If you’ve been in IT long enough, you’re bound to hear the phrase “never underestimate the bandwidth of a station wagon filled with backup tapes.” This was especially true back in the days of dialup connections and leased lines. How does it scale today? First, we have to decide what kind of media we’re going to use. Tape drives are not nearly as common as they once were and their storage density really isn’t impressive. Instead, lets use the most storage dense media that mere mortals can purchase retail (at the time of this writing): The SanDisk 64GB Ultra MicroSDXC.

Now we have our storage worked out, what are we going to haul it in? These days you can’t really get a good sturdy station wagon, but the modern equivalent seems like it would be the SUV. Since Chevrolet Suburbans have been around for so long, I’m going to pick that. According to Chevrolet.com the 2014 Suburban has 137 cubic feet of cargo space. So, how many MicroSD cards can you fit in there?

First we need to find out how big these little guys are, exactly. Wikipedia says that they are 0.59×0.43×0.039 inches. Now, time to do some math:

First, figure out how many cubic inches a Suburban can hold: 137ft³ = 236736 in³

A MicroSD card is only .01 cubic inches, so if all things were equal you could stuff 100 64gb SD cards into a cubic inch of space! But, that does not seem realistic. In fact it doesn’t even seem remotely possible. Plus, Micro SD cards are oddly shaped. Plus, nothing ever stacks just perfect and we have to put them in boxes of some sort that can support the weight of thousands of cards. So, we’re going to reduce that number by 20% when we’re done.

Given the dimensions of the MicroSD card and the size of a cubic foot, a little math later… 174,646 MicroSD cards per cubic foot. Now trim 20% off and you have 139716 MicroSD cards per cubic foot. Remember, we have 137 cubic feet to fill.

## 19 Million Micro SD Cards

19,141,092 MicroSD cards will fit in a 2014 Chevy Suburban. What does that come out to in Capacity?

1,225,029,888 GigaBytes (Yes, that’s over 1.2 * billion *GB’s!)

Or

1196318.25 TeraBytes

Or

1168.279541016 PetaBytes

Or

1.14089799 ExaBytes

Or

1/12 Google (per xkcd.com)

Or

398,772 3TB hard drives

Now let’s talk about Bandwidth. We’ll use the old “New York to Los Angeles” as a benchmark. According to Google, that’s a 2,790 mi, 40 hour trek. Since bandwidth is generally measured per second, we’ll measure the trip that way too. A 40 hour trip is 144,000 seconds. Now lets measure the bandwidth:

1,225,029,888 GB in 144,000 seconds = 68057Gbps

In comparison, ATT’s OC-768 fiber optic cable transfers 38.486016 Gbps which is only a 1/1768th of the bandwidth of the Suburban full of MicroSD cards.

That’s a whole lot of bandwidth! One thing we haven’t considered though is actually reading and writing the data off of all those cards, and then putting it back in order. That would take far more time than the actual trip itself, but we’re not counting it because the original saying “Never Underestimate the Bandwidth of a Station Wagon Filled with Backup Tapes” didn’t count it either.

So, when you’re trying to move lots of data, Never Underestimate the Bandwidth of a Suburban Filled with MicroSD cards!

## 31 comments

Skip to comment form

Although it’s a fun mental exercise, the practicality is far different. In particular, how long does it take to get the data onto those SD cards? Well, let’s calculate that:

So, first we need to make some assumptions about the transfer speed of the SD card. As I understand it, a class 10 SD card operates at above 50Mbit/sec. So let’s use that as the minimum speed.

If we’re using 16Gbyte SD cards, then we need to convert that to Mbits. I’ll spare you all the math and just say that 16GB == 131072Mb (notice the capital and lowercase B’s).

Now we need to find out how long (in seconds) it would take to fill up the SD card at the minimum transfer speed:

131072Mb / 50Mbps = 2621.44 seconds * 60 secs per minute = ~43 minutes

So, it would take about 43 minutes to fill up a single SD card. The article says that a suburban can fit 19,141,092 MicroSD cards into an SUV. So:

19,141,092 cards * 43 minutes = 823,066,956 minutes

That’s approximately 1,565 YEARS to fill up all the SD cards. So, even though a OC-768 fiber circuit is “only” 38Gbps, using a direct connection between two locations begins providing immediate results. Whereas, SD cards have to be prepped then transferred physically.

You’re also assuming one source and one destination.

How many times is that actually true of an OC-768 connection? Not typically used with one source computer and one destination computer over an extended amount of data.

Anyway, the point is that if you have (at the extreme) 19,141,092 sources at one end and 19,141,092 destinations at the other end (say, moving data between something optimized, like a cluster of smartphones), you can write and read the data at each end much faster than your 1,565 years. At optimum, add 43 minutes at each end.

If you really needed to moves lots of data, you’d have enough readers at each end of the SUV connection and you’d space the SUVs out about an hour apart to allow for proper latency and reading/writing adjustments. Still a lot faster than the OC-768, albeit also a lot more expensive for hardware once you factor in the endpoints.

nooono nono no. if it takes about 43 minutes to fill up a single SD card, then to fill up 19,141,092 SD cards takes…. 43 minutes. all you need is 19,141,092 MicroSD card readers at the source and the destination.

that gives you a transfer speed of 50Mbit/s * 19,141,092. which is more like it!!

@Sharper, I’m assuming one source and one destination because we’re talking about carrying them in a single SUV. If we were using multiple SUVs, then the calculations by the author of the article would be very different.

@lkcl, that’s true. If you have enough SD card readers, you could fill up multiple SD cards at one time. I’ll give you that.

“A MicroSD card is only .1 cubic inches, so if all things were equal you could stuff 100 64gb cards into a cubic inch of space!”

10, not 100.

Actually the mistake is that they’re .01 cubic inches, not .1, so 100 is correct.

I always brace for impact when I post anything with math- I’m not very good at with numbers! Thanks for the correction.

It’s always strange to me how everyone underestimates modern tape. Current (Dec 2012) LTO6 cartridges hold 2.5 TB natively, and with hardware compressions easily achieve real time storage of over 4 TB. Now these are physically larger than an MicroSD card, but the biggest advantage of tape is automated libraries! A standard library can hold between a few dozen and hundreds (or more) of tapes, so reading and writing the data off them is no problem at all (for example Amazon Glacier is LTO based). https://en.wikipedia.org/wiki/Linear_Tape-Open

Interesting. Probably unlikely we’ll develop a 2-4TB MicroSD cards anytime soon. But an automated MicrSD card library… hmmmmm.

I think the largest modern tape at present is actually the Oracle/Sun StorageTek T10000 T2. It’s been out for about three years, holds 5TB natively, and where I work we typically reach compression ratios of at least 2.2:1, or about 11TB per tape. The T10000 T3 tape with 8.5TB native capacity is available to order today, along with the T10000D tape drive that supports it.

Let’s assume we only get 2:1 compression and use the standard Sun shipping boxes (2′ by 2′ by 1′) that hold 20 tapes per box. In 137 cubic feet of cargo space, we’d fit 34 containers, or 6,800 petabytes of data in a manner that the tapes are adequately protected and safe even from minor fender-benders.

The individual tapes are 1″x4.29″x4.29″, or 18.4 cubic inches per tape. If the goal was simply to stuff the SUV full of tapes, we dispense with the packaging, shove them in, and hope for the best despite the crushing weight and danger to the data as the author does. We come up with 13,122 tapes @ 10TB apiece, resulting in 131,220TB of data in that SUV. Given the physically identical T10000 T3 tapes available to order today, you’d fit 223,078TB of data in that SUV. About a quarter the data density of shipping things on SD cards.

Not quite the amazing data density the author gets, but shipping 131 to 223 petabytes of data by SUV remains much, much faster than any link that exists today.

The original “no data connection in the world beats a station wagon full of backup tapes” is equally apropos today: “No data connection in the world beats a SUV full of backup tapes”.

And here’s my reality: we MASSIVE quantities of tapes world-wide every day for this exact reason. No link on the planet can keep up with the data rate, so as long as your situation tolerates 48 hours of latency for read/ship/playback time, and you plan sufficient tape drives for the job, you can enjoy absolutely massive throughput… at a corresponding cost in latency!

Crap, I mis-spoke above. It’s 6,800 TERABYTES, or 6.8 petabytes per SUV if you ship things safely. Typing too fast, not proofreading…

“Cubic square inches”?? Please correct that.

Already done. http://what-if.xkcd.com/31/

Of course, the thing that always annoys me about these comparisons is that they /also/ fail to consider “continuous transmission.” That is, your SUV is just like “one really big packet,” but the thing it’s being compared against sends zillions of small packets continuously. As a result, the distance you’re sending it matters far too much.

What do I mean by that? A fiber link has the same bandwidth, whether you’re going across town or across the country. Sure, latency goes up, but that only affects the networking protocol design, not the actual bandwidth of the medium. Your hypothetical SUV-filled-with-microSDs, however, is 2 orders of magnitude faster making a 24-minute cross-town drive by this methodology, and that seems… counterintuitive at best. Because you can only send one packet in this example, your bandwidth score is unduly penalized by the transmission latency.

To take distance out of the equation, you’d have to instead consider a fleet of vehicles, and make an assumption of how quickly you could load and unload the vehicle at both ends (and thus the intra-vehicle timing). Or, you could try to calculate the available “bandwidth” in terms of how many vehicles could you get on the road simultaneously at reasonable speeds and following distance, without causing a traffic jam. Either of those would take distance out of the equation, *and* tilt numbers further in favor of vehicles over fiber.

Something else to consider, this whole exercise scales to even faster speeds if we even think about not moving this across the country, but even in a local context (ie: offices across the city). The max transfer speed across that OC-768 cable is still the same, but the Suburban goes from a 40 hour drive to a 30 MINUTE drive (even counting stopping at McD’s for a happy meal or something)

30 minutes x 60 seconds = 1800 seconds = approximately .64PB/second

These ~20,000,000 microSD cards weigh about 0.5 g each. So that would be about 10,000 kg. Seems a bit heavy even for a Chevrolet Suburban.

68,057Gbps, that’s about 9 strands of fiber for one way, using modern day tech. Another 9 strands for return traffic. So 18 strands for a bit over 68tb/s of symmetrical bandwidth.

Good point regarding the weight limitation, Mr Mint Ape. Perhaps the station wagon is just too current. If we return to the age of steam and use rail transport, each rail car should be able to carry twice that load, making each rail car a “super-sized” packet. Now for a steam-powered computer …

Hmph. “These days you can’t really get a good sturdy station wagon” Yes, you can. You can even get it with AWD, and there’s a really sporty version. Check out the Mercedes-Benz E-class wagons.

Since I really like wagons, I’ll add to the list. Subaru’s new Outback is a bit of a border case, but there’s also the iconic Volvo V70. BMW sells its 3-series as a wagon in the US. Cadillac CTS offers a domestic alternative. Acura has the TSX. If you’re willing to call vehicles in the Jetta Sportwagen and Audi relatives size category wagons, the field opens up even more.

As a point to the original story — my place of employment has been known to send people with portable hard drives to another datacenter an hour’s drive away as the most expedient way to get date from here to there. Also, many of the online backup services allow you to mail in hard drives, or mail you hard drives if you want to move a lot of data into or out of the cloud.

The payload capacity of a Suburban 2500 is 2828 pounds. 137 cubic feet = 3880 liters, so if the density of the microSD cards is the same as water (just barely float in water), the cards would weigh about 3.9 metric tons, or about 8500 pounds. You could fit them in, but you’d be greatly exceeding the safe carrying capacity.

19,141,092 MicroSD cards * $29 / 2,790 miles = $198,957.59 per mile (excluding gas, wear and tear on the car, and the time of the driver).

19,141,092 MicroSD cards * $29 / 2,790 miles = $198,957.59 per mile (excluding gas, wear and tear on the car, and the time of the driver).

68057Gbps!!

So thats why Google is so interested in self driving cars.

Now extrapolate that to a fleet of autonomous cars, with purpose built data centers in a box that drive around the country picking up petabytes of data in large science facilities, and then distributes them to data centers elsewhere, and special software for loading and unloading these megadisks. In certain scenarios, the cost of such data nets might be lower than building new permanent infrastructure.

At the other extreme, smaller quantities of data might be flown on unmanned aerial vehicles like carrier pigeons between labs and data centers.

UHS-1 does 104 MB/s, UHS-2 cards will do 312MB/s max according to spec. Add 630*2 seconds or 210*2 seconds to write the entire card and read it back at the end of the trip.

For shorter distance trips across town I’d also suggest dropping the density down to 0.8 or so and hardwiring all the cards into the $0.8 billion dollar van (assuming an imminently reasonable ~$40/card at ~19m cards).

http://en.wikipedia.org/wiki/IP_over_Avian_Carriers

Why would you go to the trouble using microSD cards when just a trunk full of cheap terabyte drives would carry much more information and be much less weight? Unless you are relying on RFC 1149.

Data density. Your average 3.5″ drive is 4″ x 5.75″ x 1″, and holds a current maximum of 4TB. This yields a data density of ~178GB per square inch. Three 64GB microSD cards, with a volume total of 0.03″ total, hold more data than that at 192GB.

You would hit the weight capacity of the Suburban before the volume capacity, at around 1.4 million microSD cards.

https://aws.amazon.com/snowmobile