But what was the connection to the PC?
We's gettin' Cable! YAY!!!!
- Thread starter StevieC
- Start date
- Status
Originally Posted By: brianl703
I'm getting 14mbps downloads with Comcast these days (roughly 1.4 megabytes/second). My old router was only 10mbps. It died one day. I'm glad it did, because I replaced it with a Linux router (with a 100mpbs interface) and I got greatly improved speed.
In order to get 13 megabytes/second, you need something faster than 100mpbs.
Living in the rural area I'm in... I'm just happy to have this speed and to not have to worry about bad storms degrading my bandwidth or wiping it out all together...
I'm getting 14mbps downloads with Comcast these days (roughly 1.4 megabytes/second). My old router was only 10mbps. It died one day. I'm glad it did, because I replaced it with a Linux router (with a 100mpbs interface) and I got greatly improved speed.
In order to get 13 megabytes/second, you need something faster than 100mpbs.
Living in the rural area I'm in... I'm just happy to have this speed and to not have to worry about bad storms degrading my bandwidth or wiping it out all together...
Originally Posted By: brianl703
I'm getting 14mbps downloads with Comcast these days (roughly 1.4 megabytes/second). My old router was only 10mbps. It died one day. I'm glad it did, because I replaced it with a Linux router (with a 100mpbs interface) and I got greatly improved speed.
In order to get 13 megabytes/second, you need something faster than 100mpbs.
Living in the rural area I'm in... I'm just happy to have this speed and to not have to worry about bad storms degrading my bandwidth or wiping it out all together...
I'm getting 14mbps downloads with Comcast these days (roughly 1.4 megabytes/second). My old router was only 10mbps. It died one day. I'm glad it did, because I replaced it with a Linux router (with a 100mpbs interface) and I got greatly improved speed.
In order to get 13 megabytes/second, you need something faster than 100mpbs.
Living in the rural area I'm in... I'm just happy to have this speed and to not have to worry about bad storms degrading my bandwidth or wiping it out all together...
OVERKILL
$100 Site Donor 2021
Originally Posted By: brianl703
But what was the connection to the PC?
Gigabit.
Though using a 3COM 905C 10/100 NIC, you could still get better than 100Mbit, which I thought was odd (that'd doing the math on a sustained transfer using the logged rate in KB/sec)
But what was the connection to the PC?
Gigabit.
Though using a 3COM 905C 10/100 NIC, you could still get better than 100Mbit, which I thought was odd (that'd doing the math on a sustained transfer using the logged rate in KB/sec)
Last edited by a moderator:
OVERK1LL is the router you recommended up to the task of my 8mps speeds? I think so!
Originally Posted By: OVERK1LL
Though using a 3COM 905C 10/100 NIC, you could still get better than 100Mbit, which I thought was odd (that'd doing the math on a sustained transfer using the logged rate in KB/sec)
Odd? It's impossible. Maybe whatever calculation the software was using to come up with KB/sec was a little off...
I figure out throughput by looking at total bytes transferred over a given time period.
With Linux:
ifconfig eth0 ; sleep 10 ; ifconfig eth0
Subtract the first set of numbers from the second, you get total bytes sent or received in 10 seconds...from there you can figure out throughput.
Though using a 3COM 905C 10/100 NIC, you could still get better than 100Mbit, which I thought was odd (that'd doing the math on a sustained transfer using the logged rate in KB/sec)
Odd? It's impossible. Maybe whatever calculation the software was using to come up with KB/sec was a little off...
I figure out throughput by looking at total bytes transferred over a given time period.
With Linux:
ifconfig eth0 ; sleep 10 ; ifconfig eth0
Subtract the first set of numbers from the second, you get total bytes sent or received in 10 seconds...from there you can figure out throughput.
OVERKILL
$100 Site Donor 2021
Originally Posted By: brianl703
Originally Posted By: OVERK1LL
Though using a 3COM 905C 10/100 NIC, you could still get better than 100Mbit, which I thought was odd (that'd doing the math on a sustained transfer using the logged rate in KB/sec)
Odd? It's impossible. Maybe whatever calculation the software was using to come up with KB/sec was a little off...
I figure out throughput by looking at total bytes transferred over a given time period.
With Linux:
ifconfig eth0 ; sleep 10 ; ifconfig eth0
Subtract the first set of numbers from the second, you get total bytes sent or received in 10 seconds...from there you can figure out throughput.
Switches were 3COM Superstack's with fibre uplinks. This wasn't junk equipment. And not the first time I've heard of 3COM equipment performing "better than advertised" either.
Originally Posted By: OVERK1LL
Though using a 3COM 905C 10/100 NIC, you could still get better than 100Mbit, which I thought was odd (that'd doing the math on a sustained transfer using the logged rate in KB/sec)
Odd? It's impossible. Maybe whatever calculation the software was using to come up with KB/sec was a little off...
I figure out throughput by looking at total bytes transferred over a given time period.
With Linux:
ifconfig eth0 ; sleep 10 ; ifconfig eth0
Subtract the first set of numbers from the second, you get total bytes sent or received in 10 seconds...from there you can figure out throughput.
Switches were 3COM Superstack's with fibre uplinks. This wasn't junk equipment. And not the first time I've heard of 3COM equipment performing "better than advertised" either.
The problem is that the IEEE 802.3 spec that this equipment is compliant with puts an upper bound on the throughput..
Here's what Cisco has to say on the subject:
Quote:
Ethernet clocking is asynchronous. IEEE 802.3 clock tolerance allows some links in a network to be as much as 200 ppm (parts or bits per million) slower than other links (0.02%). A traffic stream sourced at line rate on one link may traverse other links which are 0.02% slower. A fast source clock, or slow intermediate clocks, may limit the end-to-end throughput to only 99.98% of the source link rate.
So the maximum clock, and hence throughput, difference can only be 0.02%, hardly noticeable.
Unless your equipment is violating IEEE 802.3 specs, and then it IS "junk equipment" (which may well fail to interoperate with compliant IEEE 802.3 devices).
Here's what Cisco has to say on the subject:
Quote:
Ethernet clocking is asynchronous. IEEE 802.3 clock tolerance allows some links in a network to be as much as 200 ppm (parts or bits per million) slower than other links (0.02%). A traffic stream sourced at line rate on one link may traverse other links which are 0.02% slower. A fast source clock, or slow intermediate clocks, may limit the end-to-end throughput to only 99.98% of the source link rate.
So the maximum clock, and hence throughput, difference can only be 0.02%, hardly noticeable.
Unless your equipment is violating IEEE 802.3 specs, and then it IS "junk equipment" (which may well fail to interoperate with compliant IEEE 802.3 devices).
OVERKILL
$100 Site Donor 2021
Originally Posted By: brianl703
The problem is that the IEEE 802.3 spec that this equipment is compliant with puts an upper bound on the throughput..
Here's what Cisco has to say on the subject:
Quote:
Ethernet clocking is asynchronous. IEEE 802.3 clock tolerance allows some links in a network to be as much as 200 ppm (parts or bits per million) slower than other links (0.02%). A traffic stream sourced at line rate on one link may traverse other links which are 0.02% slower. A fast source clock, or slow intermediate clocks, may limit the end-to-end throughput to only 99.98% of the source link rate.
So the maximum clock, and hence throughput, difference can only be 0.02%, hardly noticeable.
Unless your equipment is violating IEEE 802.3 specs, and then it IS "junk equipment" (which may well fail to interoperate with compliant IEEE 802.3 devices).
I should probably clarify, I read back through and it looks like I fubar'd what I was trying to say. I didn't mean it exceeded the 100Mbit SPEC, rather OBSERVABLE (what I apparently omitted) throughput.
In my experience with 100Mbit Ethernet, most devices are not capable of "actual" 100Mbit throughput (including overhead) with the exception of higher-end equipment. 100Mbit being 12,800KB/sec as reference here for those confused by the bits/bytes conversion stuff.
Most consumer-level switches (what I call cheap junk) do "100Mbit" at ~8,000KB/sec. I've seen lower (sometimes a lot) as well, and some that have done slightly better. This is using decent quality NIC's and good cable.
The 3COM equipment (Superstack) in question would do ~12,000KB/sec (sustained) using a 3COM 905C NIC. The only thing close/equal has been my Cisco catalyst stuff with INTEL NIC's.
This is impressive (to me) if you factor in the overhead of TCP or UDP and how close to the "max spec" of the standard you are getting.
The problem is that the IEEE 802.3 spec that this equipment is compliant with puts an upper bound on the throughput..
Here's what Cisco has to say on the subject:
Quote:
Ethernet clocking is asynchronous. IEEE 802.3 clock tolerance allows some links in a network to be as much as 200 ppm (parts or bits per million) slower than other links (0.02%). A traffic stream sourced at line rate on one link may traverse other links which are 0.02% slower. A fast source clock, or slow intermediate clocks, may limit the end-to-end throughput to only 99.98% of the source link rate.
So the maximum clock, and hence throughput, difference can only be 0.02%, hardly noticeable.
Unless your equipment is violating IEEE 802.3 specs, and then it IS "junk equipment" (which may well fail to interoperate with compliant IEEE 802.3 devices).
I should probably clarify, I read back through and it looks like I fubar'd what I was trying to say. I didn't mean it exceeded the 100Mbit SPEC, rather OBSERVABLE (what I apparently omitted) throughput.
In my experience with 100Mbit Ethernet, most devices are not capable of "actual" 100Mbit throughput (including overhead) with the exception of higher-end equipment. 100Mbit being 12,800KB/sec as reference here for those confused by the bits/bytes conversion stuff.
Most consumer-level switches (what I call cheap junk) do "100Mbit" at ~8,000KB/sec. I've seen lower (sometimes a lot) as well, and some that have done slightly better. This is using decent quality NIC's and good cable.
The 3COM equipment (Superstack) in question would do ~12,000KB/sec (sustained) using a 3COM 905C NIC. The only thing close/equal has been my Cisco catalyst stuff with INTEL NIC's.
This is impressive (to me) if you factor in the overhead of TCP or UDP and how close to the "max spec" of the standard you are getting.
Last edited by a moderator:
I've been getting ~12,000KB/sec (sustained) from Kingston KNE100TX cards (no longer made) that use the Digital (DEC) Tulip chipset (no longer in production since Intel bought DEC's semiconductor business) for several years. In fact, back when the 3C905 was still in production, I ripped all of them out of the Linux machines at work and replaced them with the Kingston cards and the Linux machines stopped crashing. That was around 10 years ago. There was some sort of bug with the 3C905 card or Linux driver that's since been resolved, but it pretty much tainted my view of 3Com[1]. (It also helped that the Kingston cards were half the cost of the 3Com cards, and they worked so much better I got a raise for making the suggestion!)
About two years ago I started using Sun quad port PCI NICs (X1034 cards) with Linux and they get full throughput as well, as do Intel Etherexpress 100 cards. Those are about the only Ethernet cards I use, other than Gigabit Intel Etherexpress cards.
Incidentally, I recall reading that one thing some Ethernet devices do for higher throughput is reduce the interframe gap. This may be a violation of IEEE 802.3 specs, but probably doesn't matter when using a switch.
Customer grade switches are much better than they used to be. The technology is pretty much mature now, 100mb Ethernet having been around for some 15 years now.
[1]Robert Metcalfe's asinine statements about Linux didn't improve my view of that company either.
About two years ago I started using Sun quad port PCI NICs (X1034 cards) with Linux and they get full throughput as well, as do Intel Etherexpress 100 cards. Those are about the only Ethernet cards I use, other than Gigabit Intel Etherexpress cards.
Incidentally, I recall reading that one thing some Ethernet devices do for higher throughput is reduce the interframe gap. This may be a violation of IEEE 802.3 specs, but probably doesn't matter when using a switch.
Customer grade switches are much better than they used to be. The technology is pretty much mature now, 100mb Ethernet having been around for some 15 years now.
[1]Robert Metcalfe's asinine statements about Linux didn't improve my view of that company either.
OVERKILL
$100 Site Donor 2021
Oh man, I have a few of those Kingston NIC's too!!! They were very solid, I have a LAN for a client that had some older P3's with those cards in them. VERY reliable.
I've always had an affinity for 3COM, right back to the ISA 3C509 card. Ran one in my 386. The Intel stuff for me has also always been rock-solid.
And yes, consumer grade switches have improved vastly, but a newer Linksys with a couple Realtek NIC's still yields, at least in my experience, somewhere around 8,000KB/sec. That's quite a ways off from 12K+
Interesting point about the 3COM bug; when I started running Slackware in the early 90's, I don't recall ever running into it? Though I do remember them revising the module..... Likely an indication of an issue I simply never encountered.
My main OS of choice back then was FreeBSD though, as Linux felt cumbersome and lacking refinement in comparison. BSD was faster, and felt more refined.
Do you remember the AMD-based NIC's?
I've always had an affinity for 3COM, right back to the ISA 3C509 card. Ran one in my 386. The Intel stuff for me has also always been rock-solid.
And yes, consumer grade switches have improved vastly, but a newer Linksys with a couple Realtek NIC's still yields, at least in my experience, somewhere around 8,000KB/sec. That's quite a ways off from 12K+
Interesting point about the 3COM bug; when I started running Slackware in the early 90's, I don't recall ever running into it? Though I do remember them revising the module..... Likely an indication of an issue I simply never encountered.
My main OS of choice back then was FreeBSD though, as Linux felt cumbersome and lacking refinement in comparison. BSD was faster, and felt more refined.
Do you remember the AMD-based NIC's?
The problem with that might be the Realtek NICs...I don't get the impression they're very good. I avoid them. Since the choice these days (for new cards) seems to be between either an Intel card or a card with a Realtek chip, I'll go for the Intel card. A gigabit Intel card runs about $25 whereas a 10/100 Realtek card runs about $10. I'll happily pay the extra $15 even if I don't need gigabit speeds.
The 3COM bug may just have been something that happened under high server load. The server locked up with no video and no logfiles to tell you what happened. There were also problems with the Intel driver around that time, but it didn't result in the machine locking up, just loss of connectivity for a short period of time.
I actually have some AMD-based NICs, they're from an IBM server. I haven't used them for anything yet.
The 3COM bug may just have been something that happened under high server load. The server locked up with no video and no logfiles to tell you what happened. There were also problems with the Intel driver around that time, but it didn't result in the machine locking up, just loss of connectivity for a short period of time.
I actually have some AMD-based NICs, they're from an IBM server. I haven't used them for anything yet.
OVERKILL
$100 Site Donor 2021
Originally Posted By: brianl703
The problem with that might be the Realtek NICs...I don't get the impression they're very good. I avoid them. Since the choice these days (for new cards) seems to be between either an Intel card or a card with a Realtek chip, I'll go for the Intel card. A gigabit Intel card runs about $25 whereas a 10/100 Realtek card runs about $10. I'll happily pay the extra $15 even if I don't need gigabit speeds.
The 3COM bug may just have been something that happened under high server load. The server locked up with no video and no logfiles to tell you what happened. There were also problems with the Intel driver around that time, but it didn't result in the machine locking up, just loss of connectivity for a short period of time.
I actually have some AMD-based NICs, they're from an IBM server. I haven't used them for anything yet.
Oh, they are junk for sure. But they are "consumer grade" NIC's, and many from companies like D-LINK also have Realtek chips on them. It is uncommon to see anything other than Marvell, Atheros, Realtek or Broadcom in a Consumer-Grade computer system.
The other issue is that much of the cheaper consumer switching equipment also has Realtek chips on it
The problem with that might be the Realtek NICs...I don't get the impression they're very good. I avoid them. Since the choice these days (for new cards) seems to be between either an Intel card or a card with a Realtek chip, I'll go for the Intel card. A gigabit Intel card runs about $25 whereas a 10/100 Realtek card runs about $10. I'll happily pay the extra $15 even if I don't need gigabit speeds.
The 3COM bug may just have been something that happened under high server load. The server locked up with no video and no logfiles to tell you what happened. There were also problems with the Intel driver around that time, but it didn't result in the machine locking up, just loss of connectivity for a short period of time.
I actually have some AMD-based NICs, they're from an IBM server. I haven't used them for anything yet.
Oh, they are junk for sure. But they are "consumer grade" NIC's, and many from companies like D-LINK also have Realtek chips on them. It is uncommon to see anything other than Marvell, Atheros, Realtek or Broadcom in a Consumer-Grade computer system.
The other issue is that much of the cheaper consumer switching equipment also has Realtek chips on it
Broadcom's name is on the main chip on the 3C905-TX card.
Looking at what's for sale at Newegg, there is pretty much nothing other than Intel and a bunch of D-Link, NetGear, Trend-Net, etc. that all have Realtek chips on them.
I think Realtek designed their switch controllers much later than they did their RTL8139--the RTL8139 dates back at least 10 years and was a budget solution even back then.
Looking at what's for sale at Newegg, there is pretty much nothing other than Intel and a bunch of D-Link, NetGear, Trend-Net, etc. that all have Realtek chips on them.
I think Realtek designed their switch controllers much later than they did their RTL8139--the RTL8139 dates back at least 10 years and was a budget solution even back then.
OVERKILL
$100 Site Donor 2021
Originally Posted By: brianl703
Broadcom's name is on the main chip on the 3C905-TX card.
Looking at what's for sale at Newegg, there is pretty much nothing other than Intel and a bunch of D-Link, NetGear, Trend-Net, etc. that all have Realtek chips on them.
I think Realtek designed their switch controllers much later than they did their RTL8139--the RTL8139 dates back at least 10 years and was a budget solution even back then.
The Realtek 8139 is definitely antique, found on older cards like the D-Link 538TX for example.
I've never seen a Broadcom chip on a 3COM 905x series NIC.
Though there was a Marvell NIC that was branded "3COM" and used for on-board applications.
A few I had kicking around on the desk in my basement:
I HAD a 3C-905B-TX there somewhere, but cannot find it ATM. It is the one that has the MUCH larger main chip, with "3COM Parallel Tasking" on it, and three large arrows.
Broadcom's name is on the main chip on the 3C905-TX card.
Looking at what's for sale at Newegg, there is pretty much nothing other than Intel and a bunch of D-Link, NetGear, Trend-Net, etc. that all have Realtek chips on them.
I think Realtek designed their switch controllers much later than they did their RTL8139--the RTL8139 dates back at least 10 years and was a budget solution even back then.
The Realtek 8139 is definitely antique, found on older cards like the D-Link 538TX for example.
I've never seen a Broadcom chip on a 3COM 905x series NIC.
Though there was a Marvell NIC that was branded "3COM" and used for on-board applications.
A few I had kicking around on the desk in my basement:
I HAD a 3C-905B-TX there somewhere, but cannot find it ATM. It is the one that has the MUCH larger main chip, with "3COM Parallel Tasking" on it, and three large arrows.
Last edited by a moderator:
Can we see the rest of the car/computer parts on your workbench?
OVERKILL
$100 Site Donor 2021
Originally Posted By: StevieC
Can we see the rest of the car/computer parts on your workbench?
Not until I find my 905B [censored] it!
I've owned every variant (that I know of) of this NIC since its inception and have never seen a Broadcom chip on one.
Can we see the rest of the car/computer parts on your workbench?
Not until I find my 905B [censored] it!
I've owned every variant (that I know of) of this NIC since its inception and have never seen a Broadcom chip on one.
I love the 905's and was watching the thread and your posts... I have my own collection of them!
Best darn cards ever made IMO!
NOW SHOW ME THE BENCH!
Best darn cards ever made IMO!
NOW SHOW ME THE BENCH!
There's two versions of the 3C905C. One has a big chip, one has a small chip.
This is the one with the big chip that says BROADCOM on it:
http://farm1.static.flickr.com/28/37975076_7ebdd29718.jpg?v=0
This is the one with the big chip that says BROADCOM on it:
http://farm1.static.flickr.com/28/37975076_7ebdd29718.jpg?v=0
Originally Posted By: OVERK1LL
Not until I find my 905B [censored] it!
I've got one somewhere that won't work in 10mb mode. All I know is...I had set it to operate in 10mb mode so I could use it with cat3 cable to connect to a WAP..I unplugged it and left it unplugged for a week or two while I was on vacation...when I got back I plugged it in, it didn't work, and further testing showed that it would not work at 10mb, only 100mb. It won't even get a link when set at 10mb. Maybe something fried it while I was gone.
Not until I find my 905B [censored] it!
I've got one somewhere that won't work in 10mb mode. All I know is...I had set it to operate in 10mb mode so I could use it with cat3 cable to connect to a WAP..I unplugged it and left it unplugged for a week or two while I was on vacation...when I got back I plugged it in, it didn't work, and further testing showed that it would not work at 10mb, only 100mb. It won't even get a link when set at 10mb. Maybe something fried it while I was gone.
- Status
Similar threads
