Exploring Time Capsule: theoretical speed vs practical throughput
Time Capsule, announced earlier this year, is a base station with an integrated hard drive and power supply. The previous segment of this series exploring Time Capsule in depth looked at the differences in members of the AirPort family. This segment, the second of six, compares the differences between the hypothetical maximum data transmission speed and typical real world performance of Time Capsule's SATA, USB 2.0, Ethernet networking, and WiFi Wireless networking interfaces.
The theoretical speed rating of a given wired or wireless connection can be betrayed by a number of factors, from the overhead of the protocols involved to signal interference. Network data throughput is usually measured in megabits per second, which are an eighth of a megabyte per second. Disk speeds are typically cited in megabytes per second; here, I'll list both numbers to make it easier to compare disk and network throughput speeds.
SATA, or Serial ATA, has a theoretical maximum of 1200 Mbits/sec (150 MB/sec). However, existing hard drives can't even deliver data that fast; top disk output speeds are closer to 40 to 100 MB/sec, depending on whether the data is being read from the inside or outside of the disk platter, the disk spin speed, and other factors.
USB 2.0 has a theoretical maximum of 480 Mbits/sec (60 MB/sec). A USB hard drive is typically a standard ATA or SATA drive attached to a USB bridge chipset. The actual speed of the USB interface depends upon the performance of the chipset used as well as the performance of the computer the drive is attached to. That's because USB transfers most of the heavy lifting to the host computer's CPU.
USB has a faster theoretical maximum than Firewire 400 (400 Mbits/sec; 50 MB/sec), but Firewire 400 is actually much faster than USB because it uses smarter peer to peer interface hardware rather than pushing low level work onto the PC host's CPU as the simpler master to slave architecture of USB does.
On a Mac, Firewire is typically around twice as fast in real world transfer rates, with USB hitting around 18 MB/sec and Firewire reaching 35 MB/sec throughput. Windows' implementation of USB has historically been faster than Mac OS X's, with Windows' USB reaching throughput closer to 33MB/sec. That also explains why Firewire is more popular on the Mac than on the PC side; it's simply far more dramatically faster than USB on the Mac, while Firewire offers less of a noticeable boost in Windows. Macs also have Firewire Target Mode, which PC users lack. For more details on why USB is faster in Windows compared to the Mac, see the footnote: USB Performance in Windows vs Mac OS X at the end of this article.
Time Capsule doesn't use Firewire; it's USB only. There are two reasons for this. First, USB chipsets are cheaper than Firewire, because they do less (USB peripherals have less intelligence on board and transfer more work to the CPU). Second, Time Capsule and the AirPort Extreme are both designed as wireless network appliances, so the difference in performance between attached Firewire and USB drives typically wouldn't be noticeable. Test results presented in the next segment bear that out.
In reality, USB doesn't simply run at a given speed. The performance of a directly connected USB drive can be affected by a number of issues, from the performance of the host computer to interference caused by other USB devices on the same bus, to the overhead related to the drive's file system.
Ethernet Networking introduces even more complicating factors. There is the overhead of Internet Protocol addressing, as well as the file sharing protocols used, such as AFP on the Mac or SMB used by Windows, neither of which play into direct, non-networked protocols such as USB. There are also architectural issues such as the quality of the cables used and the performance of any switches (or old fashioned hubs) involved. All of these issues eat into the theoretical raw data transfer rate of Ethernet.
Fast Ethernet has a theoretical speed of 100 Mbits/sec (12 MB/sec), while Gigabit Ethernet has a theoretical speed of 1000 Mbits/sec (120 MB/sec). That suggests a double speed advantage of Gigabit Ethernet over USB (60 MB/sec), but neither protocol hits its maximum. In reality, a typical USB connected disk is roughly equal to or lesser than the throughput of a shared drive attached over a Gigabit Ethernet network.
Wireless Networking has all the complexity of traditional wired networking with the additional complications of signal strength issues such as radio interference and barriers, as well as additional overhead related to wireless transmission that commonly halves its real world throughput over the theoretical raw data rate.
- 802.11b has a theoretical speed maximum of 11 Mbits/sec with a typical transfer rate of around 4.5 Mbits/sec (0.5 MB/sec) with an ideal signal.
- 802.11g has a theoretical speed maximum of 54 Mbits/sec, with a typical transfer rate of around 23 Mbits/sec (2.5 MB/sec) with an ideal signal.
- 802.11n has a theoretical speed maximum of 300 Mbits/sec, with a typical transfer rate of around 74 Mbits/sec (9.25 MB/sec) with an ideal signal.
As the signal strength of a wireless network drops, the connection speed is automatically renegotiated and slower and slower rates until no connection is possible. The transfer rates of wireless networking make it ideal for browsing the web, as most US residents have a connection speed of around 1.5 Mbits/sec for DSL, or from 3 to 6 Mbit/sec with cable Internet service. Any version of WiFi is much faster than that.
However, very fastest wireless networking is required to perform intensive data transfers such as Time Machine backups, general file sharing, and media streaming, particularly if more than one client is using the network at once, or if one user is trying to do more than one thing with their wireless connection, such as backing up files while streaming audio to Apple TV, for example.
A Visual Speed Comparison
This chart shows the relative difference in throughput of the interfaces described above, with theoretical raw data rates in blue, and typical real world throughput in red. Note that these real world numbers are ideal peak maximums, not the average throughput users will see at all times. As detailed above, there are lots of factors that can eat into the actual real world performance. Time Capsule has performance limitations of its own, which are related to its design to primarily serve wireless clients. An upcoming segment will detail what Time Capsule itself can do.
Direct connection interfaces, such as SATA and USB, commonly deliver closer to half their theoretical maximum raw data rate, but as interfaces and drive mechanisms improve, the real world data throughput will rise. Ethernet networking interfaces, such as Fast Ethernet or Gigabit Ethernet, can hit peak transmission rates close to their maximums, but suffer from greater overhead compared to a direct connection interface.
Wireless networking throughput depends more on external factors to reach its full potential. Ideal signal strength is critically important to reach anywhere near the high end of real world throughput numbers. There are other factors that make a huge difference in wireless performance; Time Capsule and AirPort Extreme both support new features unique to the new 802.11n wireless networking protocol, including the use of multiple antennas (a technology referred to as MIMO) and the use of the 5 GHz radio spectrum. The next segment will look at the pros and cons of using this alternative frequency, which depending on the circumstances can either decrease signal range or deliver a major boost in your wireless data rate.
Footnote: USB Performance in Windows vs Mac OS X
In addition to the cabling and protocol specifics, there are other reasons for Windows PCs to outperform Macs in USB transfers. The testing done by BareFeats in the article USB 2.0 versus FireWire compared 2004 PowerPC Macs against 3 GHz Pentium 4 PCs; since USB pushes much of its work to the CPU, the speed of the host made a big difference in how fast USB performed on the two platforms.
Their testing also revealed that the first generation of the PowerMac G5 delivered poor I/O across the board, scoring lower than even the mobile PowerBook and low cost eMac in both Firewire and USB. That indicates that the theoretical expectations for USB (or any protocol) are nearly meaningless when compared to the actual speed of the disk, processor, the implementation of the protocol itself, and other factors that might cause interference or otherwise eat up the expected maximum throughput speeds. In other words, USB does not ever run at its maximum theoretical speed rating.
Additionally, Windows file sharing and disk protocols are simpler than on the Mac, because Windows handles and presents less metadata. This lightness makes for faster disk operations at the expense of the sophistication of the Mac's higher quality file icons, richer file type and creator codes, and other features missing in Windows.
There are other factors that affect cross platform throughput as well; Mac OS X suffers some degree of overhead from new features such as Spotlight indexing, while Windows PCs are typically burdened with running anti-virus scanning software that peels away a significant edge in performance. Clearly, there are lots of factors to account for in making direct performance comparisons, and neatly presented numbers can easily hide those details in a misleading way.
Previous articles related to Time Capsule and its AirPort Extreme cousin: