In ros V5 we have new intel drivers (igb xxx), and i whant to know what parameters used in this drivers ???
If i use intel card with chipset 82576E (8 queue on rx and 8 queue on tx) how many queue per port use new driver ? and does its use MSI-X interrupts ?
Can we see this information in Ros V5 beta ? and can we change this parameter in driver ?

Normis, can you answer for this question ?

I don't understand the question

Ok ))
I have Intel 82576 Gigabit Network Controller which supports 8-receive queues per port. (2-port 1Gbps PCI-express package (E1G142ET))
This ethernet card works with MSI-X interrupts (MSI-X enables the Ethernet controller to direct interrupt messages to multiple processor cores), but in ROS V5 we see :

irq=17 owner="[0 0 IO-APIC-fasteoi eth1]"

irq=19 owner="[0 0 IO-APIC-fasteoi eth2]"

This card use old fasteoi interrupt.

Question :
1) How can i enable MSI-X interrupt ?
2) How i can enable 8-receive queues per port on i82576 ethernet card ?
3) How i can see with what parameter load 'igb' driver ?

This screen from linux router that use same i82576 ethernet card ('igb' driver load whith parameter - "modprobe igb IntMode=3,3")

# dmesg
Intel(R) Gigabit Ethernet Network Driver - version 1.3.8.6
Copyright (c) 2007-2008 Intel Corporation.
PCI: Enabling device 0000:01:00.0 (0000 -> 0003)
ACPI: PCI Interrupt 0000:01:00.0[A] -> GSI 16 (level, low) -> IRQ 169
PCI: Setting latency timer of device 0000:01:00.0 to 64
igb: 0000:01:00.0: igb_validate_option: Interrupt Mode set to 3
igb: eth0: igb_probe: Intel(R) Gigabit Ethernet Network Connection
igb: eth0: igb_probe: (PCIe:2.5Gb/s:Width x4) 00:1b:21:2e:9c:a4
igb: eth0: igb_probe: Using MSI-X interrupts. 4 rx queue(s), 1 tx queue(s)
PCI: Enabling device 0000:01:00.1 (0000 -> 0003)
ACPI: PCI Interrupt 0000:01:00.1 -> GSI 17 (level, low) -> IRQ 193
PCI: Setting latency timer of device 0000:01:00.1 to 64
igb: 0000:01:00.1: igb_validate_option: Interrupt Mode set to 3
igb: eth1: igb_probe: Intel(R) Gigabit Ethernet Network Connection
igb: eth1: igb_probe: (PCIe:2.5Gb/s:Width x4) 00:1b:21:2e:9c:a5
igb: eth1: igb_probe: Using MSI-X interrupts. 4 rx queue(s), 1 tx queue(s)

# cat /proc/interrupts | grep eth
140: 23 63 0 0 PCI-MSI-X eth0-Q0
148: 23 0 0 63 PCI-MSI-X eth0-Q1
156: 23 43 20 0 PCI-MSI-X eth0-Q2
164: 23 0 0 63 PCI-MSI-X eth0-Q3
172: 2 0 0 0 PCI-MSI-X eth0
180: 21 67 0 0 PCI-MSI-X eth1-Q0
188: 21 0 67 0 PCI-MSI-X eth1-Q1
196: 21 20 0 47 PCI-MSI-X eth1-Q2
204: 21 0 67 0 PCI-MSI-X eth1-Q3
212: 1 0 0 0 PCI-MSI-X eth1

Important question.
There is no reason to use new Intel NICs if mikrotik uses only 1 CPU per NIC.
Many people wait for Ros5 cause they want to increase routing speed in mikrotik.
Yes, many people use mikrotik as a router.
Now routing limited by cpu power.
To increase speed we have to use all processors.
To use all processors we have to buy NICs with multicpu (MSI-X+queues) support like Intel E1G142ET (82576),
but if driver will use only 1 CPU/NIC - where is no sense in Ros5 for routing purposes.

igb driver parameters:
IntMode 0-3
0 - Legacy Interrupts, single queue
1 - MSI interrupts, single queue
2 - MSI-X interrupts, single queue (default)
3 - MSI-X interrupts, multiple queues !!!!!!!!!!!!!!!!!!!!!!

RSS 0-8
1 0 - Assign up to whichever is less, number of CPUS or number of queues
X - Assign X queues where X is less than the maximum number of queues

NOTE: for 82575-based adapters the maximum number of queues is 4; for 82576-based and newer adapters it is 8.

----
we need next parameters:
"intmode=3,3" for E1G142ET(2 port NIC)
"intmode=3,3,3,3" for E1G144ET (4 port NIC)

+ we want to see string like "Using MSI-X interrupts. 4 rx queue(s), 1 tx queue(s)"
+ load of each cpu, like "Cpu0: 30%, Cpu1:30%, Сpu2:33%......"

Hello Normis, now you understand our question ??? ))

One queue per interface. Basically we are using defaults in all of the drivers except for our cards.

But why ??? more than one queue can increase router perfomence !!
And what is "our cards" RB816 ?

all routerboards have ethernets, and also we do have pci ethernet cards

But why you dont want use MSI-X interrupts and more than 1 queue ?? Please explain.
And what mikrotik ethernet card have Gigabit speed ? I only see 100M card on the site.
RB1000 does not have enought power fo routing my traffic

normis, please for the love of god enable multiple queues and msi-x for hardware that supports it. there is no other way RouterOS can ever reach 10G throughput and acceptable packets per second figures...

i can't exactly remember where i saw that video (from some new zealand network operators group meeting), but two guys from new zealand built an 8-core x86 linux-based router with Intel 10G cards that could push upwards of 3,5 million pps and about 40G traffic in total.

one critical part of this was using the multiple queues and msi-x interrupts for the intel 10G cards. if mikrotik doesn't want to completely fall behind in raw throughput performance - then please enable the msi-x and multiple queue per interface support.

especially the newer intel cards and (i think) some broadcom cards heavily benefit from that on multicore systems.

yes, please at least allow it a configuration item. we wont be able to continue to routeros our network in the coming years without it. end users are already getting 50-100mb internet connections, ISPs need way more than that.

It's funny that, until this thread, there were no problems

Its simple:
1. then many users loot at "CPU load" in mikrotik they see usage less then 25% and they think that all ok. But they don't know that load /per processors like '90/0/0/0' (1 ethernet/quad core) and it is very bad.
2. Every year traffic in networks dramatically increases, so now mikrotik perfomance not enougth to handle with new traffic.

To solve problem with high speed forwarding on PC people use "new" Intel NICs.

If ROS will not fully support new NICs, people will use freebsd, linux, vyatta,..., those who have enougth money - cisco routers.
Because of traffic increase in a year ROS usage will be limited to wireless or small networks

I'd even say that for ten years nobody seemed to notice any problems, so your statement doesn't make any sense (ie. nothing changed)

Problems are not present because it should not be )

For that time people dont have a big traffic, for now many of us have traffic more than 3-5 Gb/s. If you have 100 - 500 Mb/s traffic - than you dont need MSI-X and queue, but in other case you cant work without this.
Thats why many vendors (like intel, broadcom) create drivers and chip that supports many queue per port (to use not only one cpu core), and thats why many linux distribution start to use this drivers.
Vyatta, for example, says that Vyatta much better than Cisco and many people start to use Vyatta, especially after that document http://www.vyatta.com/downloads/whitepa ... ef_r04.pdf

Why Mikrotik dont whant to be a Cisco concurent ??? You dont ned money ??

You are the first to raise this question, so I guess it's not as important as you think it is. We will look into the matter though, thanks for the suggestion.

I'd even say that for ten years nobody seemed to notice any problems

Clearly you must have skipped over those countless reports of multi-core systems only using one core for packet routing using those drivers and enabling multiple queues would let multi-core systems use all available cpu power for routing. It should help a lot with packets per second throughput, and especially with 10G interfaces.

Enabling msi-x and multiple queues is definitely one step into the right direction

I raise this question 2 or 3 times, but no one intresting it, beacose no one intresting router perfomence with traffic more than 1 Gb/s (in real network, not syntetic test).
Normis - you can compile experimental package fo ROS, like Xen or wireless test, and if people need real perfomence thay can use this package...
I think many of us can test this and send feedback to mikrotik.

no one intresting router perfomence

I thought you said everyone needs it?

)) yes, my english is bad. I say that when i early create topic like this - no one need MSI-X and queue, beacose traffic speed in networks was less than 1 Gb/s and we dont have any trouble..
But now we all intresting that mikrotik use all network card futures !

intel keeps coming out with new chips and technologies, they know people will use it. routeros should implement these because if its working then people will use it : )

It's funny that, until this thread, there were no problems

I'd even say that for ten years nobody seemed to notice any problems, so your statement doesn't make any sense (ie. nothing changed)

well, what I have dealt with is "oops... one core of our quad system is overloaded... sh*t! we should add one more router =("

so, people silently suppress their despair and add more and more hardware - just because they know: RouterOS do NOT support multiqueueing, and MT developers are conservative, so they shouldn't expect that feature in near future...

but it would be extra-sweet if v5 come with this special thrill =)

Thanks Chupaka ))

After some research of RouterOS i found that forcedeth ethernet driver (Nvidia ethernet chip) have parameter that can enable MSI\MSI-X interrupts, but this module need to be insmod with parameters to enable MSI-X, it can be done via starting rc. scripts in RoS.
Intel driver compiled without any parameter, and use default slow settings.

I just want to raise my hand here as a person for whom this matters! We recently had this issue on our multicore linux routers - luckily it was adjustable there - and if we are to put RouterOS on them in the future, we want to be absolutely sure that they can balance the load around the available cores in the most efficient manner. Driver default does not always work best.

We have no plans to support multiqueue right now. However, we are working on the feature where there is one core per interface.

So what does

- updated SMP support for multicore

exactly mean ?

new kernel, improved stability.

We have no plans to support multiqueue right now.

I am sad to hear this
Just enabling one core per interface won't help at all if confronted with 10GbE or packet flooding... because one core is just not sufficient to fill a 10GbE interface, and in times of flooding every single cpu clock counts, even on 1Gb interfaces.

Edit: Also I can't see how it would be easier to implement some strange one-core-per-interface option if there is a better option available which can be implemented by just changing a driver setting.

We have no plans to support multiqueue right now. However, we are working on the feature where there is one core per interface.

huh... all our routers have 1 ethernet interface %) and I don't think VLAN interfaces will work on per core basis

We have no plans to support multiqueue right now. However, we are working on the feature where there is one core per interface.

huh... all our routers have 1 ethernet interface %) and I don't think VLAN interfaces will work on per core basis

You might want to have more than one interface then and distribute your Vlan's over them. At least it is a solution, that is usable to some extend.

As the traffic that is coming into your router also has to leave it, you should be able to archieve a nearly even split over 2 cores when Mikrotik has implemented this feature.

/M

yes, but the main traffic is via one interface, so there almost won't be the difference - outgoing packets don't need IRQ, do they? =)

im guessing the developers will look at implementing their own solution, if it doesnt work they will find out and implement standard intel configuration switches. Or their solution might be better ; ) Im sure they will get feedback and it will either be good or bad.

My preference, even without multiqueues, would be to just have a string field to allow driver configuration options on Intel NICs right on the interface window in winbox. Let us decide what the best performance parameters are for our situation.

huh... all our routers have 1 ethernet interface %) and I don't think VLAN interfaces will work on per core basis

Me too .

im guessing the developers will look at implementing their own solution, if it doesnt work they will find out and implement standard intel configuration switches. Or their solution might be better ; ) Im sure they will get feedback and it will either be good or bad.

My preference, even without multiqueues, would be to just have a string field to allow driver configuration options on Intel NICs right on the interface window in winbox. Let us decide what the best performance parameters are for our situation.

I agree . they implement their strange solutions . What is the matter with a configuration string or some parameters ?
Good programming way is not too hardcode everything .

他们说的问题确实存在
MSI-X 是必须的

其实MSI-X已经过时了!!! 但ros仍没有支持!

很多人cpu根本连25% 都没用完.

但是真正在用ros的都是小型ISP 他们需要MSI-X 希望能早日支持! 这对网卡性能很有帮助!

我们很需要MSI-X 或者更加好的解决方法.比如 I/OAT 但是我很无奈.我用英文根本不知道怎么表达! 世界上很多人也是跟我一样无法表达......

implement standard intel configuration switches

have a string field to allow driver configuration options on Intel NICs right on the interface window in winbox

Please tell me which options exactly do you want to see and how will these options help. Martini, Changeip, omidkosari, hedele, Eising, please share your opinion what exactly should be turned on like you say.

Most drivers have been preconfigured to support all what the hardware supports, so if you think there is something that needs to be turned additionally, please clarify what exactly you mean and we will research it.

http://www.intel.com/support/network/sb ... parameters

I for one would like to tune my cards with InterruptThrottleRate, Tx and RX descriptors. Also the MSI-X stuff looks very nice.

This readme talks about MSI queues and some other things related to kernel versions, etc.
http://downloadmirror.intel.com/13663/eng/README.txt

IntMode
-------
Valid Range: 0-2
Default Value: 2
0 - Legacy Interrupts, single queue
1 - MSI interrupts, single queue
2 - MSI-X interrupts, single queue (default)

Here is a nice writeup about reducing latency on the intel nics:
http://download.intel.com/design/networ ... 322819.pdf

Basically we are just asking to know how it's configured now, and if there is an option to use something like ethtool if we want to tweak it. The MSI stuff might only work if you can change kernel options on boot, I am not 100% sure if that can be turned on and off without recompiling or not.

It would be nice to know what version of the intel e1000, e1000e and igb drivers are in RouterOS.

Intel driver shipped with the all kernel have no parameter, and it use legasy function (fasteio), to use all functions you must download linux driver from intel site and compile it. Than you can use multiqueue, msi-x and other..
http://downloadcenter.intel.com/detail_ ... ldID=13663

this document describe tuning system fo 10G link
http://people.redhat.com/dshaks/Mark_10GbE_summit08.pdf

we need parameters :

InterruptThrottleRate
The driver can limit the amount of interrupts per second that the adapter
will generate for incoming packets. It does this by writing a value to the
adapter that is based on the maximum amount of interrupts that the adapter
will generate per second

IntMode
Allows changing interrupt mode and MQ status at load time, without requiring
a recompile. If the driver fails to enable a specific interrupt mode, it will
fail back to lesser capabilities.

RSS
Assign X queues where X is less than the maximum number of queues

QueuePairs
TX and RX will be paired onto one interrupt vector

All this options help us to attain max perfomence of 1G and 10G, especialy in system with more than 4 ethernet port. Its help us to reduse cpu load on high load link (high PPS) and much more..

Well since martini has already rolled up everything I can think of - I second his explanations on what would be useful for the intel cards.

How does InterruptThrotleRate differ from NAPI used by the driver in kernel? On high load NAPI will completly disable IRQs and use polling instead.

Why do you need to change IntMode, if all you want is MSI? MSI is going to be used anyway if it's supported.

Standard ixgbe kernel driver enables RSS by default (if card supports it) and number of queues will match the number of cpu cores.

So we don't see what exactly you want to configure there and how exactly it's going to help you.

try to run command - modinfo igb.ko
On kernel driver for intel in ROS 3.X - 5.X (and on non ROS kernel same) it show none of parameter.

But if you compile extended driver (from intel site) - modinfo show all parameters that i describe, and all will work.

I have many i82575-82576 based cards installed on different motherboard and none of them use MSI interrupt, beacose kernel shipped driver use only legacy mode (fasteio)

InterruptThrotleRate - on some routers it help to reduse packet drop.

I also need a more powerful engine, needs intel IOATA

Today we've setup a server - Intel® Server System SR1690WB with X5660 CPU to manage foreign traffic.
And installed MikroTik v5.0 beta 2 on it.
Right now two ethernets are connected and MikroTik shows 20% load (which should be read as 60%/60%/0%/0%/0%/0% per core) and 140 thousands pps.
Most likely it's connected to the new protocol µTP used by torrents.
Considering that four of six cores are idle in this situation - it would be more efficient and less system recourse demanding to assign more than one core per ethernet.
I hope there would be a multi-core per ethernet support in the next router OS version.

I hope there would be a multi-core per ethernet support in the next router OS version.

new intel driver in v5 beta 2 will use as many cores as there are available, as already written above

beta2?.. but he's testing beta2 =)

the distribution of load to cores was his imagination entirely, apparently there is simply not enough load for his machine.

maybe a bit offtopic, but... any news about per-core CPU monitoring?..

What about my question (really old topic)
http://forum.mikrotik.com/viewtopic.php?f=2&t=38519

new intel driver in v5 beta 2 will use as many cores as there are available, as already written above

Somebody have tested it?
You compiled drivers from Intel site as martini wrote?

I don't understand, I see R5 beta2 release 30/04/2010.
You ask about people opinion 6/05/2010, and 13/05 you say that b2 supports new drivers. Something changes with b2 from 30/04?

no, nothing changes, normis mean that v5 beta already have new drivers shipped with kernel, not new drivers from intel site

Well.. After reading this thread I understood clearly that MT customers has to get down on their knees beg and cry for a new feature to be only even considered by MT as a "needed feature". LOL

Well.. After reading this thread I understood clearly that MT customers has to get down on their knees beg and cry for a new feature to be only even considered by MT as a "needed feature". LOL

Well, maybe we will even have to bend over - we will see

Sorry, for offtipoc, but does MT going to support igb driver in ROS 4.x?
Looks like it better to use stables ROS version. And I've bought the card (E1G44ETBLK) already.

Ready to provide this card for testing to MT Team (if this needed).

Well, maybe we will even have to bend over - we will see

hahaha, are your fingers touching the ground ?

Hey at least there is no soap down here!
On the downside, there's also no better intel drivers there

I tested standart kernel 2.6.31-1 with compiled in igb driver 1.3.16-k2 (this is old driver on intel site driver version 2.2.9)
All works good even without setting parameters
cat /proc/interrupts
59: 0 0 1 0 0 0 0 0 PCI-MSI-edge eth2
60: 0 4714920 4 3 0 0 1 1 PCI-MSI-edge eth2-TxRx-0
61: 4 4750567 0 1 2 0 1 14 PCI-MSI-edge eth2-TxRx-1
62: 0 2 1 4705266 0 0 0 4 PCI-MSI-edge eth2-TxRx-2
63: 2 4 1 4673417 3 0 0 0 PCI-MSI-edge eth2-TxRx-3
66: 0 0 0 0 0 0 0 1 PCI-MSI-edge eth3
67: 2 0 2 0 0 3780613 0 0 PCI-MSI-edge eth3-TxRx-0
68: 8 0 0 0 0 3780403 3 0 PCI-MSI-edge eth3-TxRx-1
69: 0 1 0 3 2 0 0 5742831 PCI-MSI-edge eth3-TxRx-2
70: 1 2 3 2 0 0 0 5459227 PCI-MSI-edge eth3-TxRx-3

Why in mikrotik with compiled igb in kernel (new kernel igb driver should be the same I used) /proc/interrupts shows
using IO-APIC-fasteoi
using IO-APIC-fasteoi
instead PCI-MSI-edge
?

Are there still no plans for MSI-X support? Lack of support for that would be an immediate deal-breaker for me, so I would like to know for sure. I hope you have reconsidered, RouterOS otherwise looks to be quite capable.

It's already supported, it's automatically used if it's supported by the card. See previous posts.

Ok. If you say that msi-x suppported, i trust you, but how can I see it in winbox? hardware interrupts show only 1 interrupt per interface.
My card E1G44ET and supports MSI-X

you won't see it, it's internal driver feature, not RouterOS feature.

Do you have any issues with performance?

On standart linux with my Quadcore cpu proc/interrupt show 4 rx interrupt per interface (1/rx queue)
With high load I have drops/errors on interfaces.
I can't see what causes theese drops. I can tune system only by enabling or disabling some features and then I have to wait some time to see results. I always see cpu usage 20-40%
In standart linux I can see per/core usage, per process cpu usage.
How in mikrotik I can test router bottle neck?

hmm.. if MSI-X used in v5 beta - why i see only 'faste-io' interrupts used by ethernet card with cat /proc/interrupts ??
tested on 3 motherboard with intel E1G44ET and E1G42ET, and yes i did a little hack and use linux console to see what interrupts use ROS

It's already supported, it's automatically used if it's supported by the card. See previous posts.

Ah ok, I wasn't quite clear. Thanks!

you won't see it, it's internal driver feature, not RouterOS feature.

Normis, how do you know that msi-x works? What messages/logs you see?

Yesterday I installed new hardware,new kernel 2.6.32 (as in Ros5), driver automaticaly assigned 4 interrupts per interface. Then I set in flash with Ros5.02b. Ros shows
1 [0 0 IO-APIC-edge i8042]
4 [2 2 IO-APIC-edge serial]
8 [1 0 IO-APIC-edge rtc]
9 [0 0 IO-APIC-fasteoi acpi]
14 [0 0 IO-APIC-edge ide0]
15 [0 0 IO-APIC-edge i1e0]
16 [90123113 89633038 IO-APIC-fasteoi pata_marvell]
16 [uhci_hcd:usb3]
16 ether1
17 [41190614 41369363 IO-APIC-fasteoi eth2]
17 ether2
18 [33111985 33176336 IO-APIC-fasteoi ehci_hcd:us1]
18 [uhci_hcd:usb5]
...

. Only 1 int/per iface. There is just no place for additional 6 interrupts.
Or system/resources/irq shows invalid data, or igb driver in ROS 5.02b in some reason use non-msi interrupt (legacy mode).

dmesg, ethtool shows network card mode, if needed I can send supout

I cant belive in this !! ))) You did it !! ) MSI-X and IRQ balance support !
Normis and all mikrotik guys - thank you !

any experience with those features? =)

tomorrow post results of update

dont recomment to update production routers with intel ethernet that has 4, 8 or 16 queue per port (82575 or 82576) to ROS V5 beta 3
Have some strange issue, but router still work. (supout.rif sent to support)

could you delve deeper into what problems you have?

ticket 2010061966000026 and 2010061966000017

and another problem - on intel nics simple queue and queue tree leds router to crash (kernel panic)

and another problem - on intel nics simple queue and queue tree leds router to crash (kernel panic)

any updates on this topic?

on 5v beta4 queue on intel nics work better, if enable queue tree router just reboot, but dont kernel panic

on 5v beta4 queue on intel nics work better, if enable queue tree router just reboot, but dont kernel panic

I can confirm that my x86 has been rebooted on next sec, after I applied first simple queue rule.
Router started to reboot once again after reboot. The only one way to stop that was quickly login from console after reboot and remove queue rule.

If needed I can send my autosupport.rif.

Could this be fixed in 5B5?

One more thing :
My 5B4 cross connected to Ubuntu 10.04 TLS with Intel 1G card (I use 50sm. 1G croosscable for this). Ubuntu use ibg linux driver.
For some reason I have RX error on my 5B4 BOX intel card, and as result packet delay and packet lost between MT and Ubuntu.
Where can be the problem?

mangust: yes please send a supout to support. i would love to see this resolved.

strange, here i have about 270 simple queues and no reboot yet ( x86) and intel nics

strange, here i have about 270 simple queues and no reboot yet ( x86) and intel nics

It's only for 5B4. Believe me

strange, here i have about 270 simple queues and no reboot yet ( x86) and intel nics

It's only for 5B4. Believe me

still strange...

b4 here too..
[admin@mk_border] /queue simple> print count-only
272

[admin@mk_border] /system resource> pri
uptime: 1w6d10h21m49s
version: "5.0beta4"
free-memory: 1888708KiB
total-memory: 1944236KiB
cpu: "Intel(R)"
cpu-count: 2
cpu-frequency: 2660MHz
cpu-load: 35%
free-hdd-space: 153642568KiB
total-hdd-space: 153818816KiB
write-sect-since-reboot: 37002210
write-sect-total: 37002210
architecture-name: "x86"
board-name: "x86"
platform: "MikroTik"

still strange...

May be it's only for some specific Intel chips, (for example with MSI support)
What kind of Intel card do you use?

One more thing: I believe that MT Team know this problem...
Because if not, they asked me to send them support.rif ...

All intel gigabit , the last one if i recall correctly is have MSI support.

0 device="11:00.1" name="82546GB Gigabit Ethernet Controller (rev: 3)"
vendor="Intel Corporation" category="Ethernet controller" vendor-id="0x8086"
device-id="0x1079" irq=7 memory=0xED160000-0xED17FFFF,0xED180000-0xED1BFFFF
io=0x3040-0x307F

1 device="11:00.0" name="82546GB Gigabit Ethernet Controller (rev: 3)"
vendor="Intel Corporation" category="Ethernet controller" vendor-id="0x8086"
device-id="0x1079" irq=11 memory=0xED140000-0xED15FFFF,0xED100000-0xED1FFFFF
io=0x3000-0x3FFF

4 device="09:00.0" name="82572EI Gigabit Ethernet Controller (Copper) (rev: 6)"
vendor="Intel Corporation" category="Ethernet controller" vendor-id="0x8086"
device-id="0x10b9" irq=3 memory=0xEC120000-0xEC13FFFF,0xEC100000-0xEC1FFFFF

on 5v beta4 queue on intel nics work better, if enable queue tree router just reboot, but dont kernel panic

Any feedback for 5B5?

RoS V5beta5 still have issue with queue tree.
All problem with queue tree "global-in" parent, it seems that RoS cannot work with many rx queue on ethernet interface.. On "global-out" parent all works fine, because ethernet use only one tx queue.

yes it still drops the links after a while like beta 4

there´s anyway to support Broadcom MSI-X ?

The BCM5718 provides a PCI Express v2.0-compliant host interface, which can operate at 5GTs at x1 link width or at 2.5GT/s at x2 link width. The PCI Express supports MSI and MSI-X capabilities and Function Level Reset (FLR). The host interface supports a separate PCI Express function for each LAN interface. The BCM5718 includes I/O Virtualization (IOV) features such as 17 receive and 16 transmit queues, 17 MSI-X vectors with flexible vector-to-queue association.

yes it still drops the links after a while like beta 4

It's just drop links or reboot the router also?
What is your status of watchdog (On or OFF)?

this new drivers supports the Intel 82580EB chipset?

IMHO:
After days of googling I’ve decide that MSI/ksoftirqd are too fresh for v.5 Beta.
Looks like v5 really need support of new Intel chipsets, but support for MSI can wait.

If MT do not find way to fix kernel crash (what is starts from 5.0beta3 - 2010-Jun-17) in nearest mouth, then may be it will be good idea for MT to remove this feature (MSI) until v6 .

In any case QOS support was in ROS “dosen years” and if there will be a choice, then I choose support of QOS, but not support of MSI.
For anyone who want to use v5 BETA (for new Intels card) and QOS I can suggest use magic command :

/system hardware set multi-cpu=no

2 mangust - yes, but this magic command not need on heavy loaded routrer, because CPU load will be 100%. MSI need many cpu core, in other way you can use V4

is the v5 with multiple intel nics and cores is somewhat stable? we´re about to put 2x 1gbit transit providers (bgp / mpls / ospf) and some simple queues.

2 mangust - yes, but this magic command not need on heavy loaded routrer, because CPU load will be 100%. MSI need many cpu core, in other way you can use V4

a) V4 do not support new Intel chipsets (igb driver).
b) I suggest this command only as a workaround until some stable solution...

very strange i tried v5b6 and 4.11 and 3.30 my on Intel quad port Pro 1000/vt (82575gb) in a dell r200 and it does not detect. I ditto the msi-x requirement because we are seeing increasing bandwidth utilization by customers and eventually beyond 1GBps this is going to be needed

3 ethernet card i82575 works ok on different routers and on ros 5.02 and 5.06

What about support MSI? Debian 6.0 kernel 2.6.32-5-686 without modifications, use the PCI-MSI-edge for the AOC-SG-I2 (Dual i82575)

Result: 82575EB (AOC-SG-I2) 950Mbit () .... Debian cpu loaded to 3%, but Mikrotik loaded CPU to 50%!! (C2D E5200)

Tested on the same hardware and the same situation.

Is that so hard? basic Debian kernel can do it but not Mikrotik?

Hello,
i've just got a new intel mobo (S1200BTL - http://www.intel.com/products/server/mo ... ations.htm) for testing.
It would be a great for routing (and price also). It has 2 onboard NICs (as viewable on specs): i82574L and i82578M
the first nic is recognized, last one is no. The tech guy at distributor said it should work with same driver, so please include vendor/device id into source!

Details:
Vendor: Intel Corporation
Vendor ID: 0x8086

Name: unknown device (rev: 4)
Device ID: 0x1502

I would be pleased to test in beta ROS version, thanks!

I want to buy that board but also don't know if i82578M is supported in latest ROS 5.5 or 5.6. Anyone could check?

Hello,
i've just got a new intel mobo (S1200BTL - http://www.intel.com/products/server/mo ... ations.htm) for testing.
It would be a great for routing (and price also). It has 2 onboard NICs (as viewable on specs): i82574L and i82578M
the first nic is recognized, last one is no. The tech guy at distributor said it should work with same driver, so please include vendor/device id into source!

Details:
Vendor: Intel Corporation
Vendor ID: 0x8086

Name: unknown device (rev: 4)
Device ID: 0x1502

I would be pleased to test in beta ROS version, thanks!

Really??? Pls export the pci devices list.... It is very bad, if so...

Confirmed. Second built-in eth controller doesn't work on latest ROS v5.6

Edit: This one doesn't work i82578M.

Confirmed. Second built-in eth controller doesn't work on latest ROS v5.6

Edit: This one doesn't work i82578M.

wow... too bad...

Anybody have any experience with 10GbE cards?

Any news with support for the ethernet controller on Intel S1200BTL Motherboard?


12 device="00:19.0" name="unknown device (rev: 5)" vendor="Intel Corporation" category="Ethernet controller" vendor-id="0x8086" device-id="0x1502" irq=10
memory=0xC0A00000-0xC0A1FFFF,0xC0A60000-0xC0A60FFF io=0x3020-0x303F


Its not working on Mikrotik 5.6

send a supout.rif file to support, we will see if we can add support, or if it needs a newer kernel

Hey, I wrote same thing few posts earlier...

Hi,

After almost 1 year, NIC i82578M still doesnot work on v5.18

When it will be supported? ????

Regards

Have you tried 6.0 beta2?

Have you tried 6.0 beta2?

Using beta version is not recommend in our company policy

Have you tried 6.0 beta2?

Just upgrade to 6 beta2. Yes it can detect i82578M. Great! !!

But I cannot use beta for reallive. I really hope Mikrotik Team can support i82578M in stable version....very soon

Using beta version is not recommend in our company policy

In that case, you have to forgot using mikrotik i think. In many cases, the new version is wrong than the previous ones.
The changes are not well documented, so maybe you can think, the changes in the changelog does not counts in your side - but some changes not documented in log - so you don't know if the problem is in your side or mikrotik ROS, or hardware.

If you want to be sure, do NOT use the newest version, just those what qualifed and worked for weeks for other forum members. Yes, I also sold a new intel mobo based router - what worked only with 6.0b2. The customer is understand:
- he have to take the risk
- or buy an expensive PCIe multiport NIC
- or buy an old and much slower (and EOL'd) motherboard.

But I cannot use beta for reallive. I really hope Mikrotik Team can support i82578M in stable version....

v6.0 will be 'stable' =) soon...

Sorry, for offtipoc, but does MT going to support igb driver in ROS 4.x?
Looks like it better to use stables ROS version. And I've bought the card (E1G44ETBLK) already.

Ready to provide this card for testing to MT Team (if this needed).

Did You test this card (E1G44ETBLK) ?
I want to buy one the same - but found no information on how it will work...

this new drivers supports the Intel 82580EB chipset?

Did you ever figure this out?