BER), block error rate (BLER) [1] ... of a flexible clock domain
Aug 6, 2013 - Release 1.03. Zihan Chen. 1. , Anton Deguet. 1. , Russell Taylor. 1. , Simon DiMaio .... the high-speed serial network (IEEE-1394a) and the I/O hardware. In this design .... of services: isochronous and asynchronous transfers.
Fax: (716) 645-6963. E-mail: ... PAGE EST: 7.5. Published on Web 00/00/0000. 1. 2. 3. 4 ... solution of kanamycin A free base (2.57 g, 5.7 mmol) was dissolved in ...
Mar 17, 2017 - In view of the guidelines mentioned in the above circular the Exchange has reviewed the lot size of the security Jet Knitwears Limited on SME ...
library of C++ code for real-time vision processing, into FVision (pronounced \ ssion"), a fully-featured domain-speci c language embedded in Haskell. The re-.
Systems." yFunded by the ESPRIT BRA ProCoS project: no. 3104, \Provably Correct Systems." zOn leave from the .... represents the data space available to the ML0 program. ..... can be found using the psi clause (see later for more details).
The development of computer-based systems can benefit from a formal approach at all levels of ... This phase is popularly called the optimization phase.
issues. The IPC platform also has a large number of. PCI and ISA expansion slots available, 16 in total, and provides the potential for greater expansion than a.
Evolutionary algorithms are abstract formalisations of natural processes. In a sense they have been removed from their natural context and transplanted into the ...
A Method for the Rapid Prototyping of Custom Contoured Cushions ... Such a system would be capable of producing a suitable custom-contoured cushion in.
will explore a data-drive, evidence-based approach to modeling and simulating ... will gain experience with data collection, data visualization, and dynamic.
computer, a programmer has to do low-level device ... Even when this is done, the programmer ... SDGToolkit appears to the SDG application developer, and.
Apr 20, 2017 - Review of experience with scientific approaches to co-development ... Implementation of the 2016 Notice on the application of the Orphan.
May 19, 2017 - Advances in the co-development process for personalised medicines ... of engagement for marketing authorisation applications involving ...
Oct 30, 2009 - measure the endpoint position and provide force feedback in three .... was packaged into major CAD systems and it would speed up the ...