Welcome to TP-LINK Tech Support Forum
+ Reply to Thread
Results 1 to 7 of 7
  1. #1

    Question T2600G-28TS - Problem with LACP and Jumbo Frame

    Model :

    Hardware Version :

    Firmware Version :

    ISP : [/COLOR]

    Hi guys,
    I'm writing here because I need your precious help.
    I think to probably have a little problem of configuration on my network and I'llI try to explain in simple way the problem and my network's configuration. Thanks to all for support and patience !
    I have an ASUS AC87U Router, a T2600G-28TS and two identical client with AMD FX 8350 processor, 32 Gb of ram and a SSD as primary disk.
    In each of these two clients I have installed a
    HP 331T Quad Port 1GbE Network Adapter (which is moved by a Bradcom chipset and which is managed using Bradcom Advanced Control Suite 4).
    Using Broadcom Control Suite I create a Team using all 4 lan port and on my T2600G-28TS I did a link aggregation (4 lan port of the switch for the first pc and other 4 lan port of the switch for the second PC).
    The cables are CAT6A FTP,
    The problem is that when I try to transfer data from PC1 to PC2 (form a source folder that on PC1 is on SSD to a target folder that on PC2 is always on a SSD) I don't see throughput of the Team adpater which I can see regularly working under lan device of my Windows 10.
    Team device show 4Gbit connection while data transfer is of 130 Mbyte (so more near to a 1Gbit connection).
    In my idea I think to reach 300-400 Mbyte of speed but it isn't. There is something wrong in my idea?

    I did some test using iperf and i decide to work on the same pc as server and as client infact from a DOS command line I launched this command :
    iperf3.exe -s -i 5

    and from a second DOS command line :
    iperf3.exe -c ipserver -t 60 -i 5

    and this is the result:

    [ ID] Interval Transfer Bandwidth
    [ 5] 0.00-60.00 sec 0.00 Bytes 0.00 bits/sec sender
    [ 5] 0.00-60.00 sec 26.2 GBytes 3.74 Gbits/sec receiver

    It seem to be correctly aggregated but I can see this speed when I move data from PC1 to PC2. Can be a reason? Could I check something on the switch?

    The second problem is jumbo frame.
    I set MTU 9000 to all 4 lan adapter of the HP 331T Quad Port 1GbE Network Adapter, so Team adapter has 9000 MTU.
    I try this command “ping IP address –f –l 9000" but the answer is that packet need to be fragmented and I don't undestand the reason.

    On the switch (for the first PC) in LAG session I have

    1 Active LACP 1/0/17,1/0/19,1/0/21,1/0/23



    Hash Algorithm SRC IP + DST IP

    Detail info :

    Group Number:
    LAG1

    LAG Type:
    Active LACP

    Port Status:
    Enable

    Speed:
    Auto

    Flow Control:
    Disable

    Ingress Bandwidth (bps):
    --

    Egress Bandwidth (bps):
    --

    Broadcast Control (bps):
    --

    Multicast Control (bps):
    --

    UL Control (bps):
    --

    QoS Priority:
    CoS 0

    Join VLAN:
    1



    I don't know reason why bps field are blank.

    I hope to be sufficiently clear in the exposition. If you need more info please tell me .
    Thanks you very much for time and support that you will dedicated to me.






  2. #2
    Members R1D2 is on a distinguished road
    Join Date
    Dec 2015
    Posts
    1,637
    Quote Originally Posted by AINEMORF View Post
    It seem to be correctly aggregated but I can see this speed when I move data from PC1 to PC2. Can be a reason? Could I check something on the switch?


    You selected SRC IP + DST IP as the LAG hash algorithm. Should use SRC MAC + DST MAC to make use of more than one LAG channel.

    As for the jumbo frames: sometimes the 9kB specification includes TCP headers, sometimes not, so try to set a slightly lower MTU.

  3. #3
    Thanks a lot for your answer.
    I made the change you indicated to me but I can't see any improvement but I have a doubt.
    I know that 802.3ad standard (which is the standard for Link aggregation) does not increase the bandwith for a single conversation.
    This means that also if I have a quad lan port on each of the two computers if a transfer a single file of 4 Gb, from a pc to another, the speed is of 1 Gbit/s.
    I could see the improvement only if I transfer 4 file of 4 Gb. In this situation I could see 4 transfer process which can go ad 1 Gbit/s.

    Now the doubt : BASF (Broadcom Advanced Control Suite driver) has 3 option when I go to create team :

    1-Load balancing and failover (it is not my case because I'm trying to improve performances of speed)
    2-802.3ad Link Aggregation using Link Aggregation Control Protocol (LACP)
    3-FEC/GEC Generic Trunking

    I choose option 2 and the suite create my virtual lan adpater (called Team 1).
    Doing this, if I go on the web pages of the switch under SWITCHING-LAG i see the Member correctly populated:


    Global Config
    Hash Algorithm: SRC MAC + DST MAC
    LAG Table
    Select Group Number Description Member Operation
    1 Passive LACP 1/0/18,1/0/20,1/0/22,1/0/24 Edit | Detail
    2 Passive LACP 1/0/17,1/0/19,1/0/21,1/0/23 Edit | Detail


    In other way do I have a change to see for a SINGLE transfer a speed > 1 Gbit/s? (working on configuration of my switch or on configuration of my pci lan quad adapter)

    On youtube there is a video (linustechtips link aggregation) which speaks about RSS and SMB3.0 and he tells that he was able to transfer a single file at 3.60Gbit (you can also in the video) because his configuration splits traffic on the 4 cable. Infact if we see the video and his W10 perfromances monitor we can see 4 lan apdator and not 1 (he has no teaming). I try also in this way enablin RSS and SMB3-0 but nothing (I have tremendous performances).

    Thanks a lot fot your support.
    Last edited by AINEMORF; 04-17-2018 at 09:05.

  4. #4
    You are correct for the question about jumbo frame
    Infact if I try this command from PC1 : ping PC2 -f -l 8970 it gives us a correct result (also if from PC2 I try : ping PC1 -f -l 8970).
    We are very very near to the limit of 9000 for maximun MTU.

  5. #5
    Members R1D2 is on a distinguished road
    Join Date
    Dec 2015
    Posts
    1,637
    Quote Originally Posted by AINEMORF View Post
    In other way do I have a change to see for a SINGLE transfer a speed > 1 Gbit/s? (working on configuration of my switch or on configuration of my pci lan quad adapter)
    IMHO this depends on implementation of the interfaces used for the LAG. If the 4 ports are independent interfaces they should have different MACs even in a LAG, at least on the system I did use channel bonding on (server running IRIX). But Windows and even Linux might use another scheme. I'm not familiar with those. Currently I just use LAGs between TP-Link switches, not servers or even client devices.

    Maybe it's easier to consider buying a 10Gbit/s interface card to increase bandwidth of a single system/single session.
    Last edited by R1D2; 04-19-2018 at 00:38.

  6. #6
    Hi,
    thanks a lot for your answer.
    I took some steps forward and I wanted to share with the results.
    Let's start by saying that, as the standard says, LACP does not increase speed but increases troughtput and these are extremely different concepts.
    If, as in my case, we have quad lan adapters we have a transfer rate of 1 Gbit / s. The difference is that we have more lanes (many as the ports made available by the network card that we will aggregate and configure on the switch) .
    This operation allows you to transfer 4 files simultaneously at a rate of 1Gbit / s.
    If you want to increase the speed you do not have to teaming and use the LACP protocol, but you have to work on the individual adapters focusing on the parameters: jumbo frames, receive buffers, transmission buffer, RSS queus.
    You need to have enabled the SMB 3.0 protocol and have a processor that has a good number of physical cores (for example 8).
    In this scenario with my FX8350 (4 core) and this setup

    Ethernet 15 802.3az EEE (Copper Only) Disable
    Ethernet 15 Flow Control Auto Negotiation
    Ethernet 15 Interrupt Moderation Enabled
    Ethernet 15 Jumbo Mtu 8970
    Ethernet 15 Large Send Offload V2 (IPv4) Disabled
    Ethernet 15 Large Send Offload V2 (IPv6) Disabled
    Ethernet 15 Maximum Number of RSS Queues RSS 4 Queues
    Ethernet 15 ARP Offload Enabled
    Ethernet 15 NS Offload Enabled
    Ethernet 15 Priority & VLAN Priority & VLAN Enabled
    Ethernet 15 Receive Buffers Maximum
    Ethernet 15 Receive Side Scaling Enabled
    Ethernet 15 Speed & Duplex (Copper Only) Auto Negotiation
    Ethernet 15 TCP/UDP Checksum Offload (I... Rx & Tx Enabled
    Ethernet 15 TCP/UDP Checksum Offload (I... Rx & Tx Enabled
    Ethernet 15 Transmit Buffers 600
    Ethernet 15 Virtual Machine Queues Disabled
    Ethernet 15 VMQ VLAN Filtering Disable
    Ethernet 15 Wake on Magic Packet Enabled
    Ethernet 15 Wake on Pattern Match Enabled
    Ethernet 15 EEE Control Policies (Coppe... Optimal Power and Performance
    Ethernet 15 Network Address D89D672DA0D9
    Ethernet 15 VLAN ID 1
    Ethernet 15 Ethernet@WireSpeed (Copper ... Enable
    Ethernet 15 WOL Speed (Copper Only) Lowest Speed Advertised

    I have 220 Mbyte/s transfer rate. The video on youtube showed 450MByte/s but that guys have 8 processor so I think 220 MByte/s can be a good results.

    BUT this was only an experiment because i want to team the 4 lan port into a virtual team card and i want to use LACP protocol

    NOW THE new PROBLEM :

    I want to create on TPLINK T2600G-28TS switch three different LAG group. Every of this group is made by three ports.

    LAG Table

    Select Group Number Description Member Operation




    1 Passive LACP 1/0/17,1/0/19,1/0/21,1/0/23 Edit | Detail
    2 Passive LACP 1/0/18,1/0/20,1/0/22,1/0/24 Edit | Detail
    3 Passive LACP N/A Edit | Detail






    Group number 1 is the port of PC1, group number 2 the port of pc 2 and group 3 the port of the NAS that at the moment I don't have but I'll buy in the future.

    Port Config
    UNIT: 1LAGS



    Select Port Type Description Status Speed Duplex Flow Control Jumbo LAG
    DisableEnable 10M100M1000MAuto HalfFullAuto DisableEnable DisableEnable



    1/0/1 Copper Enable Auto Auto Disable Disable ---
    1/0/2 Copper Enable Auto Auto Disable Disable ---
    1/0/3 Copper Enable Auto Auto Disable Disable ---
    1/0/4 Copper Enable Auto Auto Disable Disable ---
    1/0/5 Copper Enable Auto Auto Disable Disable ---
    1/0/6 Copper Enable Auto Auto Disable Disable ---
    1/0/7 Copper Enable Auto Auto Disable Disable ---
    1/0/8 Copper Enable Auto Auto Disable Disable ---
    1/0/9 Copper Enable Auto Auto Disable Disable ---
    1/0/10 Copper Enable Auto Auto Disable Disable ---
    1/0/11 Copper Enable Auto Auto Disable Disable ---
    1/0/12 Copper Enable Auto Auto Disable Disable ---
    1/0/13 Copper LAG_3_Nas Enable Auto Auto Disable Disable ---
    1/0/14 Copper LAG_3_Nas Enable Auto Auto Disable Disable ---
    1/0/15 Copper LAG_3_Nas Enable Auto Auto Disable Disable ---
    1/0/16 Copper LAG_3_Nas Enable Auto Auto Disable Disable ---
    1/0/17 Copper LAG_1_Matteo Enable 1000M Full Enable Enable LAG 1
    1/0/18 Copper LAG_2_Serena Enable 1000M Full Enable Enable LAG 2
    1/0/19 Copper LAG_1_Matteo Enable 1000M Full Enable Enable LAG 1
    1/0/20 Copper LAG_2_Serena Enable 1000M Full Enable Enable LAG 2
    1/0/21 Copper LAG_1_Matteo Enable 1000M Full Enable Enable LAG 1
    1/0/22 Copper LAG_2_Serena Enable 1000M Full Enable Enable LAG 2
    1/0/23 Copper LAG_1_Matteo Enable 1000M Full Enable Enable LAG 1
    1/0/24 Copper LAG_2_Serena Enable 1000M Full Enable Enable LAG 2



    If I try in this configuration a single transfer this is slow only 40 Mbyte/s and if I try multiple transfers, for example 4, speed decrease and it is divided a 10 MByte/s for single transfer .
    Obviously here there is something wrong because it works opposite to the standard of 802.3ad...
    But were is the error? Do you see something wrong?
    Please help me...
    Thanks
    Last edited by AINEMORF; 04-20-2018 at 19:56.

  7. #7
    Members R1D2 is on a distinguished road
    Join Date
    Dec 2015
    Posts
    1,637
    I don't know whether this is the error, but AFAIK all your systems need to process Jumbo frames if one device does to get full benefits of using Jumbo frames. IMHO, LAG makes only sense if you have at least a number of different sessions (with different IPs or MACs) equal the number of ports in the LAG. That's not the case if you transfer 4 files from the same server system to the same client system.


 

Tags for this Thread

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts

Copyright © 1996-2018 TP-LINK Technologies Co., Ltd. All rights reserved.