How many Google machines

An inter­est­ing tid­bit com­ing out of the Google S-1 fil­ing is that they have spent about $250 mil­lion on hard­ware equip­ment. From there, we can get a few guesses at the mag­ni­tude of the Google sys­tem. Based on quick back of the enve­lope cal­cu­la­tions, it looks like Google is man­ag­ing between 45,000 and 80,000 servers. Here’s how I arrived at this conclusion:

Accord­ing to cal­cu­la­tions by the IEEE, in a paper about the Google clus­ter, a rack with 88 dual-CPU machines used to cost about $278,000. If you divide the $250 mil­lion fig­ure from the S-1 fil­ing by $278,000, you end up with a bit over 899 racks. Assum­ing that each rack holds 88 machines, you end up with 79,000 machines.

How­ever, one must rec­og­nize that equip­ment is not all CPUs. As a result, you must dis­count the fig­ure of $250 mil­lion to account for routers, fire­walls, machines for employ­ees, etc… So let’s assume for a minute that only about $200 mil­lion is going to the CPUs. That still leaves us with 719 racks or a bit over 63,000 machines.

Even if we dis­count other equip­ment to be cost­ing $100 mil­lion, we end up with a bit over 31,654 machines on 359 racks.

So how much pro­cess­ing power is that? Well, once again, the Google clus­ter doc­u­ment pro­vides some inter­est­ing tid­bits. Per the doc­u­ment, the racks that were used were

88 dual-CPU 2 Ghz Intel Xeon servers with 2 Gbytes of RAM and an 80-Gbytes hard disk.

That means that, on the low end, the Google clus­ter has the fol­low­ing stats:

• 359 racks
• 31,654 machines
• 63,184 CPUs
• 126,368 Ghz of pro­cess­ing power
• 63,184 Gb of RAM
• 2,527 Tb of Hard Drive space

In the mid­dle range of my esti­mates, the clus­ter would have:

• 719 racks
• 63,272 machines
• 126,544 CPUs
• 253,088 Ghz of pro­cess­ing power
• 126,544 Gb of RAM
• 5,062 Tb of Hard Drive space

And on the high end of my estimates:

• 899 racks
• 79,112 machines
• 158,224 CPUs
• 316,448 Ghz of pro­cess­ing power
• 158,224 Gb of RAM
• 6,180 Tb of Hard Drive space

Assum­ing that the 1Ghz chip is going at about a third the gigaflops of a 2Ghz proces­sor (3.3Gflops), we can then guess at the size of the Google super­com­puter. Just for the sake of argu­ment, let’s go with 1 Gigaflop per proces­sor. This means that the Google super­com­puter has about 126 ter­aflops of power on the low end of my esti­mates, 253 ter­aflops on the mid­dle end, and 316 ter­aflops on the high end. This would eas­ily put it on top of the list of fastest com­put­ers in the world.

Any way you slice it, that’s a lot of power.