Sony in talks on commercial use for PS3

[Lazy8s]

Sony wouldn't sell PS3s for as low as retail price for commercial, distributed computing use since those buyers wouldn't be needing games for those systems and therefore wouldn't be supporting the loss-leading console business model. PS3s would have to be sold for a profit to them.

Would that allow the PS3 to be a good value compared to solutions for commercial and scientific computing, like Clearspeed, requiring double precision FLOPS?

 

[patsu]

It's a difficult proposition but it's also fairly "new". I don't think we should write the "Utility Computing" concept off yet.

Both IBM's and Sun's utility/grid computing efforts years ago failed. There were a few problems (e.g., software, trust). These people may also pull out because the market (then) was not big enough. Do you know what exactly happened ?

A worthy utility computing effort would be Amazon's EC2 and S3. I think both are doing quite well. P2P platforms like Skype might be interesting too.

Besides running small SPE programs in the background...
I hope because of this, they make Linux image/apps runnable within Game OS.

 

[one]

I don't know if the PS3 can do well for a commercial scheme if the entire platform didn't already have that as a primary goal.

The vision that Cell processors scattered in the world form a single computer has been repeatedly mentioned by Kutaragi since 2000 but in this aforementioned interview of Kutaragi conducted 2 years ago is more related to utility computing.
http://www.eetimes.com/showArticle.jhtml?articleID=163701734

Kutaragi: We can pursue a lot of dreams based on the Cell, but it will take time. My mission [at Sony] is to concentrate on implementing the Cell in practical applications and propagating it.

I am sure that a technology revolution is about to occur, not only within Sony but throughout the digital consumer electronics industry.

An enormous demand for computing will be born in homes. A part of the Cell's computation power could be sold to, for example, information service providers and could be purchased when large processing power was necessary. This is already realized as a part of grid computing, but the Cell could realize it in much faster and more effective way. This would encourage the emergence of a new business.

 

[aldo]

Perhaps a bit naive, but I’d be interested in if or how this impacts Sony’s offering of a free online service.

Will corporate usage be used to offset Sony’s offering of its free online service?

Does this obligate Sony to maintain an ongoing free online service, e.g, how would corporations react to Sony announcing X-live like rates which would most likely cause a drop in accessible units?

-aldo

 

[patsu]

I see them as 2 different creatures though.

Participating units will still be able to "fold" commercially regardless of whether PSN is free or chargeable.

 

[3dilettante]

The vision that Cell processors scattered in the world form a single computer has been repeatedly mentioned by Kutaragi since 2000 but in this aforementioned interview of Kutaragi conducted 2 years ago is more related to utility computing.
http://www.eetimes.com/showArticle.jhtml?articleID=163701734

It's not a question so much fo Cell's applicability but the platform and infrastructure it's on.

PS3s are personal appliances users may or may not leave on, hanging off who knows how many crappy home networks, on thousands of ISPs, all without guarantees of uptime, network performance, or authentication.

In the wide range of applications that would wind up being bound by network latency (or the limited storage and variable availability of a PS3), Cell's other advantages are mitigated.

As a charitable donation of cycles, Folding at Home is a lot less demanding.

Are there other projects that can use the computation? Sure.
Are any of them meant to make money?

 

[patsu]

It depends on how the deals are structured.

What is the expected service level for specialized grid computing applications ? e.g., How long before they expect the result to be sent back, how many Gb are the dataset ? Are there applications that will take time to raise funds/schedule to run on a supercomputer anyway without a cheap enough alternative ?

We need more details about the needs.

For serious jobs, the end product may not be something we run in the background while hanging out in Playstation Home. It may mean running an image full-time (e.g., when the user sleeps), or even handling data on Blu-ray discs and external drives (I hope not)

 

[3dilettante]

It depends on how the deals are structured.

What is the expected service level for specialized grid computing applications ? e.g., How long before they expect the result to be sent back, how many Gb are the dataset ? Are there applications that will take time to raise funds/schedule to run on a supercomputer anyway without a cheap enough alternative ?

The client will still need to go through the expense of data gathering and packaging of work units.
Its apps would have to be customized, and the workflow must now include means of verification and error checking.

Is the savings in front-end costs for building and maintaining an on-site system worth the consistently higher overhead of distribution to 3rd parties through middlemen?

Maybe, if the client just has this one big non-proprietary job and never intends to use the system again.

Otherwise, the only way to save money is to not pay somebody somewhere full cost (the user). One gets the reliability one can expect from such a source.

Any proprietary data would have to be sent over an open network to thousands of insecure locations.
Aside from network interception, any such data would have to stay out of the PS3's hard disk and reside entirely in its tiny RAM.

This would put a limit on the size of the work unit, and place a premium on network bandwidth and latency.

For serious jobs, the end product may not be something we run in the background while hanging out in Playstation Home. It may mean running an image full-time (e.g., when the user sleeps), or even handling data on Blu-ray discs and external drives (I hope not)

I hope not as well. No way my financial data is going on a Blu-ray disk mailed to someone's house.

I'm not getting into legal and political issues of data crossing jurisdictions.

 

[patsu]

The client will still need to go through the expense of data gathering and packaging of work units.
Its apps would have to be customized, and the workflow must now include means of verification and error checking.

Is the savings in front-end costs for building and maintaining an on-site system worth the consistently higher overhead of distribution to 3rd parties through middlemen?

Maybe, if the client just has this one big non-proprietary job and never intends to use the system again.

I think it depends on the problem areas. In general packing/unpacking data and distributing it securely can be done via a middleware independent of the actual solving.

Now that Sony has Linux running on PS3, customers may be able to run these custom apps on a stripped down Linux image compiled for Cell. Then they will have to find some way to run that image within Game OS (or in the worst case, dual boot). If so, this is similar to Amazon's EC2 service: General and cheap.

Otherwise, the only way to save money is to not pay somebody somewhere full cost (the user). One gets the reliability one can expect from such a source.

Any proprietary data would have to be sent over an open network to thousands of insecure locations.
Aside from network interception, any such data would have to stay out of the PS3's hard disk and reside entirely in its tiny RAM.

This would put a limit on the size of the work unit, and place a premium on network bandwidth and latency.



I hope not as well. No way my financial data is going on a Blu-ray disk mailed to someone's house.

I'm not getting into legal and political issues of data crossing jurisdictions.

I don't think they want to handle financial data. The OP only mentioned pharmaceutical applications for a start. Not sure if it's useful for a malicious third party to obtain a small part of the data (when calculating all the data is a big problem to begin with). But agree that in general, the infrastructure has to support good enough security.

 

[one]

It's not a question so much fo Cell's applicability but the platform and infrastructure it's on.

PS3s are personal appliances users may or may not leave on, hanging off who knows how many crappy home networks, on thousands of ISPs, all without guarantees of uptime, network performance, or authentication.

Uptime is not important, F@H is mostly offline.
As for security architecture, Cell supports it in hardware as I mentioned in my first post in this thread.
http://www-128.ibm.com/developerworks/power/library/pa-cellsecurity/

IBM's security SDK for Cell (which is not directly related to PS3) is currently not available in public though.
http://www.alphaworks.ibm.com/tech/cellsw

The Cell Broadband Engine Security Architecture can now be explored using the newly-released Cell Security SDK. The Cell Security SDK is released under a special proprietary license and is not downloadable here at alphaWorks. For further information, please e-mail alphaWorks licensing support.

 

[3dilettante]

As for security architecture, Cell supports it in hardware as I mentioned in my first post in this thread.

It's more secure, not invincible.

One problem:

The infrastructure outlined by Kutaragi is not between the client and the PS3. The server belongs to the ISP. What part of the SDK secures the server?

What part of Cell's design keeps a server from being spoofed into thinking it's transmitting to a PS3?

Like I said, it's not Cell that's at issue.

 

[one]

It's more secure, not invincible.

One problem:

The infrastructure outlined by Kutaragi is not between the client and the PS3. The server belongs to the ISP. What part of the SDK secures the server?

What part of Cell's design keeps a server from being spoofed into thinking it's transmitting to a PS3?

Like I said, it's not Cell that's at issue.

I think what you are trying to describe is man-in-the-middle attack and it's prevented by server authentication by a server certificate.

EDIT: I might took it wrong, but you meant wiretapping by ISP? It's prevented by PKI.

 

[3dilettante]

Not in the last post. The two scenarios are a compromised ISP server and a non-authorized system masquerading as a PS3.

 

[Scott_Arm]

I think this is a cool idea as long as people are able to decide which projects they are lending cycles to. What I wouldn't want is a system where Sony has deals with X companies to lend them computing cycles of people playing games in a Sony Arcade or something like that. I wouldn't want my background processing going towards any projects that I didn't agree with. I highly doubt a situation like that would happen, but it was something that crossed my mind.

 

[one]

Not in the last post. The two scenarios are a compromised ISP server and a non-authorized system masquerading as a PS3.

If Cell has a hardware secret key embedded in it and SCE knows it, a compromised ISP server or a PS3-impersonating system can't decrypt a message from PS3 or SCE.

 

[3dilettante]

The server in question is the ISP if the ISPs are the ones selling the service, not the company that is buying processing power.

Routing everything through SCE is not what I'd call a desireable solution.

 

[one]

The server in question is the ISP if the ISPs are the ones selling the service, not the company that is buying processing power.

Routing everything through SCE is not what I'd call a desireable solution.

Hmm. Maybe SCE can lease a security hardware to a customer? If it's stolen, it's revoked.

 

[3dilettante]

The question is how the security hardware can work without first getting in touch with a PS3. Since this is more like grid computing, the individual PS3 isn't known beforehand.

I'm not even sure if it's in keeping with the supposedly abstract nature of a computing grid for the client to ever know anything about the processing machine on the other end.

There needs to be a way of authenticating the ISP's grid-task assignment server.
The info I've seen doesn't seem to show that this part of the infrastructure is present yet.

 

[one]

The question is how the security hardware can work without first getting in touch with a PS3. Since this is more like grid computing, the individual PS3 isn't known beforehand.

Participants are PS3s registered via Playstation Network accounts. If extra security is required, credit card information can be also requested to join in this initiative.

Then, via network, a security hardware leased from SCE can access the user database that contains PSN user IDs and associated hardware IDs of Cell processors.

 

[one]

I think this is a cool idea as long as people are able to decide which projects they are lending cycles to. What I wouldn't want is a system where Sony has deals with X companies to lend them computing cycles of people playing games in a Sony Arcade or something like that. I wouldn't want my background processing going towards any projects that I didn't agree with. I highly doubt a situation like that would happen, but it was something that crossed my mind.

Unfortunately I think that's very probable and a secret project for atomic bomb simulation is processed alongside a life science project! But hey, it's like watching a TV program sponsored by a company you don't like.