Truth in advertising -- it's important even in networks.
So when you bought that Gigabit Ethernet NIC for your server, how much throughput did you think it was capable of? 1,000 Mpbs? Or at least 900 Mbps? Why would you think that; just because Gigabit means 1,000,000,000 bps and it was written on the network interface card (NIC) packaging?
Buyer beware (or "Caveat Emptor" for the Latin geeks)!
Like everything in consumer life, you shouldn't make the assumption that the packaging corresponds to the contents. NIC cards, or more specifically the drivers provided for them, won't necessarily perform at Layer 2 specification. In fact, the phrase "your mileage may vary" should be kept firmly in mind -- as well as the fact that there are many ways to improve your mileage and most of them are simple.
Of course this applies to any NIC card -- Gigabit Ethernet isn't the culprit. It's just the latest, and therefore least mature, desktop network technology. Let's take a quick look at what should be possible from a Layer 3 data payload perspective based on the Layer 2 frame rate.
The Layer 2 link speed as it appears on the packaging is partly consumed by Layer 3 headers, suggesting a maximum theoretical value. Waving hands a little to account for maturity of technology (such as the case of Gigabit), we can arrive at some realistic estimates of peak performance a NIC might be able to offer.
A sampling of a popular 10 and 100 Mbps NICs shows that, despite being quite mature, performance can be highly variable. Focusing on Windows 2000 as a typical end-station OS, the peak transfer rate is a reflection of Layer 3 (i.e. pre-TCP) performance as measured by AppareNet. Using a methodology distinct from raw data throughput, it exposes the NIC's sensivities and limitations as they would affect a typical application. (NOTE: the names have been changed to protect the author):
|
NIC
|
NIC
driver version
|
OS
|
Peak Transfer Rate
(Mbps)
|
|
Vendor A
10/100 model AX
|
MS 5.5.0.0
|
Win2K
|
96
|
|
Vendor B
100-only model BX
|
VB
5.41.27.0
|
Win2K
|
96
|
|
Vendor
B 10/100 model BY
|
VB 6.1.3.0
|
Win2K
|
92
|
|
Vendor A
10/100 model AX
|
MS
5.0.2170.1
|
Win2K
|
92
|
|
Vendor
B
10/100
model BZ
|
VB
5.41.27.0
|
Win2K
|
90
|
|
Vendor
B
10/100
model BZ
|
MS
4.1.67.0
|
Win2K
|
48
|
|
Vendor A
10/100 model AX
|
VA
1.10.14.0
|
Win2K
|
89
|
|
Vendor A
10/100 model AY
|
VA 4.8.0.0
|
Win2K
|
88
|
|
Vendor C
10/100 PCMCIA 32 bit model CX
|
MS 5.5.0.0
|
Win2K
|
75
|
|
Vendor D
10/100 PCMCIA 32 bit model DX
|
MS
2.58.2.2
|
Win2K
|
59
|
|
Vendor D
10/100 – model DY
|
VD 3.12
|
Win2K
|
55
|
|
Vendor A
10/100 PCMCIA 16 bit model AZ
|
MS
2.0.3.4000
|
Win2K
|
6
|
Gigabit can be even more variable.
|
NIC
|
NIC
driver version
|
OS
|
Peak Transfer Rate
(Mbps)
|
|
Vendor
E 1000 model EX
|
n/a
|
Solaris 9
|
820
|
|
Vendor
E 1000 model EX
|
VE
7.43.0.0
|
Win2K
|
650
|
|
Vendor
E 1000 model EX
|
VE
6.63.0.0
|
Win2K
|
650
|
|
Vendor
B
1000
model BX
|
VB
7.4.19.0
|
Win2K
|
470
|
|
Vendor
B
1000
model BX
|
VB
6.2.22.1
|
Win2K
|
440
|
|
Vendor F
1000 model FX
|
VF
1.2.905.2001
|
Win2K
|
460
|
|
Vendor G
1000 model GX
|
VG 6.2.2.0
|
Win2K
|
400
|
|
Vendor A
1000 model AW
|
VA
1.0.0.64
|
Win2K
|
460
|
Often the solution is as simple as installing the latest drivers. But occasionally the latest drivers are sometimes the culprits, as manufacturers push out not-ready-for-prime-time versions, to solve other problems. Or the drivers automatically installed by the operating system are not optimal.
Before you start wasting time fine-tuning your network, check your driver versions. And choose tried-and-true NICs that have proven performance. And whatever you do, don't be taken in by what's written on the outside of the package.
Chief Scientist for Apparent Networks, Loki Jorgenson, PhD, has been active in computation, physics and mathematics, scientific visualization, and simulation for over 18 years. Trained in computational physics at Queen's and McGill universities, he has published in areas as diverse as philosophy, graphics, educational technologies, statistical mechanics, logic and number theory. Also, he acts as Adjunct Professor of Mathematics at Simon Fraser University where he co-founded the Center for Experimental and Constructive Mathematics (CECM). He has headed research in numerous academic projects from high-performance computing to digital publishing, working closely with private sector partners and government. At Apparent Networks Inc., Jorgenson leads network research in high performance, wireless, VoIP and other application performance, typically through practical collaboration with academic organizations and other thought leaders such as BCnet, Texas A&M, CANARIE, and Internet2. www.apparentnetworks.com
