The era of Fast Ethernet was a highly important and memorable (at least, to me) period in computing history. Introduced in 1995 (i.e. 20 years ago), after most competing networking technologies were starting to fall by the wayside, it added another nail into the coffin of its main competitor Token Ring who never made the jump past 16Mbit/s, and bought a much needed speed improvement from the 10Mbit/s “shared medium” hub/concentrator-based Ethernet which was common up till then. It seems that the Fast Ethernet standard was very well adopted, and by the turn of the century, most businesses were already running 100Mbit/s.
It became one of the most popular and inexpensive solutions for networking for well over a decade, and saw the “democratization” of cabled Ethernet networking as not something only for the corporate businesses, but something that every home should have. The standardization of the physical layer and protocols meant an even playing field for which many vendors competed against one another, resulting in a plethora of choice and rapid affordability. The ubiquity of the Fast Ethernet standard, offering 100Mbit/s over CAT-5 unshielded twisted pair cable, was so great that it drove Ethernet ports to be integrated into motherboards as standard, and resulted in expansion cards for Fast Ethernet being sold at under AU$10 as of 2003. By the end of the decade, Gigabit Ethernet cards would be sold for around AU$15.
While networking infrastructure was initially expensive in the 1990’s, home users with two-node requirements resorted to the cross-over cable which allowed operation between two cards with no need for an intervening hub or switch. The bandwidth on offer was quite impressive, given V.90 dial-up modems ratified in 1999 offered 56kbit/s, and serial direct cable connection rarely operated faster than 115.2kbit/s. Most hard drives at that time were still connected using IDE, in UDMA/33 mode, and would have an effective speed of roughly the same as the 100Mbit/s network (about 10-12 Mbyte/s).
Progress in integrated circuit technology pushed the cost of networking infrastructure down, and thus switched Ethernet became the norm around the early 2000’s, and hubs (or repeaters, concentrators, etc) became completely obsolete as there was no longer any cost reduction with the inferior solution. Full duplex operation became the norm. Some higher end products began to offer “automatic” cross-over operation, termed MDI/MDI-X, which allowed for interconnection of hubs without using cross-over cables or even the mis-application of cross-over cables when a straight-through was required. Since the introduction of Gigabit Ethernet, this has become a standard feature on Gigabit Ethernet ports where all four pairs are used, instead of just two in Fast Ethernet.
As broadband connections became more widespread, storage requirements increased and multimedia streaming became a reality, the network became a standard feature of most homes. The “CAT 5” or “blue” cable, so affectionately known, was a term coined in the Fast Ethernet era – when the cable requirement was Category 5 and many cables came with a blue outer sheathing. This cable was cheap to manufacture compared to coax, costing AU$0.50 per meter in many cases. By the mid 2000’s, it was rare to find a computer without an Ethernet port as standard. Operating system support for networking was “out of the box”, without messy configurations necessary for binding protocols to adapters as TCP/IP had become the de-facto standard, for example. This made networking accessible to many who would not have otherwise had it.
The mid-late 2000’s saw affordable integrated network gateway devices, offering DHCP, NAT/routing, and firewall features become very common, and became standard provision with some broadband ISPs. These devices simplified network creation even further, resulting in “literal” plug and play operation. Internet connection sharing configuration, by using a computer as a router/gateway was no longer a requirement.
The legacy lives on today – but the speed has ratcheted up another notch. Today, we are seeing Gigabit Ethernet at 1000Mbit/s become as affordable as Fast Ethernet was, and thus, usurp the market space once held by Fast Ethernet. While some of the lowest cost products today (e.g. routers, switches) still feature Fast Ethernet, the majority of computers have already moved onto Gigabit Ethernet, and the rest is likely to follow suit. But the contribution that Fast Ethernet has made in making affordable, speedy connectivity available to a large number of home and business users has enabled the low-cost IP connected technologies we rely on today such as NAS, IPTV, IP Cameras, VoIP, VPNs, DLNA Media Servers, network printing, etc, with many use cases still sufficiently serviced by Fast Ethernet today.
To commemorate the era of Fast Ethernet, which is on its way out, I’ve decided to make a two-part posting about the various Fast Ethernet cards in my possession. The vast majority of these are for the PCI bus, which itself, is on the way out too, although there are a few exceptions. Generally, Fast Ethernet cards for the ISA bus were never encountered, mainly because the ISA bus wasn’t really fast enough to handle the data rate especially when the hard disk controller and graphic adapter had been accounted for. While a lot of what I said is related to 100BASE-TX, Fast Ethernet featured other media as well, including other copper and fibre options, but these were generally much less popular.
3Com Fast Etherlink XL PCI
These adapters were somewhat iconic in the Fast Ethernet era as they were commonly installed by OEMs such as HP as well as by smaller computer shops. The adapters are of 1998 to 1999 vintage, and at that time, 3Com was a reputable brand in networking equipment, backed by Lucent Technologies.
Several different versions of the card are seen in my collection, with very subtle differences. This 3C905B-TX has a code of 03-172-400 and in Made in Singapore with a white label. This part appears to be an HP OEM part. The main chip is all integrated, meaning the MAC and PHY are in the same chip (as most of the Fast Ethernet products were). A “bel” branded transformer is used for the magnetics (as Ethernet is galvanically isolated by transformers). The PCB seems to be dated Week 30, 1999, with a chip date of Week 27, 1999.
A second variant is marked with a model code of 3C905B-TXNM and utilizes the yellow silkscreening to provide the data about the card, instead of a label. The product code is 03-172-110 and this example is Made in USA. Premium stuff! The chip has a slightly smaller mark on it, and is dated Week 48 of 1999, with a PCB made by KCE dated Week 46 of 1999. This would have been installed by a local computer shop by the looks of it.
The final example is a 3C905B-TX with silkscreened codes of a different order. The product code is 03-0172-400 and is an older example with Pulse branded magnetics (equivalent part) and a chip date of Week 10, 1998 and PCB date of Week 9, 1998. The PCB is made by KCE, and is marked Fab 02-0172-000 Rev 01, as opposed to Fab 02-0172-004 Rev A of the ones above. The PCB itself is a glossy solder resist finish unlike the matte boards above.
All of the cards featured similar backplates, with the oldest one eschewing the 3Com, CE and no-telephone-line marks. In what appears to be a cost cutting measure, the LEDs are surface mounted to the card and shine through holes made in the backplate. No light-pipe is used, however, three LEDs providing activity and link rate indication are provided.
A common feature is the Parallel Tasking II Performance logo on these chips, which indicates the provision of more advanced multiple-data transfers over PCI to reduce CPU usage by the network card. In some sense, this can be thought of as a way of “bus mastering” burst transfers of data to reduce interrupt rates. This idea has been taken further by modern hardware with provision of hardware accelerators for calculating frame checksums, packetization etc.
Reception seemed to be good, although it appears to have been a pricey option, as most 3com products were.
Realtek RTL8139x-based Adapters
From quality, we move straight to the other end of the spectrum – that of the highly cost-reduced adapters. These were the adapters you would often get if you were a cost conscious home-PC builder, and you wanted to spend the absolute minimum money on a network card. While they were generally not unreliable, they often showed sub-optimal performance due to their simple hardware design which relied heavily on drivers and host CPU to perform tasks. In fact, it seems it has no abilities whatsoever – e.g. no scatter/gather, hardware checksum, unaligned accesses, IRQ moderation, hardware packet fragmentation or encryption. These were, however, very adequate in the late 2000’s as computing power was in surplus, and they had an added benefit of being widely supported by many later OSes due to their popularity. Several generations of RTL8139-based adapters have been seen in my collection.
The first card is pretty non-descript, with the exception that it’s clearly made in Taiwan in Week 37 of 1999 and features an RTL8139A chipset (5V only) and YCL branded magnetics. The FCC ID leads us to the conclusion that it was built by CIS Technology Inc. as a likely “generic” adapter.
No branding is found on the backplane, and two 3mm LEDs are used to indicate link and speed. This would be a considered a somewhat “older” generation of generic adapter, as it has a sizeable PCB and not entirely spartan set of inclusions, although the chipset itself is probably not famed for their speed.
In fact, this latest adapter shows you just how spartan a generic adapter can be. Towards the latter end of the Fast Ethernet era, this adapter is dated Week 5 of 2002, and was made in China. It features a minimally sized PCB, using a later RTL8139C chipset with dual voltage support, hence the dual notches, and missing the option ROM socket despite having the footprint for it. It also features an Ethernet jack missing wipers for the other 4 unused contacts, just for some additional cost savings. No WOL header is populated either.
Again, two 3mm LEDs are used for link and activity display, with nothing fancy on the backplate. The card itself is marked SN-5300TX/N, although that doesn’t really bring anything up. Searching the ACN seems to suggest this was a D-Link branded adapter at one stage. The PCB also has “OASIS” written on it, which suggests this may be the PCB manufacturer.
A little bit later still along the cost reduction path and we can see a newer chipset, an RTL8139D with a missing option ROM socket (again). This is dated to Week 7 of 2004, and the model number suggests it’s another generic adapter, with a brand name of Xnet. The three-pin WOL header is not populated either, and this one has the same pin-reduced plug. Further cost optimization is had by cutting the PCB shorter than the full PCI length.
The final cost optimization is had in eliminating the second LED, resulting in just one LED to signal link and activity.
A question readers may have is whether these reductions really had any great impact on the user experience, and the answer is – most likely not. The socket which allows for a DIP chip to be inserted is rarely used by consumers, as it was intended for an Option ROM containing net-booting software (e.g. PXE, RPL, NetWare, etc) for systems without any internal boot devices. This is unlikely to be used in home devices, but furthermore, when NICs became integrated on motherboards, these Option ROMs were generally standardized to PXE or RPL only, and were pre-loaded onto the BIOS firmware so as not to be needed to be supplied separately. Users who did not have Option ROMs could still net-boot provided they booted using a physical CD-ROM or floppy disk with the required software to continue the booting from the network.
The other socket for Wake on Lan was rarely used by older home machines. This was partially because of the prevalence of AT supplies in the late 90’s which did not have a “soft off” power setting which allowed for “waking” from, required appropriate motherboard support and because it had a reputation for being flaky. It was most useful for servers, and in the later days, the WoL feature could be provided through the PCI bus rather than by a dedicated cable.
And finally, the elimination of LEDs proved to be inconsequential for most users. Very few users looked at the LEDs after they lit up the first time indicating link, as the state was able to be determined through software diagnostics (e.g. tray icons, driver utilities). Many of the other LEDs provided by other cards were not normally lit during operation anyway, or were so rarely lit to be unlikely to be useful.
As a result, these cost reduction strategies generally had no functional impacts on most users, and were happily accepted for lower prices in return.
This is a middle-to-late D-Link branded adapter dated Week 4 of 2002. This adapter seems to be a very generic adapter, but built a little “better”, without the removal of Option-ROM and WOL headers and the use of a shielded jack, but still retaining the use of a cheaper branded transformer. The chipset seems to be rather interesting, marked as DL10038C, for which I suspect DL stands for D-Link. This may be an attempt to “disguise” a generic chipset as something “better”. There is some evidence that it’s just an RTL8139 in disguise or possibly a VIA Rhine.
The extra touch of class is provided by the branding on the backplate – that’s really all you get for paying about twice as much as the generic branded adapters.
I have no clue who MPX is, but these chipsets were used in some other cards under different branding. This one uses an MPX EN5038, with Deltra branded magnetics, with a full complement of jacks and is dated Week 3 of 2001. From the labels, it was supplied in an IBM system, and the ACN points to this being an Accton product.
It generously has three 3mm LEDs on the rear, two for speed, one for activity.
SMC EZCard 10/100 SMC1211TX
In a case of deja vu, doesn’t the card above look like the Accton card above? It seems to be based on the same PCB, albiet an older revision chipset labelled EN5030B1. SMC was a decently reputable networking brand, being Standard Microsystems Corporation, but were then accquired by Accton.
It has no firm date I can see, but I’d assume it’s a little older than the card above. It does have the same three LED arrangement, although it also has a slight improvement in that there is a model name and branding stamped into the bottom of the backplate.
This is the only low-profile card in my collection, and is a later one with a much smaller PCB but not skimping on the features. Similarly, an EN-series chipset is used, marked EN5251BE. The PCB is dated Week 1 of 2003.
The LED is surface mounted on the board, and a single LED is provided for indication of link and activity. No metal shielding is provided on the jack, making this card incompatible with shielded twisted pair, which is rarely used anyway.
This is another generic card from another “generic” vendor, although C-Net (unrelated to the online news and download site) was trying to get established as a networking brand. This one had no Wake-on-LAN connector, but did have an option ROM socket. The socket footprint is slightly longer, as a larger option ROM can be fitted to some cards. This is powered by a different chipset – namely a Davicom DM9102AF which appears to have fairly similar characteristics to the Realtek competitors. The card is quite dusty, but it seems to be dated Week 53 of 2000.
Two LEDs are used, 3mm, and the backplate is stamped with the branding for an additional touch of class.
Follow along with the next part to see some more Fast Ethernet hardware – including some more “interesting” examples. Coming soon!