Intel 82574L looses to Realtek RTL8111E Ethernet
-
i tend to disagree with posts #2 and part of #3.
i have Tested the 8111E in the lab extensively on different mobo mfg's motherboards and find they Bottleneck
LONG before the Intel Nics do. you have to understand (especially in where im employed) that with Gig-E links
at 80% utilization 24/7 , the 8111E just cant hack it 24/7 before we see crashes from the NIC… i could barely
push 60% utilization down a 8111E throat and i could easily make it crash... understand the tests in the lab are
(like Realworld in Datacenters and warehouses and our offices) are done FULL duplex.Yes Realtek has a bad history from the 10/100 days but its NOT much better now. im tired of having to deal
with this JUNK and having the management want us to to test the newest mobo from our distributors and they show
up with these junk Nics onboard the mobo. (and these systems are cheaper cost wise but we cant afford ANY down
time in the enterprise i work in)ive had better luck with Broadcom than i have had with Realtek. at least we dont see crashes from the Nics from
broadcom hardware even though i dont have even one machine deployed in our enterprise that has Realtek on it.so from my perspective (in a Million Dollar plus Enterprise) , we cant sacrifice downtime...Downtime costs us severely
so we stick with the Same system's that are ROCK solid and they all Feature Intel on the mobo...also of Note. I am a apple fanboy as you want to call it, i have LOTS of mac and a iPad or 2... BUT I dislike the iphone
.. i have 2 macs on my desk at work (4 monitor'd Mac Pro and a MacBookPro (Non Retina)
and at home im worse... more macs.. NOT a pc to be seen... the iphone/ipad has problems with Wifi for SURE and i
can see it at work. (employees have have the right to use there wifi devices on our network (isolated of course) and we field
probably 20 calls a day about my apple iphone or ipad doesnt connect to the network... apple has a mess on there hands
with some of the ipads and ALL of the iphones... Wished work didnt have the allowing employees to bring wifi devices and use
them on the network at work but its not my call... only certain iPads seem to have wifi issues.in the lab ive tested IOS 6.1.2 LONG before anybody else had access to it, and its Better on our network but theres still issues..
and im sure 6.1.3 will break the fixes in 6.1.2 -
Ha, got you to click.
Bah! ::)
@SunCatalyst:
Yes Realtek has a bad history from the 10/100 days but its NOT much better now.
I have to disagree with this, the Gigabit Realtek NICs are much better than, say, the RTL8139. However that's not saying much because:
The RealTek 8139 PCI NIC redefines the meaning of 'low end.' This is probably the worst PCI ethernet controller ever made…
If you have 8139 NICs you cannot assume they will stay 'up' under pfSense. They may just stop working. The gigabit NICs may bottleneck at less than 1Gbps but they are at least reliable, in my experience.
The Intel 82574L is an old NIC, yes, but because of that it's probably more widely used and tested than almost anything else. It has now been replaced by the Intel I350/340 (I'm not really up to speed on those yet). The number one reason why Intel NICs are recommended though is that Intel contribute the drivers to FreeBSD. I gurantee there will be (or already are) drivers for the I350 NICs for FreeBSD. I cannot say the same for new Realtek NICs.
Steve
-
Yup.. it all comes down to how good or bad the drivers are written for any NIC.
-
As a matter of fact, I'm actually experiencing a living hell with Realtek RTL 8111EVL (onboard NICs on NF9D-2700 mobo) and pfsense 2.1 beta…
Keep getting "watchdog timeout" underload and can't get NICs back to working state unless rebooting.Had no problem with previous "older" realtek gigabit chips.
And Intel is no 100% magic that will works in all cases. Recent epic fail demonstrate this. -
And Intel is no 100% magic that will works in all cases. Recent epic fail demonstrate this.
Especially if the motherboard manufacturer doesn't follow the design / implementation specs. :)
-
Yep, bad third party code can kill anything. ;)
Steve
-
bad Code does hurt products and in that case , Intel didnt do the right thing and release
the patch for everybody to apply to that particular Motherboard … was limited to that one
motherboard from the certain MFG for NOW...that case was the MFG fault for not programming
the EEPROM with correct contents.now with Realtek... its a Hardware problem. i have well over 150+ servers
siting on pallets in the back room of the lab with different models of RealTek nic's in them..
they all fail in one sort or another or fall flat on there faces.also of interesting note... i also have another 2 pallets full of servers as well that failed
the lab tests for different reasons and SOME of them have intel nics and Broadcom nics...all 3 of these pallets of equipment is destined to be destroyed by our corp policy..
everything is tested extensively for 90 days in the lab (under our realworld data tests) and
decided whether to deploy it or not. -
That's some fairly extensive testing, I guess uptime is really important to you, ;)
You've clearly tested more hardware than I have probably ever seen!
Steve
-
we have a 99.9% uptime requirement… with the exception of the fiasco of the boss
that moved us to Cisco without testing in the lab first, we have had 100% uptime
but we have built redundancy into every possible place.my budget just for testing new machines each year is around 1Million Dollars...
Budget on what ive already spent this year on new hardware to deploy...
1 Million in new hardware...but you have to consider we have:
8 warehouses around the world we ship from (7 of the warehouses are in the US)
equipment in 20 DataCenters around the world
20 office buildings full of employees taking orders 24/7its a shame that all the equipment that fails our tests (that is BRAND new) gets
destroyed but its corp policy. -
@SunCatalyst:
its a shame that all the equipment that fails our tests (that is BRAND new) gets
destroyed but its corp policyUsed to work under similar rules, discarded equipment is destroyed, at least 99% is fully working equipment that should be donated.
-
@SunCatalyst:
my budget just for testing new machines each year is around 1Million Dollars…
:o OK
I'll upgrade my earlier statement to 'definitely more hardware than I've ever seen'.Steve
-
I've used 3 RealTek 8139 10/100BaseTX NICs (2 WAN in failover + 1 LAN) for the past 3-4 years. It's a very old non-server platform that mostly does NAT for 30-40 clients and barely pushes the 100/100 mbit WAN to its limit, and not for longer than a couple of hours a day. The office-hours utilization would be around 30/30 mbit on average.
It's nothing like the above scenarios, but still I have to say that I have NEVER experienced any problems with the NICs. Before I had to upgrade to 2.0.2 I logged over 1 year uptime.
-
i have to agree that the equipment should be donated somewhere or resold… not destroyed...
MOST of our Gigabit connections are at about 80% utilization (which is really damn close to
what the physical hardware can transmit on the wire after the tcp stack).we HAVE to have systems that dont fail under the load or crash... currently looking for 10GE Nics at the
moment to get some relief as our traffic gets to be more and more each year...
10Ge seems to be ALOT thinner pickings that are supported under FreeBSD 8.3 (pfsense 2.1 Beta)and i DOUBT we will be running intel Nics at 10GE..
-
@SunCatalyst:
MOST of our Gigabit connections are at about 80% utilization (which is really damn close to
what the physical hardware can transmit on the wire after the tcp stack).we HAVE to have systems that dont fail under the load or crash… currently looking for 10GE Nics at the
moment to get some relief as our traffic gets to be more and more each year...
10Ge seems to be ALOT thinner pickings that are supported under FreeBSD 8.3 (pfsense 2.1 Beta)and i DOUBT we will be running intel Nics at 10GE..
Have you considered configuring 2 x GbE ports in LACP ? http://en.wikipedia.org/wiki/Link_aggregation
-
we have thought about it BUT…... most of our uplink providers DONT support it..
OR its cheaper to do 10GE drop to us and having to have redundancy built into
everything , that complicates stuff tremendously -
Well FreeBSD-HEAD (what will eventually become FreeBSD 10) has recently updated drivers for several 10 GbE NICs, such as Chelsio, Emulex and Intel. I'm not sure how much of that code has been backported to FreeBSD 8.x though …
-
@SunCatalyst - I volunteer with an organisation in Nepal - www.inf.org - we have nothing like "western country" bandwidths available, just getting WiMax technology in the last few months with Nepal Telecom guaranteeing >1Mbps to corporate customers! We get our first WiMax connection on Monday. We are not going to saturate any ethernet chip any time soon, so equipment that has some obscure high-performance issue is no big deal. If your corporate has a change of heart, please PM me. We have fund-raising offices in various countries and can easily accept donations of electronics equipment and get it out to Nepal.
-
@phil.davis. if i had a choice i would do something like your suggesting BUT corp policy dictates otherwise.
i will be sending you a PM about other possibilities i MAY have for you. -
Put me in the list, I work for an NGO that is always struggling with tight budgets for IT infrastructure.
-
