Saturday, April 25, 2009

Why Did Sun Have To Sell?

A bit of a departure from what I intended this blog to be about, but I wanted to discuss what's going on with Sun Microsystems and why it was up for sale. First Oracle and HP were going to purchase Sun and split it up so that Oracle would get Sun's software portfolio for $2 billion and HP would get Sun's Solaris and hardware business. The deal was supposedly blocked by IBM, who then made their own $6.5 billion bid for Sun.

When IBM walked out of talks with Sun, Oracle came in with an offer to buy Sun for $7.4 billion.

But why did Sun need to sell at all? The answer will surprise you. It doesn't have to do with Solaris or SPARC or Java or Open Source. This turned out to be a long post, so let me just give the answer to the impatient. It has to do with Notes. (Yup, you heard me. Notes!)

To help answer the question of why Sun needed to sell, I dug into some of Sun's financial reports to see how the company was doing. The chart on the right shows annual financial numbers from 1985 to 2008. It includes Sun's product, services and net revenues, as well as net income and net cash provided by operating activities.

A lot of people seem to think that the dot-com burst is what hurt Sun. I think blaming Sun's problems on the dot-com bubble bursting is too simplistic. It was the bubble itself that caused the problems. But just like with the latest housing bubble, people didn't realize it was a bubble and started investing/spending money that wasn't real. That's what I think the case was with Sun.

During the dot-com bubble, venture capitalist (VC) money was flowing freely. People could afford to buy anything they wanted for their startups, and what they wanted was Sun hardware. Lots of Sun hardware. Sun likely had to ramp up manufacturing capabilities and increase staff. But when the VC money dried up, there wasn't anything to take it's place, and demand for Sun's servers fell.

At the same time, Linux was maturing, and what a startup could do with Sun hardware and Solaris, they could now do with Linux and Intel based servers. I don't think it's so much that Linux and x86 servers had caught up to Solaris and Sun's big iron servers, but people started being realistic about their needs. A large number of start-ups could probably be hosted on nothing more powerful than your desktop because they never took off. There was no need to build out for some theoretical level of traffic that will never be obtained.

Sun's hardware, and its Solaris operating system are still well regarded and have their place. According to IDC's Worldwide Quarterly Server Tracker, Unix servers accounted 36.2% of server revenue compared to 13.6% for Linux servers. They didn't break down by Unix vendor, but in the past, Sun has usually been in the number one or two spot for Unix server revenues. If you look at job trends on indeed.com, you'll see that demand for Solaris is greater than IBM's AIX and Hewlett-Packard's HP-UX combined in the job market and Solaris jobs are growing.

While Sun's SPARC server revenues have been declining, they still had over $900 million dollars in billings the last reported quarter. That includes their new line of CMT (chip multithreading aka CoolThreads, Niagara, T1, T2, T2 Plus) based servers which is seeing revenue growth in the SPARC line, with revenues nearly doubling to $1.12 billion in FY08 compared to $608 million in FY07.

While Sun was late to the Intel based server game, they have created some very innovative hardware including the first 8 socket Quad Core Opteron Server, the X4600. I like this video of Andy Bechtolstien demonstrating the X4600 server but the marketing produced video of the x4600 might be more informative.

Sun's blade systems have also been showing very good revenue growth according to the latest IDC report.

5 Straight Years of Losses

OK, that all sounded good, but Sun did have 5 straight years of net loss following the dot-com bubble bursting (as was seen in the chart above.) The largest being in 2003. One thing that former CEO (current Chairman) Scott McNealy used to always say is that it's always been cash flow positive. If you look at the net cash provided by operating activities, that figure has always been positive. Even in recent quarters. The last two quarters though, they took some big hits in cash flow from investing activities to total a loss of over $630 million from investments. The last 3 quarters had negative cash flow.

If you look at net revenues, they've also started to recover. One thing many people kept saying is that Sun needed to reinvent themselves and it really looks like they did. While server revenue continues to decline, they made up quite a bit in services revenue. Now services revenue makes up almost 40% of Sun's total revenue.When you also look at all the great software Sun has been putting out, it has become much more than just a hardware company, though hardware is an important part of it's business.

They have fallen quite a bit from their dot-com highs, but revenue is still impressive at nearly $14 billion annually.

Unfortunately, the latest economic crisis is hitting Sun hard, just as they were getting back on their feet with 2 back to back years of positive net income.

Some people think that Sun didn't downsize quickly enough after the dot-com bubble, and I think that might be a fair assessment. Sun has the money coming in, they just aren't keeping it, and in recent quarters had some big losses.

The Real Reason Sun is Selling is Because of Notes

At least that's my feeling. For the most part, Sun has a lot of cash, large revenues, demand for it's products, a large customer base, they are innovative, they have some good partnerships, so why are they selling?

Sun is selling because of Notes! At least that's what it looks like to me.

You may be confused, but also thinking that's why IBM was so interested in Sun, but no, I don't mean Lotus Notes. During the dot-com era (1999), Sun issued $1.5 billion dollars in unsecured senior debt securities in four tranches. The only remaining tranch is $550 million in Senior Notes that mature in August 2009. These notes pay interest semi-annually at a rate of 7.65%. When they reach maturity this summer, Sun will need to pay back the $550 million.

This is really bad timing. With Sun's revenue posting losses again, it's going to eat into their cash reserves. While they should be able to cover it, it's going to be a big dent. If they don't cover it, it's going to kill their credit rating (which isn't that great right now) and would make it very difficult for them to raise capital in the current economic climate.

In addition, Sun has $700 million in Convertible Notes issued to KKR PEI Solar Holdings, II, Ltd. Half due in 2012, the other half in 2014. While those notes aren't looming over sun like the $550 million due in August are, it's something to consider.

Sun has also been restructuring lately (laying off employees). It will save the company plenty of money, but right now they have to deal with severances and other associated costs, which just makes the situation worse. Had they restructured earlier, and gotten the costs out of the way by now, they would have been in a better situation. The executives at Sun might have not anticipated just how inflated their sales were as a result of the bubble and that those days were never coming back.

Mind you, I'm not a financial analyst, expert, or someone you should trust with your money, so I may be completely off base here. I also haven't seen anyone else reference this situation, but it looks to me like that's the kingpin here. It's not that alone, it looks like a perfect storm of financial misfortune and bad decisions. Sun borrowed during the good times, struggled, then came out of it and would have had no problem paying off the notes. But the current conditions hurt them again. They can't borrow to pay off the note. Getting new investors to pay off old investors is what got Bernie Maddoff into an orange jumpsuit.

They could probably afford it, but combined with the expected revenue shortfalls and the costs of restructuring too late, it would likely kill their credit rating, which Standard&Poor's already put on watch. You can't run a multi-billion dollar business, especially in hard times, with bad credit.

It's pretty sad, had the notes been due just a few quarters earlier, things might have been different.

Like other companies, Sun doesn't keep all their money in cash. Only 11% is in cash and cash equivalents. In addition, they have $268 million in asset backed securities, including $118 million in mortgage backed securities. We know how shaky those have been. Even though 80% of what Sun holds is rated AAA by S&P, many MBS ratings have been questionable. The expected weighted average maturity of their asset backed securities is more than 2 years, so that's not money they can tap into right away.

How could Sun have Saved Itself?

There are a lot of things Sun could have done. At their lowest point post bubble, Sun's annual revenues were $11 billion dollars. That's a whole lot of money! They just didn't use it effectively. The probably should have laid off more staff earlier. In 2002, when their revenues fell by over 30%, they should have laid off 30% of their staff right away or found other ways to cut expenses. They did something to account for losing 1/3 of their revenue, but clearly not enough.

In my opinion though, the biggest mistake Sun made was buying MySQL for $1 billion. That money could have come in handy to keep them afloat during this time. Half would have gone to pay off the Senior Notes, the rest to deal with the financial crisis. Since so many of Sun's customers are in the financial sector, maybe they could have told them that the party was going to be over soon. Some of them actually knew and started making money shorting the housing market.

MySQL revenues were only around $50 million a year at the time Sun was buying them. The price tag of MySQL AB was 20 times revenues! That's just insane. Sun is selling itself for 50% of revenues.

Sun should have continued partnering with MySQL AB, but focused more on PostgreSQL. PostgreSQL is a more feature rich open source database. It doesn't have the market share that MySQL has, but converting that market share into Sun customers that buy expensive hardware and justify the $1 billion price tag is risky and even if it works, would take too long.

With PostgreSQL, they have a database that could get clients off of Oracle and DB2 and Sun has already invested in the software and had some success with Solaris/PostgreSQL over Oracle and Linux.

IBM made the right move to invest in PostgreSQL based EnterpriseDB after Sun bought MySQL and again after Oracle announced it will be buying Sun.

While there were likely many bad decisions, the timing and cost of this one, seems to be the worst in my view.

Friday, March 27, 2009

Great deal on Dual, Dual-Core Opteron Server

There's currently a great deal on a dual processor motherboard on Newegg.com so if you're looking for a dirt cheap 1u server and willing to build your own server. For a single cpu setup with 8 gigs of ram and 1 hard drive in a 1u non hot swap chassis the components are less than $600. Here's the components you will need:

TYAN S2912G2NR Extended ATX Server Motherboard

The base of this build is the TYAN S2912G2NR server mother board, which is currently selling for $69.99. The board officially supports 2 Dual Core AMD Opteron processors but some people have gotten it to work with some of AMD's new Quad Core processors, though it's not officially supported and may cause you problems with certain power management features. There are 8 240 PIN DDR2 667 memory slots for up to 64GB of RAM if you're using two processors. The board only supports PATA and SATA II drives but you can hook up 6 of them with the onboard controller and select RAID 0/1/0+1/g and JBOD.

The reason the motherboard is so cheap is because it is being EOL'd. Likely because of the problems with quad core processors. If 4 cores (2 dual core CPUs) is good enough for you, then this is a great deal. The S2912 board is being replaced by a new model that has support for the Barcelona and Shanghai Opteron CPUs. Tyan Thunder n3600R (S2912-E) Server Board - nVIDIA - Socket F (1207) - 1000MHz HT - 64GB - DDR2 SDRAM - Extended ATX

Dual Core Opteron

If you want to be safe, the Dual Core Santa Rosa processors are a good fit for this board. Below is a selection of processors that will work with this board along with current prices (as of today):

CPU Cooler

This motherboard has a 4.1" mounting pitch which is less common. So you need to make sure you pick the right heatsink. A good one for this board is the Thermaltake 1U CPU Fan - 70mm - 4800rpm - 1 x Ball Bearing, 1 x Sleeve Bearing

Memory

You'll need a minimum of 2 sticks of ram per processor. 4GB DDR2 Ram is affordable enough to use these days at around $80/stick. Kingston 4GB 240-Pin DDR2 SDRAM Server Memory Model KVR667D2D4P5/4G is a good choice to use with this board.

Hard Drives

For servers in a raid configuration, I like Western Digital RE3 Drives. Below are options you can use with this board.

1u Chassis

If you don't care about having hot swap drives, the NORCO RPC-170 Black 1U Rackmount Server Case is a good deal at $79.99. You'll also need an Athena Power AP-U1ATX40L 1U 400W Server Power Supply at around $95.

Hot Swap Chassis

Because of the size of the motherboard, options are limited for hot swap chassis. Chenbro has a 2U chassis that will support 4 Hotswap 3.5" Drives. The chassis includes a power supply.

Alternately, you can go with 4 2.5" hot swap drives in this 1U chassis from Athena Power which includes a 500W power supply. For a raid setup, the 2.5" Velociraptor 10K drives are very good and provide better performance than the 3.5" Drives. Not quite SAS but pretty darn close. They're a bit pricey but great for disk heavy applications such as databases. They are available in 2 different capacities, 150GB and 300GB. Put them in a RAID 1+0 configuration, or a mirrored pool if you're using ZFS and you'll get 300GB or 600GB of capacity and incredible performance.

Sata Cables

The position of the SATA ports on the board is a bit awkward. The included 18" SATA cables that come with the motherboard won't reach drive bays that are on the right hand side of the chassis. For this reason you should pick up some OKGEAR 36" SATA ll Cables depending how many drive's you'll be using.

Wednesday, March 18, 2009

Buying vs Building a Rack Mount Server

The savings you can achieve if you build your own rack mount servers is big. This can be an attractive choice for many, but not all.

The site is still being developed but I wanted to start off the blog with going over some of the benefits and drawbacks of building your own servers. The cost savings can be tremendous, but there are things you need to consider.

What size server can I build?

Theoretically, you can build anything from a 1U single processor server up to a 4 processor 8U server. As the servers get more complex, there might be some real benefits of buying from a good manufacturer such as Sun Microsystems, Hewlett-Packard or IBM.

For that reason, this site is going to focusing primary on entry level rack mount servers. Any one who can follow instructions will be able to build a rack mount server in this class. With so much built into modern server motherboards, all you need to build a 1u server is a chassis, motherboard, cpu, memory and disks. as for tools it's nothing special. Just a screwdriver.

What you lose if you build your own server.

You can save about 50% of the cost of a server if you build it yourself, but don't be deceived into thinking that that server makers are marking up their servers by 50%. Profit margins on entry level servers are pretty thin. A good chunk of the price goes into support costs.

You buy an X2100 M2 server from Sun, for example, and even with the standard 1 year warranty that comes with the purchase, you get next day part replacement. Anything fails and you have one vendor to deal with to get replacement parts. For an extra $400 you upgrade that to 3 years same day replacement or for $700 you can get 2 hour replacement for 3 years. During the warranty period, and usually for a certain period of time they will have replacement parts available.

Need operating system specific support? That's over $1k for 3 years, but it's an option you don't generally get with building your own servers. When you build it, you maintain it.

But you're not completely shut out if you purchase your own components. Most of what you put into your server will have a manufacturer's warranty. It's a little bit more of a hassle and some manufacturers will not give you next day service, let alone 2 hour service, but you can manage hardware failure costs when a component fails.

There are other ways you can deal with hardware failures that might even give you a better response than 2 hours. This tip comes from how construction contractors buy power tools. It's nice to have an expensive Festool or Hitachi and they have their merits, but for the same price you could buy 3 or 4 Ryobi or Ridgid tools. These aren't the junk $20 tools people pick up at Harbor Freight either.

You can do the same with servers. The components in the servers that will be on this site aren't no-names or knock offs. They're from reputable manufactures that sometimes even produce the components that go into big name servers.

Lets take another look at that X2100 M2 server. It is similar to the 1u Quad Core Opteron Server that's currently on the front page of the site except the Sun server is only a Dual processor box but it has a remote management card and 4 network ports. With 2GB of Ram and 1 500GB drive it costs $1,537.

The 1U Quad Core Opteron Kit specified on the home page, with the same amount of memory and disk space costs $816 as of right now. That's a savings of a little over $700. A decent chunk of that is the cost of the chassis and power supply. The chassis is just a metal box. If it's safe and secure in a rack it should outlive the rest of the components. For that $700 you saved, you can buy an extra motherboard, CPU, heatsink, powersupply, drive and ram. If something fails, you have the parts on hand.

It's like you're getting the same day service for free. If the part is still covered by the manufacturers warranty, you can still get a replacement, but while you're wrangling with the manufacturer, your server is still up and running.

For some people, the support their vendor provides is worth giving up the savings. If your $1,000 server generates $1,000 a day in revenue, the savings is trivial and you might like the peace of mind vendor support brings.

There is something to say about the satisfaction of knowing you built and can maintain your own infrastructure.

Rack Mount Servers Are Loud

OK, this might be obvious to most, but while looking through reviews on various sites for different components, I routinely see negative reviews on components intended for use in rack mount chassis because the reviewer thought it was too loud.

They're usually talking about a fan that was meant for a 1U server chassis. 1U enclosures are very thin and you need good air flow through them to keep the components within recommended temperature ranges.

Fans that do that don't tend to be quiet. Don't even think of putting something with a fan, that was meant for a 1U server chassis in your home gaming system that's right next to your bed or under your boss's desk.

The following video shows a typical small office server room (with pretty bad cable management on the racks) to give you an idea what it sounds like.