TS-509 pro link aggregation problem (STILL not fixed!!!)

[stretch]

This may not be entirely helpful, but it might be informative. LACP is a useful protocol for load-balancing and rapid failover between two or more ports on the same switch. Generally, I use them inside a datacenter to do things like build the trunk connections between core switches. The further you get from the switching core, the less advantage you get from LACP, PAGP, or similar.

What it will not do is give you more than the speed of one link, when considering the throughput from a single source to a single destination. This is just the nature of the standard. It's just as true on my 150-pound cisco switches and very high-end servers. Where LACP shines is when you have a good number of source and destination hosts. My personal throught is, if I've got enough hosts to generate that kind of sustained load, I'm looking at a rack full of SAS drives and fiber channel, not a NAS that I can lift one-handed.

As has been said before, if you had several devices copying to and from the NAS, you'd quite possibly see over 1gb/s of aggregate throughput. I was able to observe 1.3gb/sec, using a Cisco 3560 switch and two fairly good desktop PC's connecting to an SS-839, moving largish (1-3GB) files.

There is a dirty workaround: If you are doing something like running VMware, and have multiple hosts on a single machine (especially if that machine itself has LACP capability), and the MAC hash comes up your way, you can end up watching all switch activity lights go really fast. If I absolutely needed more than 1gb/s just between two machines, I'd aim for a hack like that.

If you have more than one switch, be very aware of spanning-tree's effect on the network. I've had more than one experienced person thrash a network because they split a "teamed" pair of NIC's across two switches. The results are unpredictable and hard to troubleshoot, and unless your switch has good diagnostics...

While the novelty of seeing a box that generates almost no heat and consumes less power than some of my laptops push more than a gig of data across the network was cool, I found I'm happier using the "Balance-alb" config seems to be very stable, requires no swich config, and won't make spanning-tree freak out if I accidentally connect one leg to the wrong switch.

[jello_moto]

What it will not do is give you more than the speed of one link, when considering the throughput from a single source to a single destination. This is just the nature of the standard..

No, that is baloney. When I have a server with two teamed gigabit NIC's using LACP I get a solid two gigabits of throughput to the SAN. Please don't chime in with your observations when they aren't germane to the solution of this problem.

[stretch]

What it will not do is give you more than the speed of one link, when considering the throughput from a single source to a single destination. This is just the nature of the standard..

No, that is baloney. When I have a server with two teamed gigabit NIC's using LACP I get a solid two gigabits of throughput to the SAN. Please don't chime in with your observations when they aren't germane to the solution of this problem.

Ok. I'm only running two datacenters full of HP servers and 9 racks full of Brocade/HP SANs, and a little under two million bucks worth of Cisco network gear, so I'll yield to your greater expertise, and not inject my misunderstanding of what manufacturer-specific as opposed to standards-based options might be relevant.

PS -

IMG00082-20090822-1605.jpg

Maybe just beginner's luck, but...

[jello_moto]

Please stop "contributing" to this thread. You are not helping in any way.

[bmanfield]

Same problem replicated here. Dual port LACP trunk implemented between HP 1800-24G Procurve switch and TS-439Pro NAS. Data is only transmitting across NIC one and not load balancing let alone providing agregated throughput.

Is this still being looked into?

[petur]

Please stop "contributing" to this thread. You are not helping in any way.

At least he tried, right before you told him to get lost

Keep it positive, guys....

[Mildahp]

now i'd like to see qnap what's going to do... would be nice if at least someone from qnap says that someone is working on it and this problem will be sorted in next firmware ... or is it hardware problem?? please qnap stop foging things tell us the truth.... most of us will this more appreciated than pretending nothing is happening..... right lad??

[jello_moto]

Coming up on 6 months now. When are you guys going to implement the full IEEE 802.1AX standard? This is also referred to as LACP or "IEEE 802.3ad".

But lets go ahead and get modern by calling it IEEE 802.1AX.

[bmanfield]

Has this issue been resolved? My Qnap 439Pro is showing data moving across both ports (under the System Information status page) as of 3.1.2 1014T without any changes made to LACP on my switch.

Cheers
Ben

[jello_moto]

Has this issue been resolved? My Qnap 439Pro is showing data moving across both ports (under the System Information status page) as of 3.1.2 1014T without any changes made to LACP on my switch.

Cheers
Ben

Go look at the traffic stats for your switch and you'll find that its not distributing the load or bonding the two Ethernet channels.

If QNAP would just implement the full LACP/IEEE 802.1AX channel bonding standard, we could all just call it a day and be done.....

[AndyChuo]

Has this issue been resolved? My Qnap 439Pro is showing data moving across both ports (under the System Information status page) as of 3.1.2 1014T without any changes made to LACP on my switch.

Cheers
Ben

Go look at the traffic stats for your switch and you'll find that its not distributing the load or bonding the two Ethernet channels.

If QNAP would just implement the full LACP/IEEE 802.1AX channel bonding standard, we could all just call it a day and be done.....

Hi jello_moto,

just to check with the test method you performed again, can you describe briefly?

Andy

[jello_moto]

setup:
Windows 2008 server using intel pro 1000/pt dual port NIC teamed to 2 gigabit aggregate throughput. Connected to RAID5 iSCSI disk on the TS-509 pro through an HP procurve 1800-24 switch with 802.3ad configured.

Configure the QNAP to 802.3ad.

Copy a file from the QNAP iSCSI disk to a local disk and you will not get over 1 gigabit of throughput. If you look at the traffic on the switch (see the 1st page of this thread) you will see that almost all the traffic on the QNAP is going over only one of its ports.

If QNAP would just implement the full LACP/IEEE 802.1AX channel bonding standard it would have a higher peak throughput.

[Mildahp]

i agree with this...509 seams to be in link aggregation (trunk L8 it says) in HP pro curve 8G but i am getting terible speed...(start at 120mb/sec and in 5 sec drop down to 15mb/sec) in qnap is sellected link aggregation switch is sellected for link aggregation dual ethernet on each site Cat6 on desktop 2x500gb hard drives in raid 0 Qnap 509 upgraded procesor+RAM environmental= desktop windows7 X64 2.66GHz quad core 8GB RAM + Macbook pro 2.53MHz 4GB ram ...................... please what could be possible wrong?? have checked all + have changed all cable for CAT6e UTP (dearest ones) thank Milan

[jello_moto]

i agree with this...509 seams to be in link aggregation (trunk L8 it says) in HP pro curve 8G but i am getting terible speed...(start at 120mb/sec and in 5 sec drop down to 15mb/sec) in qnap is sellected link aggregation switch is sellected for link aggregation dual ethernet on each site Cat6 on desktop 2x500gb hard drives in raid 0 Qnap 509 upgraded procesor+RAM environmental= desktop windows7 X64 2.66GHz quad core 8GB RAM + Macbook pro 2.53MHz 4GB ram ...................... please what could be possible wrong?? have checked all + have changed all cable for CAT6e UTP (dearest ones) thank Milan

Read the thread from the beginning. The version of 802.3ad that QNAP has implemented does NOT do channel bonding it does failover and load balancing only to different destinations.

[schumaku]

Configure the QNAP to 802.3ad. Copy a file from the QNAP iSCSI disk to a local disk and you will not get over 1 gigabit of throughput. If you look at the traffic on the switch (see the 1st page of this thread) you will see that almost all the traffic on the QNAP is going over only one of its ports.

Was not looking into this for a while. At the early testing of 802.3ad, I also found it does come very near to GbE performance (including Ethernet and IP/UDP packet overhead) but was unable exceeding the one GbE when using a single data flow. My conclusion was that 802.3ad compliance requires the transmission of every packet of any particular flow over the same link. That could be wrong, but this is what I was reading from the Linux kernel bonding code.

In multiuser environments with very active access - SAMBA, nfs, AFS, ftp, ... - I found this was not a limiting factor.

A brief test transferring a bunch of big files by cp over an nfs confirms your observations, here on the v3.2.0 1207 build.

Now, when looking at say serving a single iSCSI target with one or multiple LUN ... it might be interesting to get > 1 Gb/s when 802.3ad is active. Do not know if and how the policy mentioned above has changed with 802.3ax, and do not have time and resource to review the IEEE papers nor recent kernel.org sources.

Somebody from QNAP and some interested people from the community here should look into that ASAP.

-Kurt.