Attempting workaround for MikroTik lack of support for DYNAMIC LACP

[mervincm]

I have a Synology NAS (1813+4GB with 2x1TB SSD and 6x3TB HDD with 4xGBE NICs) that I have been trying to get more that 1GBE worth of simultaneous bandwidth out of for months now. I saturate a single link in large sequential reads/writes easily. I understand that it must be multiple simultaneous loads, as any single connection can't exceed 1GBE. This limit existing even with multiple NICs connected to the NAS with an LACP bonded channel.

Synology only supports 2 options for the bonded link. IEEE 802.3ad Dynamic Link Aggregation, or Network fault tolerance only mode. Since I am looking for more performance (load balancing) only 802.3ad Dynamic Link Aggregation is helpful.

Problem #1 My switch, a MicroTik CRS226-24, does not support the dynamic messages portion of the 802.3ad spec within its switch hardware. Mikrotik suggested that I try it anyway as "it sometimes works" but I had no luck at all.

In an attempt to workaround the issue, I bought a switch to in between the NAS and the CRS226. Mikrotik support suggested that this should work. I bought Netgear GS110TP v200, by all accounts a well featured decent quality switch for the money.

Synology to Netgear
So far so good. I created a 4 port LACP channel on both sides, and both sides report a happy link. The NAS lists it as established and 4000 mbps.
Capture1.PNG - Google Drive
Capture2.PNG - Google Drive
Capture3.PNG - Google Drive
Capture4.PNG - Google Drive


Netgear to Mikrotik
Single cable, everything works just fine. I have been running it this was for weeks now without a single hint of an issue. As soon as I try to establish a bonded link between the Netgear and the Mikrotik all hell breaks loose.

I create another lag group on the GS110TP
Capture5.PNG - Google Drive
Capture6.PNG - Google Drive
Capture7.PNG - Google Drive
Capture8.PNG - Google Drive

and I also created one on the CRS226 (capture 9 and 10)
Capture9.PNG - Google Drive
Capture10.PNG - Google Drive


I quickly need to pull 3 or the four cables and reboot my Mikrotik to regain connectivity. I am assuming a loop is created / detected or something like that shuts it down.

I have emailed MikroTik support, but no response for weeks.

 

[Dk3]

After reading around, there is no real success on this switch to work in LACP. Even if success, it will likely degrade the throughput.
I had already bought two of this, intent to use it in W2012R2. With multipath enable, it simply uses all connection port and also, redundancy on another switch.

I felt that this switch is more suitable on multipath capability features like ISCSI and Windows SMB. Honestly, with such a price having 2x 10gb its really unbeatable.

 

[Entz]

Yeah LACP wont work (tried myself) , but standard link aggregation techniques do work fine with it i.e. I use balance-xor on my linux clients. Not sure how you do that switch to switch (never seen that option in a Netgear). SMB multipath works great as well but would on any switch .

From what I understand you can edit the config files and change the bonding mode on your Synology to support the standard linux modes (/etc/sysconfig/network-scripts/ifcfg-bond0 => BONDING_OPTS) even if the UI doesn't show it.

 

[mervincm]

THX Entz
I might try the NAS change, I didn't know about that option. I am really hesitant to go around the gui.. The appliance concept is why I paid the premium for it in the first place, alas I might not have any other options.

 

[mervincm]

After reading around, there is no real success on this switch to work in LACP. Even if success, it will likely degrade the throughput.
I had already bought two of this, intent to use it in W2012R2. With multipath enable, it simply uses all connection port and also, redundancy on another switch.

I felt that this switch is more suitable on multipath capability features like ISCSI and Windows SMB. Honestly, with such a price having 2x 10gb its really unbeatable.

SMB3 multipath is really nice in theory. I found implementation was not so smooth. I still have it on my "implement later pile" but my initial test showed issues with a slow ramp up in speed (1 link starts at full speed) plus asymetrical performance issues.