Optical components

Google Loosens Up on Data Center Optics

SANTA CLARA, Calif. -- Google (Nasdaq: GOOG), realizing it isn't interested in the stringent requirements of the telecom network, is trying to craft a new standard for data-center optics.

The idea -- and that's all it is so far, with no immediate plans to submit anything to a standards body -- got its first public airing at the The Ethernet Alliance 's Technology Exploration Forum this morning, in a discussion that was supposed to be low on controversy.

Vijay Vusirikala, an optical architect with Google, explained that Google thinks it can save money and speed optical module development times by eliminating some of the corner cases that telecom gear is required to meet -- loosening the requirements, in other words. A Google data center never gets to 0 degrees Celsius, for instance -- so why does it need components built to withstand that temperature?

The concept is in line with Google's overall data center philosophy, where zillions of identical, cheap servers get thrown at every computing problem -- and get quickly thrown out if they fail.

Google's immediate objective is to see if other Internet data center owners -- Amazon.com Inc. (Nasdaq: AMZN) and Facebook , specifically -- even like the idea, said Chris Cole, Finisar Corp. (Nasdaq: FNSR) director of engineering, who headed the discussion.

Cole, Vusirikala, and Google honestly don't know the answer. As Cole put it, Amazon and Facebook are as quiet about their networks as Google was two years ago.

Cole and Google tried reaching out, with emails to eight owners of massive data centers, but they got no meaningful response. (Donn Lee, who's hit the conference circuit talking about Facebook's 100Gbit/s requirements, sent "something generic," Cole said. Amazon apparently sent nothing.)

So, Google is trying a new tactic: Writing a first-draft specification to see if it's something Amazon and Facebook care about.

The "standard" mainly involves removing corner-cases from the testing and design cycle. "What we are focused on relaxing are the things that take a lot of time up front," Vusirikala said.

Some of the suggestions include:
  • Looser temperature requirements, such as designing for a maximum temperature of 60 degrees Celsius rather than 70 or more;
  • A less stringent bit error rate than telecom is used to -- maybe 10-12 versus 10-15 for DWDM equipment;
  • About 1,000 hours of accelerated aging tests, which is half the norm.

Google has no quarrel with most other Telcordia Technologies Inc. specifications, such as temperature-cycling tests or requirements for withstanding vibration.

Vusirikala told Light Reading that his talk wasn't meant to be as controversial as, say, Google's criticism of standard 100Gbit/s optics, which came out at the recent European Conference and Exhibition on Optical Communication (ECOC). (See 100G Watch: Google Complains Again.)

But some, including the Brocade Communications Systems Inc. (Nasdaq: BRCD) and Cisco Systems Inc. (Nasdaq: CSCO) representatives who'd been invited to present talks as part of Cole's session, misinterpreted Google's proto-standard, thinking it would require entire systems to be redesigned.

Easy there, Vusirikala said. All Google is talking about is a cheaper type of pluggable optical module. Google buys its optics directly from the module vendors, so this isn't something that systems vendors need to think about.

Optical components and module vendors would stand to gain the most from Google's idea, since the extreme corners of the telecom specifications help drive costs upward. As Vusirikala put it, the extra engineering margin means less profit margin for the module vendor.

The question, as Cole told Light Reading, is whether Google and its ilk represent a big enough market to make the idea work.

The best-case scenario would be that Google is a leading indicator. "If it is a successful market, and people realize there are lots of benefits to this, people could try to design their own data centers like this," he said.

In fact, the Cisco and Brocade talks showed some potential benefits of Google's approach. For example, if the maximum temperature of an optical transceiver is assumed to be lower -- meaning you're no longer building to worst-case scenarios -- that creates the leeway to put more power into certain components, possibly increasing throughput.

These are all maybes, and Google knows it. Even if other companies like the idea, it's still yet to be seen if the expected benefits come to life. "If there's no gain for us, we don't seen any need for more thrust behind this proposal," Vusirikala said.

— Craig Matsumoto, West Coast Editor, Light Reading

Page 1 / 3   >   >>
yike 12/5/2012 | 4:21:52 PM
re: Google Loosens Up on Data Center Optics

Why the telecom optical standards can not meet google or data center requirements ? too expensive of OpEx or CapEx? We don't expect too many standards... as TDM and Packet network are merging, we don't expect data center network go a different direction.

paolo.franzoi 12/5/2012 | 4:21:51 PM
re: Google Loosens Up on Data Center Optics



Google has already stated:

Capex - Google feels that the data center folks are paying for features required for telecom networks that are not required in the data center.

Opex - Google feels that by changing some requirements that lower power (and thus lower Opex) might be achieved.

You can argue with them on this issue or not, what they are saying is that the currently planned versions of the standard for 100G fiber interfaces do not meet their needs.



Duh! 12/5/2012 | 4:21:50 PM
re: Google Loosens Up on Data Center Optics

I understand the goal, but am trying to understand the strategy.

Do they think that relaxing specs will increase yield?  Use die that have been rejected for industrial/telcom spec parts?  Allow the industry to design cost-reduced packaging?  Shorten testing or reduce testing sample sizes? 

What is the business case for this from the vendors' perspective?


Pete Baldwin 12/5/2012 | 4:21:49 PM
re: Google Loosens Up on Data Center Optics

Of the things you listed, it's the next-to-last one: shorten testing time.  Google is also hoping that a looser spec will lead to faster product development.  It sounds like it's less about yields (although that might be a beneficial side effect) and more about making the design easier in general.

For the optical components guys, this could be nice. They'd have a market for less stringently engineered/tested parts, meaning they could start getting returns on some designs more quickly.

I don't know this for sure, and Vijay didn't say it to me, but: My guess is that Google, which buys its optical modules directly from the module vendors, kept getting told that this-and-such a product wouldn't be available yet because it needed to complete its testing, or that it needed tweaks in order to survive at 0 degrees. Google seems to want to get its hands on these things faster -- and, of course, wants to pay less for them.

Whether these benefits would surface is anybody's guess -- that's why Google wants input from people like Facebook, to see if the idea's got enough merit to build on.

jr_hoge 12/5/2012 | 4:21:45 PM
re: Google Loosens Up on Data Center Optics


If your understanding of Google's intent is correct, then it seems to me that their efforts to further proliferate standards is totally counter to the ends they want.

Multiple, overlapping standards mean that engineering resources get stretched and  economies of scale distroyed. One would think that their nends would better be served by minimizing the numbers of Phy for any given datarate.

Pete Baldwin 12/5/2012 | 4:21:42 PM
re: Google Loosens Up on Data Center Optics

Could be.  Google is admitting they don't know the full effect of this idea.  But Finisar sees merit in this (judging from Cole's enthusiasm), so there's a chance it could actually help both sides.

Duh! 12/5/2012 | 4:21:40 PM
re: Google Loosens Up on Data Center Optics


It almost sounds as if Google is trying to create a "relaxed spec" option in the standards, rather than creating a new standard.  It also sounds as if the "relaxed spec" option will create a branch off the NPI process with an early path to shipping product, but will merge back into the mainstream product after it gets qualified. 

Does that really require a new option in the standards?  Couldn't vendors just characterize sample parts, and if they're within acceptable parameters to a big customer like Google, do a pre-production run in parallel with the mainstream NPI process?

Also, wouldn't the window of opportunity for a relaxed spec long since have closed before an addendum to the 100G standard got to sponsor ballot in IEEE 802?

Pete Baldwin 12/5/2012 | 4:21:36 PM
re: Google Loosens Up on Data Center Optics

An offshoot spec is one possibility.  Keep in mind, Google hasn't technically done anything yet.  They're describing the way they'd like to see processors built and sold.  Whether that means a full-blown standard, an offshoot standard, is yet to be seen.  If the idea doesn't actually save time or money, then nothing happens and we go back to normal.

Someone else commented -- was it on here or in person? too much going on this week -- about the 802.3ba standards ship having sailed.  Two things about that -- first, I'd presume Google started thinking about this problem relatively recently, maybe in the past year, long after the standards work was underway. Second, even Google doesn't always get its way in standards.

The proposal here is very much in the Google/Web2.0 way of doing things. Got an idea? Just go do it and see if it flies. Google is likely to meet some resistance, and it's possible the whole idea won't work; maybe there's good reason for the stringency of telecom standards and IEEE rules.

jepovic 12/5/2012 | 4:21:34 PM
re: Google Loosens Up on Data Center Optics

Multiple standards make sense if there are truly diverse requirements. It doesn't make much sense to have products that fulfil 98% of the specs.  For example, Google will surely require 10 degrees in case the AC goes bananas, and is it really that challenging to handle operation at zero degrees? So the products must still be tested. Plus, the contracts must be rewritten etc (imagine how much work there will be for the lawyers if half of these Google-spec interfaces break down).

On the other end of the spectrum, isn't military grade specs becoming less common?

jepovic 12/5/2012 | 4:21:33 PM
re: Google Loosens Up on Data Center Optics

Google's argument that it makes sense for vendors is kind of hilarious. The key point for them is to cut capex. So they will go to vendors with a spec which has been reduced by 2%, and ask for a 25% price reduction. They're big, but I dont think they're big enough to get away with that.

Page 1 / 3   >   >>
Sign In