Thursday, September 30, 2010

UPDATED: Data Center Stocks Rise 20.3% in the 3rd Quarter

By David Gross

Data center stocks were up 20.3% this quarter, well ahead of the 12.3% gain posted by NASDAQ, and 10.7% gain posted by the S&P 500.

Leading the way was Savvis, with a 43% gain, while 7 of the 13 companies we track in the services sector were up over 20%.   Akamai bounced back strongly after a weak earnings call dropped into the 30s briefly, while the REITs have been held back a little as their dividend yields are under 4%.   While I continue to believe that Wall Street is overestimating the threat to Akamai from smaller competitors and Level 3, Limelight and Internap also posted strong gains during the quarter.

Equinix came back from its 2Q price decline, while reporting 21% year-over-year revenue growth excluding Switch & Data.  Competitor Telx filed an additional S-1 in the quarter, but still no news on when it will actually IPO.

Rackspace, continues to do very well operationally and financially, although it is now trading at over 60x annualized earnings with just 20% top-line growth.   Nonetheless, it continues to outperform its peers with a 1.6 Revenue/PP&E ratio, a benefit of staying completely out of the co-lo business.  Service focus is also helping AboveNet grow, with top line expansion in the mid-teens, and the data center-focused fiber provider is on the verge of surpassing the market cap of over-diversified Level 3, whose revenue has been mostly flat over the last 12 months. Services Index

Company Ticker Mkt Cap Sep 30 Close July 1 Open Gain %
Equinix EQIX $4,666,136,500 102.35 81.22 26.02%
Digital Realty DLR $5,386,410,000 61.70 57.68 6.97%
DuPont Fabros DFT $1,489,634,500 25.15 24.56 2.40%
Rackspace RAX $3,246,720,600 25.98 18.34 41.66%
Savvis SVVS $1,164,459,200 21.08 14.75 42.92%
Level 3 LVLT $1,555,420,000 0.94 1.09 -14.04%
Akamai AKAM $9,113,189,800 50.18 40.57 23.69%
Navisite NAVI $125,851,200 3.34 2.63 27.00%
Terremark TMRK $679,441,400 10.34 7.81 32.39%
Limelight LLNW $579,340,400 5.89 4.39 34.17%
AboveNet ABVT $1,311,105,300 52.09 47.18 10.41%
CoreSite COR $280,432,900 16.39 16.00 2.44%
Internap INAP $254,359,000 4.90 4.17 17.51%


Sep 30 Close Jul 1 Open Gain
NASDAQ 2368.62 2109.24 12.30%
S&P 500 1141.20 1030.71 10.72%

Note : CoreSite traded for six sessions in the quarter, and has been weighted accordingly

We weighted each stock's gain by end-of-the-quarter market cap.

Our market-weighted Services Index launches tomorrow at the opening bell with an index value of 100.

DARPA and NIST Research Projects Could Impact Data Center Networks

By Lisa Huff

While many private sector investments have been made in optical communications over the years, none have contributed as much ongoing support as the US federal government has. The Defense Advanced Research Projects Agency (DARPA) and the National Institute of Standards and Technology (NIST) have been at the forefront of funding for decades and are continuing to ante up now.

One of these projects being developed at NIST right now. TERAPIC™: The TERAPIC (Terabit Photonic Integrated Circuits) project has as its goal to "develop technology for optical components that can transmit and receive up to one terabit of data per second over one single-mode fiber, greatly reducing complexity and cost in high-capacity fiber-optic networks."

Privately-held CyOptics and Kotura have partnered on this project to bring Terabit connectivity to data centers and HPC centers by reducing hundreds of individual components "to less than 10." The project is expected to produce an integrated component that can "transmit and receive up to one Tbps of data over one single-mode fiber across transmission distances of up to two-kilometers." The project team has been successful with 100G and 500G prototype devices that transmit up to two-kilometers and continue to work towards Terabit devices.

The PICs will be monolithic arrays of CWDM lasers and detectors that will be automatically assembled in CyOptics' manufacturing facility in Breinigsville, PA. Kotura has developed two integrated silicon photonics chips—data multiplexing/transmission and data demultiplexing/detection that are integrated into the overall transmit optical sub-assemlbies (TOSAs) and receive optical sub-assemblies (ROSAs) from CyOptics.

The TERAPIC project is set to be completed in 2010 so products are expected to be released in 2011, but early adopter customers have yet to be identified so actual revenue-generating opportunities may still be several years off. Especially since they are targeted for Terabit Ethernet, which has not even started on its standardization tract yet. However, the 100G and 500G parts that were developed could be made production-worthy in the interim.

Wednesday, September 29, 2010

Brocade, QLogic and Emulex Endorse 16G Fibre Channel

By David Gross

QLogic (QLGC), Emulex (ELX) and Brocade (BRCD) recently endorsed the forthcoming 16G, or FC-PI-5, standard under development by ANSI. While not unexpected, it was refreshing to see Emulex and Brocade in particular support stand-alone Fibre Channel's upgrade path, without the Fibre Channel-over-Ethernet nonsense both have been pushing the last two years.

SANs and LANs are not merging anywhere outside of PowerPoint presentations, and SAN managers continue to need vendors who have expertise in Fibre Channel and storage networks. While 8G Fibre Channel products are just starting to be deployed now, it seems SAN vendors are at least listening to customers who want to know how they will upgrade their networks in the future, not how they will “converge” them.

Brocade plans to have 16G Fibre Channel switches available next year. Still, I have to wonder how they'll maintain their market share in Fibre Channel as they continue on their misguided over-diversification into all things data networking. The press release they put out on 16G was filled with enough marketing gibberish to discourage any investor who thinks they might have turned a corner.

Some of the odd statements from the Brocade PR include :

“Brocade’s heritage is data center networking”

What??? Brocade's heritage is storage networking. No reason to run away from a segment they've dominated the last ten years to catch onto the flavor-of-the-month.

Then there was this...

“delivering unmatched simplicity in managing data-intensive applications, investment protection for existing SAN infrastructure and non-stop networking”

Non-stop networking, are they borrowing bad catch phrases now from HP? Also, if the products are so simple, why can't they described in plain English?

Equinix Would Make a Terrible REIT

By David Gross

Citigroup put out a note last week arguing that Equinix (EQIX) should consider becoming a REIT in order to close a gap in its market value relative to Digital Realty (DLR) and DuPont Fabros (DFT). This would be a terrible idea though for the collocation provider. Co-lo and data center REITs remain very separate businesses, with different operating and financial requirements, and in many respects they're growing further apart.

The primary financial difference between providing co-lo space and building space is the number of people a provider has to pay. DLR generates over $2 million dollars in revenue per employee, Equinix just $600,000. DLR, DFT, and COR, like most REITs, need to generate NOI (net operating income) margins of about 65% to justify the high cash flow/high dividend strategy needed for that form of business. While less hands-on than managed services, which often entail a lot of software configuration, co-location is not a real estate business. It requires far more internal experts in areas like database management, network technology, and industry-specific sales support than a typical data center REIT has on staff. The additional staffing needed to do this in the co-lo business eats into operating margins, and makes it financially unsuitable for the REIT structure.

A co-lo provider needs to generate more revenue per dollar invested in fixed assets than a REIT to offset the higher staffing requirements. Equinix has been producing about 40 cents of revenue per dollar invested in PP&E compared to around 20 cents for DLR and DFT. Essentially, co-lo is more labor-intensive, less capital-intensive than the data center REIT business, which makes mixing the two very difficult.

It's easy to look at stock valuations and rush to judgments about quick fixes that could temporarily lift the price. But the trend is for co-location providers to lease space from REITs, as Telx is doing with DLR. Perhaps sometime in the future when it slows down its expansion, Equinix could consider paying some kind of dividend, but bringing its productivity and fixed asset ratios into line for a higher yielding REIT would require destroying existing operations, which would cause a lot more problems for the stock than its current valuation gap.

Savvis Best Performing Data Center Services Stock This Quarter

With two trading days left in the quarter, Savvis (SVVS) leads data center stocks with a 45.07% gain since July 1, outpacing runner up Rackspace (RAX) by nearly three points. Terremark (TMRK), Equinix (EQIX), and Navisite (NAVI) are all up over 25% for the quarter.

Among data center networkers, F5 (FFIV) leads the pack, up 51% for the quarter, well ahead of the 2.58% gain posted by Cisco (CSCO). If you include companies whose products connect data center to data centers, Riverbed (RVBD) leads everyone, up over 66% for the quarter.

Tuesday, September 28, 2010

HP, GigOptix Lift Guidance

By David Gross

Two companies that develop data center technologies, one very large, another very small, have lifted guidance over the last couple days.

HP raised its annual bottom line guidance this week from $4.99 to a range of $5.05-$5.15, and lifted next year's revenue guidance to a range 5-7% greater than publicly issued sales expectations a month ago. On the other end of the revenue spectrum, micro-cap GigOptix (GGOX), which makes analog components and ICs for 40/100 Gigabit networks, lifted its guidance for the quarter from $6.8 million to $7 million.

Transitioning Your Data Center from Copper to Fiber

By Lisa Huff

Companies like Corning (GLW) like to tell data center managers that with the advent of 10G, they should be transitioning their network from mostly copper connections to all fiber-optic. But, as many of you probably know, this is easier said than done. There are many things to consider:

1. How much more is it going to cost me to install fiber instead of copper?
2. Do I change my network architecture to Top-of-Rack in order to facilitate using more fiber? What are the down sides of this?
3. Is it really cost-effective to go all-optical?
4. What is my ROI/IRR if I do go all-optical?
5. Will it really help my power and cooling overload if I go all-optical?

It is very difficult to get to specific answers for a particular data center because each one is different. And guidelines from industry vendors may be skewed based on what they are trying to sell you. OFC/NFOEC management has recognized this and asked me to teach a new short course for their 2011 conference – Data Center Cabling – Transitioning from Copper to Fiber will be part of the special symposium Meeting the Computercom Challenge: Components and Architectures for Computational Systems and Data Centers. I invite your ideas on specifics you would like to see covered in this new short course.

Mellanox Up 35% Since Guiding Down on its Last Call

By David Gross

Buoyed by Oracle takeover rumors, Mellanox (MLNX) is coming back from the big hit it took earlier this quarter when it announced earnings and issued weak guidance. It closed yesterday at $19.86, up 35% from its July 22 close of $15.49, day in which it fell more than seven dollars from $22.94.

While I still think the company needs a clearer growth strategy, it remains reasonably valued compared to many data center high fliers, with an enterprise value/annualized revenue ratio of 2.75, and a pristine balance sheet with $230 million of cash and no debt.

When Sun strongly endorsed InfiniBand, no one seemed to care, and the conventional wisdom about Ethernet eventually taking over InfiniBand apps was still going strong. But now that the same architecture and same product lines are under Oracle's ownership, the endorsement is being taken much more seriously, even though the substance is no different - InfiniBand's share of top500 supercomputers has continued to grow at the same pace, and now tops 40%.

I've long argued that Mellanox needs to be a little bolder about associating itself with InfiniBand, because it dominates that niche's market for ICs, while it gets lost among a large crowd in the market for Ethernet cards, in spite of the company trying to get ahead of everyone else with 40 Gigabit Ethernet server NICs. Maybe now that there is growing appreciation for InfiniBand's role in high-speed networking, Mellanox won't be so shy about declaring its leadership in this technology.

Monday, September 27, 2010

New Japanese Data Center Will Expand Equinix's Asia-Pac Capacity by 10%

Equinix (EQIX) announced today that it is building its 3rd Tokyo Data Center, TY3, at a construction cost of approximately $880 per square foot. The new facility will expand the co-lo provider's global capacity by 960 cabinet equivalents. The company had 82,400 cabinet equivalents in its inventory as of the end of 2Q, with about 9,500 in the Asia-Pac region.

Rackspace, Equinix, Akamai, Savvis, Limelight All Near 52-Week Highs

Data center stocks have had a very strong quarter, with Rackspace (RAX), Equinix (EQIX), Akamai (AKAM), Limelight (LLNW), and Savvis (SVVS) all near 52-week highs. The REITs, however, have not shared in many of the recent gains, in spite of CoreSite's IPO last week. Digital Realty (DLR) is up just 1.4% over the last three months, while DuPont Fabros (DFT) has been essentially flat since the third quarter began.

Is this all justified? Rackspace is now trading at roughly 45x annualized earnings, on a top line growth rate in the low 20s, and a bottom line growth rate of 60% over the last 12 months. Equinix is just under 4x revenue on a 21% top line growth rate (excluding Switch & Data), and the company is barely breaking even. Savvis is up 37% over the last three months and hasn't had any top line growth over the last 12 months. While the industry's fundamentals remain very strong, there's not enough growth to support a 5x multiple on revenue.

IBM's Acquisition of Blade Network Technologies a Smart Move

By David Gross

Rumors were flying last week that IBM (IBM) was about to buy Brocade (BRCD). Well, that still hasn't happened, but the large technology supplier is going to make a much more sensible deal - acquiring top-of-rack switch maker Blade Network Technologies.

Cisco's (CSCO) proprietary VLAN protocols have far less of an impact in the top-of-rack segment than they do elsewhere, which has made this small portion of the Ethernet market more friendly to alternate suppliers than the traditional corporate LAN market. Moreover, many vendors missed out on this segment because they were so focused on building massive carrier Ethernet boxes with overcooked operating systems, while Blade and Arista have kept things simple and straight forward by focusing on latency, not accommodating massive routing tables.

In addition to an existing sales relationship with IBM, Blade has an OEM arrangement with Juniper (JNPR), who was an investor in the company. Financial terms of the acquisition have not been disclosed.

PLX Technologies Acquires Teranetics for $36 Million

PLX Technologies (PLXT), a leading supplier of PCI Express silicon, is acquiring privately-held 10GBASE-T chip designer Teranetics for $36 million. While this move solidifies PLXT's presence as a data center networking chip supplier, it is not a great exit for Teranetics' investors. The company has raised $61 million since 2003, and along with Solarflare and Aquantia, makes up the leading triumvirate of 10GBASE-T chip suppliers. While 10GBASE-T power consumption per port has dropped as low as 3 watts as production has shifted to smaller process geometries, the technology is still advancing at a slow pace in the data center.

Teranetics key customers include Mellanox (MLNX), to whom it supplies the PHY for that vendor's 10GBASE-T LOM product. PLXT reported revenues of $30 million last quarter and has a market cap of $139 million.

The Networking Index

By David Gross

Saturday I posted about our new Services Index, which will be a market-weighted average of the 13 data center service providers listen in the ticker in the right column. With the end of the quarter coming this week, we're going to start that index at 100.0 as of Friday's opening.

In addition, we will also start a data center networking index at the same time, although that will be price-weighted to avoid being dominated by Cisco, whose $123 billion market cap is four times the value of all the companies in our services index.

While most networking companies now are eager to jump on the data center bandwagon, we've picked ones with some kind of operating history working in the sector, and have looked specifically at how they are contributing to Fibre Channel, 40/100 Gigabit Ethernet short-range port types, InfiniBand, PCI Express 3.0, load balancing, data center-to-data center WAN Optimization, and cabling. With so much development now taking place at the chip/component levels, many companies in the index make chips or optical parts, not full systems. The 25 companies in this index will be:


Saturday, September 25, 2010

Aggregate Market Cap of Data Center Service Providers Tops $30 Billion

By David Gross

In preparation for the launch of our Services Index, I've calculated the aggregate market cap of 13 data center services providers at $30.23 billion, with REITs, co-lo, managed hosting, and cloud providers accounting for just over half of that at $15.93 billion. Over half of the difference between the two is AKAM's $9.24 billion market value. It also includes LVLT, INAP, and SVVS whom I didn't count among the "pure" providers because they also sell transit and office-to-office bandwidth services.

EQIX 4.66
DLR 5.29
DFT 1.47
RAX 3.06
SVVS 1.22
LVLT 1.61
AKAM 9.24
NAVI 0.13
TMRK 0.70
LLNW 0.50
ABVT 1.34
COR 0.75
INAP 0.26


EQIX 4.66
DLR 5.29
DFT 1.47
RAX 3.06
TMRK 0.70
COR 0.75


Introducing the Services Index

By David Gross

AboveNet has a marketing slogan which claims that "data centers are the central offices of the 21st century". If this is really true, then data center providers are the telecom providers of the next 100 years. With this in mind, we are going to start tracking their overall health further with the introduction of the Services Index. This will be a market-weighted benchmark of the 13 providers we track in the ticker in the right column - specifically EQIX, DLR, DFT, RAX, SVVS, LVLT, AKAM, NAVI, TMRK, LLNW, ABVT, INAP, and COR.

Stay tuned for more news on the index!

Friday, September 24, 2010

25G/40G VCSELs Driving Short-reach Optical Interconnects

By Lisa Huff

Just a few years ago, laser designers were struggling with stability of their 10G VCSELs. But now, at least one, VI Systems GmbH, claims it will have production-ready 40G VCSELs within the next few years. The German start-up has developed two products it believes will take VCSELs beyond 10G applications - a directly-modulated (DM) device and an electro-optic modulated (EOM) DBR VCSEL. Both are short-wavelength (850nm) lasers.

In a recent press release, VI Systems explains that it “developed the VCSEL products at a wavelength of 850 nm along with a range of extremely fast integrated circuits based on the SiGe BiCMOS (silicon-germanium bipolar junction transistors in complementary metal-oxide-semiconductor) technology. The company uses a patent pending micro-assembly platform for the integration of the opto-electrical components and for alignment to a standard high performance multi-mode glass-based fiber.” The start-up has been presenting data supporting its claims of highly stable devices for more than a year now. It gets there by changing the laser active region material and structure to InAs quantum dot (QD).

Not only is VI Systems working on innovative laser structures, it has also developed new electro-optic integration methods to further reduce the cost of these devices.

I’ve noted in previous posts how VCSELs are the key to low-cost optical networks in the data center. These new VCSELs and packaging methods would bring an even more cost-effective “serial” solution for 40/100G. They could also be used for very short-reach optical connections like for chip-to-chip, on-board or board-to-board. Perhaps these inventive products will rival Avago’s MicroPOD and Luxtera’s OptoPHY (also in previous posts). Based on the presentations that VI Systems has released, it sure appears that its management completely understand the needs of both the data center and optical interconnect markets so could very well give incumbents in the industry some competition.

Thursday, September 23, 2010

CoreSite Nearly Turns Over Its Float In 1st Trading Session

By David Gross

CoreSite (COR) ended the day up a penny from its initial offering price, but with over 15 million shares traded, it nearly turned over its float of 16.9 million shares. The stock got off to a good start, and was up 3% to 16.48 by 1:30 this afternoon, but fell back the last two hours of trading right back to its IPO price.

CoreSite Up 2%, 3 Million Shares Traded

COR is off to a solid start in its first 45 minutes as a public company. It's up to 16.30 from an open of 16.00. Volume has settled down after 1.5 million shares traded after the opening around 9:48 this morning.

CoreSite Raises $270.4 Million, Prices at Midpoint, IPO's Today

By David Gross

CoreSite (COR) priced at $16, in the middle of its projected $15-$17 range, and begins trading today on NYSE. The offering raised a total of $270.4 million for the data center REIT, which will be the third publicly traded stock in that category, along with DLR and DFT.

We have added CoreSite to the list of stocks tracked in the ticker in the widget in the right column.

Finally, Some Common Sense About FCoE

By David Gross

Was reading this Fibre Channel over Ethernet story at Search Networking, and found it to be a pleasant surprise compared to all the hype the protocol has generated. Specifically, I liked the following quote from a CCIE, certainly echos the feedback I've gotten about FCoE:

"The guys who run Fibre Channel SANs like having a network of their own," Banks said. "When you talk to a storage guy and say, 'Hey, we want to take your network and bring it onto this big Ethernet network that's been doing data for a long time,' they're going to get worried about performance and a loss of control."

Along with the high cost of "converged" hardware, this is a major problem for FCoE. And is exactly what slowed other attempts to force networks together. The talk of LAN/SAN integration with FCoE reminds me a lot of the LAN/WAN integration we heard about in the 90s with ATM, or the LAN/MAN integration FDDI was supposed to deliver. I wouldn't be surprised if ten years from now we are having a good laugh over this misguided technology.

Wednesday, September 22, 2010

Intel’s Light Peak OR USB 3.0?

By Lisa Huff

After Intel’s (INTC) Developer’s Forum last week, there is renewed interest in Light Peak. For those of you that don’t remember one of my first blog posts, Light Peak is an Intel-developed optical interconnect techonolgy that uses a new controller chip with LOMF and a 10G 850nm VCSEL-based module with a new optical interface connector that looks very similar to the one used in Avago Technologies (AVGO) MicroPOD transceiver. Light Peak is aimed at replacing all of your external connectors on your PC including USB, IEEE 1394, HDMI, DP, PCIe, etc. It is also targeted at other consumer electronic devices like smart phones and MP3 players.

Many in the industry think Light Peak is intended to replace USB 3.0 even before USB 3.0 is finished being standardized. I tend to disagree. USB 3.0 is a 5G data rate and to me, will bridge the gap between existing USB 2.0 (480 Mbps max) and the 10G that Light Peak can provide. Just because they are being developed at the same time, doesn’t mean they will make it to production simultaneously.

While Intel is now saying that 2011 will be the year for Light Peak to take off, I’m still skeptical. There may be some really high-end applications like video editing that may need this bandwidth, but your run-of-the-mill PC user isn’t going to want to pay the extra money for it when you probably won’t be able to actually detect the improvement. And, what might be more important – what kind of power consumption difference is there and how does this affect battery life?Or is this technology not meant for laptops? I’m not sure these questions have been sufficiently answered yet.

Tuesday, September 21, 2010

Morningstar Initiates Credit Coverage of Equinix with a BB Rating

By David Gross

Morningstar is now covering Equinix's (EQIX) credit, rating it BB. However, the debt rating agency made some odd comments in analysis, including:

"On the downside, there are numerous companies providing data center space, including giant network owners like AT&T T and Verizon VZ and other specialists like Savvis SVVS"

What? Since when have AT&T and Verizon even made a dent in Equinix's business? Or Savvis for that matter? They've been competing against one other for over a decade, and during that time Equinix has grown nearly 100-fold - from $13 million in revenue in 2000, to a run rate of $1.2 billion today.

What Morningstar doesn't get is that the key to continued success for Equinix is a focus on network neutral co-location, and at this point the provider is not veering into any dangerous overdiversifications, which are typically the biggest risk with any data center service provider.

Morningstar goes on to say:

"Pricing can come under pressure quickly if supply outstrips demand and data center owners seek to fill up unused space quickly."

In theory, I guess it could. In practice, it rarely has. In the middle of this recession, revenue per cabinet has risen 8% year-over-year. What Morningstar seems to be missing is that part of the reason revenue has been increasing is that underlying power costs are increasing, and those are unlikely to reverse course, regardless of what happens in the overall economy. Additionally, construction labor isn't getting cheaper. These two factors make EQIX very different than telcos which face declining costs.

Yahoo Opens Lockport Data Center

Yahoo held a grand opening yesterday for its new Lockport, NY data center, located about half an hour northeast of Buffalo.  A few stats on the new 155,000 square foot facility:

-1 server per 3.2 square feet of total space
-$968,000 construction capex per square foot
-1 full-time worker per 1,240 square feet
-3.23 construction jobs per square foot

Monday, September 20, 2010

About $3500 for a 10G Port at new UT Data Center

By David Gross

The University of Texas has just opened a new $32 million data center for use by the school's academic and administrative divisions.   Interestingly, the nonrecurring port fee for redundant 10G connections is about $3,500, not all that different from commercial co-location pricing.

1G port fees are still approximately $1,500 from most co-lo providers, although UT seems to be offering quite a deal at $475 per pair.   One factor to keep in mind is that there are no additional cross-connect or backhaul fees for the university's end users.

Sunday, September 19, 2010

CoreSite to Debut This Week

by David Gross

DLR is near its 52 week high, DFT is barely 10% off of its annual peak, and it looks like they will be getting a peer this week, as CoreSite prepares to becomes the 3rd publicly traded data center REIT.   Like the other REITs, CoreSite is planning to list on NYSE, and is expected to trade under the ticker "COR".

According to the S-11 CoreSite filed last week, the company is planning to offer 16.9 million shares, and raise approximately $246 million.   Like most REITs and co-lo providers, the company's portfolio skews heavily to major markets, with the Boston, NY, SF, DC, and LA metro areas accounting for 98% of its annualized rent.   This emphasis on a handful of top regions is one reason why data center geographical markets cannot be treated like office markets, although this is not stopping old school real estate firms from trying to force fit data centers into vacancy models better suited for commercial offices.

Co-location provider Telx, whose S-1 is beginning to show signs of age, has yet to price its shares in preparation for a public offering, though it would be surprising if it remained on the sidelines much longer, particularly with its peer Equinix (EQIX) up 20% since July.

Friday, September 17, 2010

10GBASE-T versus 10GBASE-SR – Tradeoffs in Costs Make the Difference

By Lisa Huff

The 10GBASE-T standard was finalized some four years ago now, but, as I’ve mentioned before, equipment using these ports is really just starting to be deployed in actual networks. The main reason being that you couldn’t get a switch with these ports on it. So early implementations of 10G are 10GBASE-SR, 10GBASE-LR or 10GBASE-LRM, with the vast majority now being the SR. But now that switch manufacturers the likes of Blade Networks, Cisco (CSCO) and Extreme Networks (EXTR) have products offering up to 24 ports of 10GBASE-T, the market dynamics may change.

With Ethernet, history tends to repeat itself so let’s take a minute to review what happened at Gigabit. Early products were 1000BASE-CX, SX and LX because the 100m 1000BASE-T had not yet been standardized. But, as soon as it was and the switch manufacturers started adopting it, it took over the shorter-reach Gigabit Ethernet market. In fact, it still dominates today.

So, why would 10GBASE-T be any different? Well, my belief is that eventually, it won’t. Even though data center managers concerns have shifted from space, to power availability per rack and cooling hot spots, when they see the price tag difference between SR and T (still about 2:1 per installed port), it causes them to pause and rethink the T scenario. So although data center managers want to reduce their headaches with fat CAT6A cables, most are still not willing to pay that much more for the optical solution until they have to because of distances. So even though the T ports may push electricity bills up, for most, the increase isn’t significant enough to justify the up-front cost of SR.

Wednesday, September 15, 2010

More on 10G Cabling in the Data Center – 10GBASE-T versus 10GBASE-CR

By Lisa Huff

Even though Augmented Category 6 cabling has been around for more than four years and has actually been installed in many data centers, it has really just started to be used for 10G applications. 10GBASE-T ports have supposedly been available from both server and switch manufactures for over a year, but trying to actually get your hands on them has been difficult. And, when you look at the power dissipation and latency specs, data center operators are choosing 10GBASE-CR (SFP+ direct-attach copper) instead. So what are the tradeoffs?

SFP+ direct-attach copper is designed to be used for short interconnects – 15m or less – which makes it a perfect solution for ToR switches. It’s also available with 30AWG cable, which makes it much more manageable than the 24AWG (or sometimes 23AWG) CAT6A – especially if you’re using a shielded CAT6A solution. Because it plugs into a SFP+ port, if you purchase a passive cable (no DSP chip to clean up the signal), there is essentially no added power consumption. If you add an optical SFP+ module, it’s still less than 1W. The 10GBASE-T port still burns about 3W. Latency of a 10GBASE-T solution could be as much as 2.6 ┬Ásec, while SFP+ is 0.3 ┬Ásec max. For some firms, like trading companies, this could mean millions of dollars made or lost.

With SFP+ DAC, you could buy a ToR solution today that will accommodate both short connections within the rack and up to 40km outside the rack. If you choose a 10GBASE-T product, it will currently only cover up to 100m within one switch – there are no products available so far that have both 10GBASE-T and SFP+ ports. But, I’m sure these are being developed. And, once they are, it will most likely be more cost effective to use 10GBASE-T within the rack with LOMF SFP+ (or maybe 40G QSFP+) uplinks outside the rack. Just like what we have today with Gigabit – 10/100/1000BASE-T within the rack and SFP uplinks.

Both SFP+ DAC and 10GBASE-T products will be needed in the long term – 10GBASE-T for inexpensive connections and 10GBASE-CR (SFP+ DAC) for lower latency and lower power consumption.

Monday, September 13, 2010

10G Copper versus Fiber – Is Power Consumption Really the Issue?

 by Lisa Huff

For decades now, fiber has been slated to take over the data networking world, but somehow, some way, copper keeps reinventing itself. But are the ways in which copper can compensate for its lower bandwidth capacity coming to an end at 10G due to what seem to be astronomical power consumption issues? Probably not. I have listened to the rhetoric from the fiber-optic companies for more than five years now, and have conducted my own research to see if what they say is true. Their argument was that at 8 to 14W per port, copper is just too costly. But, now that the chips have reduced power consumption to less than 3W per port, 10GBASE-T is a viable data center networking solution. Actually, even at 14W per port, it was viable, just not practical for switch manufacturers to incorporate in their designs because they couldn’t get the port density they needed and actually have room to cool the devices. Now that doesn’t seem to be an issue, evidenced by 24-port 10GBASE-T configurations that have been released by all the major players.

I believe decisions on copper versus fiber will be made around other parameters as well, such as latency. In a recent study, Data Center Cabling Cost Analysis - Copper Still Has Its Place, we looked at the cost of 10G copper versus fiber and added in the higher power consumption. Using the specific example focused on a rack of EoR Cisco switches, copper was still more cost-effective even when considering higher electricity costs.

But our next area to study will be a rack of servers with a ToR switch. In this scenario, the power consumption difference may be enough to justify the cost of installing fiber over copper. The above referenced report and this next research are part of a series of research reports for our Data Center Network Operator Service.

Wednesday, September 8, 2010

Is it Time to Sell F5?

by David Gross

Few stocks have performed like F5 (FFIV) over the last year. It's risen over 150% the last 12 months, while NASDAQ has eeked out a 9% gain, and rival Cisco (CSCO) has fallen 5%. Its other chief competitor, Citrix (CTXS), has risen just 71%, less than half as much as F5,  renewing calls that the stock is overvalued, and ready for a fall.

F5's rise over the last few years has climbed a wall of worry. When it hit $26 in April 2009, UBS downgraded it citing valuation concerns. Since then, the stock has risen more than three-fold.  Moreover, the fundamentals have remained very strong, with top line growth exceeding 40%, the cash balance growing and nearing $900 million, negligible long-term debt, a workforce that produces nearly 50% more revenue per head than rival Citrix, and Cisco (CSCO) still struggling to find its way in the load balancing market.

Between 2000 and 2003, F5's annual revenue was essentially flat, bouncing around between $108 million and $116 million. Since then, it's grown eight-fold, with last quarter's revenue coming in at $230 million. With the exception of 2009, when it grew just 0.5% due to the recession, the company's top line growth has come in anywhere between 24% and 64% since 2004. If it hits its guidance in its fiscal 4th quarter for 2010, which ends this month, it will report 34% growth for 2010.

From 2005 and 2007, the company's growth rate was nearly cut in half, from 64% to 33%, before the recession began. While many are expecting lower growth in 2011, Wall Street hasn't treated the stock well when its growth rate has slowed. From August 2007 to March 2008, when growth was decelerating, but still over 20% year-over-year, the stock lost 58% of its value, a period during which NASDAQ fell just 15%.

The current enterprise value/annualized earnings ratio of 41 is certainly high, but nothing like its obscene late 1999 valuation when the company was 96% smaller than it is now, generated nearly a quarter of its revenue from a soon-to-be-bankrupt Exodus, but had a market cap just 20% lower than today's soaring value. Moreover, many of the product lines it competed against then, such as Cisco/ArrowPoint, Nortel/Alteon/Radware, and Foundry/Brocade, are the same, yet have struggled amid mergers, acquisitions, and management focus on other product categories. So while the long term story for F5 continues to look very good, the sudden price drops that have accompanied its periods of slowing growth should make investors think twice before getting in at these levels.

Tuesday, September 7, 2010

Why Fiber Polarity Matters, Part 2

by Lisa Huff

If you’ve read my previous posts on the subject, you know that polarity can be a tricky matter and it’s even more complicated when you try to choose it for your data center cabling. You really have to choose based on several factors:

1. Patch cords – method A has two different patch cords that you have to stock, but the upside is that it’s pretty simple to follow where the signal is going and if you happen to be out of one type of patch cord, you can really take the one you have and just flip the fibers as a temporary fix until you can get the other patch cords. Of course this isn’t recommended, but if you’re in a bind and need to get up-and-running right away, it will work. With methods B and C you have the same patch cord on each end so no need to worry about this, but if you happen to have the wrong cassettes or backbone, nothing will work and you'll have to wait to get the correct ones.
2. Cassettes and backbone cables – you need to make sure you buy all of one method of polarity or your system won’t work. If you’re concerned about supply, all three polarity methods are available from multiple vendors, but Method A is “preferred” by most.
3. Upgradability – this is where it can get dicey. Typically your pre-terminated assemblies are running Gigabit applications today and a few may be running 10G. Any of the polarities will work at these data rates. But when you move to 40/100G, methods A and B have straight forward paths, while C does not. Also, you’ll want to make sure you use the highest grade of LOMF available, which is OM4 – this will give you the best chance of being able to reuse your backbones up to 125m. If you need something longer, you’ll need to go to SMF.

If you are thinking about installing pre-terminated cassette-based assemblies now for 10G with an upgrade path to 40 and 100G, you need to consider the polarity method you use. Unlike today's 2-fiber configurations, with one send and one receive, the standards for 40G and 100G Ethernet implementations use multiple parallel 10G connections that are multiplexed. While 40/100G equipment vendors will tell you that polarity is not an issue, care must be taken if you want to reuse this installed base.

40G will use four 10G fibers to send and four 10G fibers to receive, while 100G uses either four 25G fibers or ten 10G fibers in each direction. Because 40 and 100G will be using the MPO connector, if the polarity method is carefully chosen, you will be able to reuse your backbone cables. This is enabled by the fact that the IEEE took much care in specifying the system so that you can connect any transmit within a connection on one end of the channel to any receive on the other end.

Those selecting fiber to support 10G now and 40G in the near future need to understand what will be involved in transitioning and repurposing their cable plant. In order to upgrade using method A, you can replace the cassettes with MPO-to-MPO patch panels and MPO-to-MPO patch cords and it will enable flexibility to address moves, adds and changes as well as promoting proper installation best practices. The polarity flip will need to be accomplished in either an A-to-A patch cord or possibly with a key up/key down patch panel.

Method B multimode backbone cables can also readily support 40G applications. For a structured cabling approach, method B will still use a patch panel and patch cords, though as with current method B, both patch cords could be A-to-B configuration. While Method C backbones could be used, they are not recommended for 40G as completing the channel involves complex patch cord configurations.

It appears that 100G will use either the 12-fiber (4x25G) or the 24-fiber (10x10G) MPO connector. With transmits in the top row and receives in the bottom row, the connection will still be best made using a standardized structured cabling approach as described above.

There are many suppliers of pre-terminated optical assemblies including Belden, Berk-Tek, a Nexans Company, CommScope (CTV), Corning (GLW), Panduit, Siemon, Tyco Electronics NetConnect (TEL) as well as many smaller shops that give quick-turn assemblies like Cxtec CablExpress and Compulink.

Thursday, September 2, 2010

New TACC Supercomputer Running 40G InfiniBand

by David Gross

The Texas Advanced Computing Center is expanding its Lonestar Supercomputing Cluster with a new system that will feature 22,656 compute node cores, up from 5,840 in the existing system, connected with a 40G Mellanox (MLNX) InfiniBand network that uses a fat-tree topology.

The new system will feature:

* 302 teraflops peak performance vs. 63 teraflops on the existing Lonestar
* 44.3 terabytes total memory vs. 11.6 TB on the existing system
* 1.2 petabytes raw disk vs. 106 TB on the existing system
* 90 million CPU hours/year to TeraGrid

What's interesting here is that processing capacity and disk space are all rising at a faster rate than maximum I/O speed, which has risen a fairly sluggish 4-fold in the last eight years. This suggests that the I/O bottleneck is likely to become an even larger problem in supercomputing as time goes on.

Wednesday, September 1, 2010

Why Fiber Polarity Matters ( Part 1)

by Lisa Huff

Again, polarity is the term used in the TIA-568 standard to explain how to fiber (wire) to make sure each transmitter is connected to a receiver on the other end of a multi-fiber cable.

Many data center managers are opting to use pre-terminated fiber assemblies due to their higher-quality factory termination, ease of use and quick installation. And many are using 12-fiber MPO backbone cables with cassettes and patch cords to transition to active equipment. When doing this, they choose a polarity method that makes sense for their operation.

Polarity Method A: This is the most straight-forward method. It uses straight-through patch cords (A-to-B) on one end that connect through a cassette (LC-to-MPO or SC-to-MPO depending on what the equipment connector is), a straight-through MPO-key-up-to-MPO-key-down backbone cable and a “cross-over” patch cord (A-to-A) at the other end.

Polarity Method B: The “cross-over” occurs in the cassette. The keys on the MPO cable connectors are in an up position at both ends, but the fiber that is at connector position 1 in one end is in position 12 at the opposite end, and the fiber that is in position 12 at the originating end is in position 1 at the opposing end. Only one type of patch cord is needed – A-to-B.

Polarity Method C: This is the most complicated. There is pair-wise “cross-over” in the backbone cable in this method. A-to-B patch cords are used on both ends, the cassette uses MPO-key-up-to-key-down and the backbone cable is pair-wise flipped so 1,2 connects to 2,1; 3,4 connects to 4,3; etc.

There is a fourth propietary method that I won't go over here since it's proprietary and not standardized.

If the end user does not get this correct and use all of the proper pieces together, their systems will not work. If you don’t understand what I’ve just explained above, you're not alone. There are diagrams in the TIA-568 standard as well as many white papers from leading structured cabling companies explaining fiber polarity in arrayed cabling systems. Here’s a link to Panduit’s white paper that may help. In the next post, I’ll explain how to upgrade to 40/100G and reuse your pre-terminated backbone.