Unleash Networks Blog

Just when I thought the Cell Processor was the future platform for high performance + low cost CPU intensive applications – there is more hot action in the multicore processor space.

First a note about Sony’s much anticipated cell processor board 

I am disappointed by the low-key launch of the Sony Cell Processor Board at SIGGRAPH 07. More negative news hit us in the form of performance problems with Madden 2008 on the PS3 vs XBox360. Could this be due to the programming complexity ? To top it, Sony/Toshiba/IBM do not seem to promote the cell processor for networking applications.

Tilera founded by MIT Professor-turned-entrepreneur Dr Anant Agarwal announced a 64-core chip called Tile64. 

The Register UK Reports : The Silicon Valley-based start-up’s first product links together 64 RISC-like cores running at up to 1.0GHz. The real magic, however, stems from the five-lane switches used to link each core in an 8X8 grid that provides up to 32 terabits per second of data bandwidth across the whole chip. You end up with a product – Tile64 – that can tear through software threads.

Chip architecture :

A quick look at the chip architecture.

 

1. Tiles are arranged in a 8×8 grid (contrast with the architecture of the Cell Processor which featured two counter rotating rings)

2. An absolutely cool feature is the On Chip Four DDR2 controllers and Dual Gig E On Chip.

3. Each “tile” has a L1 and L2 Cache which should make programming for it somewhat easier. The SPE’s in the Cell Processor do not have these caches, which contribute to the programming complexity.

Networking applications

This is where I am most interested in, the Tilera site claims that

• 10 Gbps Snort® processing
• 20+ Gbps iptables (firewall)
• 20+ Gbps nProbe
• 16 X 16 SAD at 540 MBlocks/s
• H.264 HD video encode for two streams of 720p @ 30 Fps

Tilera explicitly says that the Tile64 chip is most suited for networking. I am going to follow this closely. Currently, Unsniff Network Analyzer with all options turned on (TCP session tracking, PDU analysis, User Objects) is running out of CPU juice around 6-10Mbps. 

What I would be really interested in

The first question is : well how easy is it to program.

Tilera also sells a product called the Multicore Development Environment (MDE) which claims to ease software development.

• iLib standards-based parallel programming library for efficient inter-core communication
  • Socket-like communication
  • Message passing
  • Streaming channels
• Graphical multicore application debugging and profiling

 

Interesting times ahead.

[tags] Tilera, Tile64, Cell Processor, Unsniff multicore [/tags]

 

I am just jotting down these notes as I go along. Some points may be inaccurate – it just reflects my vague understanding at the time of writing the post. 

I have not yet been able to install the cell processor simulator from IBM. We dont have a powerful enough box to spare at the moment. These are my initial impressions about the cell processor and its suitability for our use.

1. There is a central processor with two hardware threads called the PPE (Power Processing Element). It is either exactly the same as a PowerPC or a slight variant of it.

2. There are eight other processors connected by an extremely high speed ring. These are called SPE (Synergistic Processing Elements). They are kind of crippled compared to the PPE – they dont have branch prediction, or caches. They only have a “scratchpad memory” of 256K.

3. Porting software that works with the PPE is easy. We could probably get by with nothing more than a recompile for the PowerPC. This could give us a major boost over running on another custom hardware platform – both cost and performance wise.

4. Using the SPE is a whole another ball game altogether. We have to rip open our software design from the ground up. The main focus would be to have the PPEs do nothing but feed the SPE with “work” and keep em busy at all times. If we could pull this off – the gains promise to be fantastic. We could do a ton of things at wirespeed.

5. The PS3 is an impressive platform to stage this work. The toolkit “MultiCore Plus SDK” from Mercury seems to be well worth a purchase. It seems to cost less than $500 (I have to verify it).

6. There is very little news about Sony’s Cell BE based board product launch at SIGGRAPH 07. I did find one mention about “Taking the Play out of Playstation“. The features I am interested in the 1U board are :

• Power consumption (the PS3 sucks about 400W)
• Memory type (RDRAM?) + architecture
• Hard drive storage and option for Fiber Channel
• Networking – the PS3 has only 1 GigE port
• Option to purchase without the graphics hardware (HDMI, etc)
• Cost + whether they will sponsor evaluation systems (high hopes here)

 

I am waiting for the unveiling of the cell processor board by Sony and Nvidea at SIGGRAPH starting tomorrow (Aug 5th to 9th). This promises to be one of the most exciting moments in recent computing history.

According to a short statement by Sony, the so-called “Cell Computing Board” incorporates the high-performance Cell Broadband Engine microprocessor developed jointly by IBM, Sony and Toshiba and RSX graphics processor created by Nvidia and Sony. The combination of the two key-components of the PlayStation 3 “results in high computational performance capable of handling large amounts of data at high speed while also achieving reductions in size and energy consumption”.

I am in the process of staging a Lanner FW1200 1-U rack unit with a new probe software. It seems the cell board from Sony uses 400W and can fit in a 1-U Rack mount as well.

Peak processing power of the Cell Computing Board is 230GFLOPS, according to the statement, while peak energy consumption is 400W, which is 20W higher compared to the PlayStation 3 game console. The current Cell Computing Board can be embedded in a 1U (unit) sized server and mounted on a 19” rack, which points to enterprise usage scenario of the device.

I was aware about the Playstation 3’s use of the cell processor, but an article in Dr Dobbs Journal greatly enhanced my interest in this architecture. See “Programming the Cell Processor“. I believe this is an architecture that can take network / security analysis to a completely new level without involving supercomputer type costs.

I am still learning about the power of this platform,  I will blog about it as and when I make progress learning it. This ought to be fun.

My starting points :

The Cell Processor Home Page at IBM Research

I want to see if it is worthwhile to buy a PS-3 and use it as a learning platform.

[tags] Sony, Cell Processor, Playstation – 3 [/tags]