What's new

Tips for NAS Link Aggregation testing

  • SNBForums Code of Conduct

    SNBForums is a community for everyone, no matter what their level of experience.

    Please be tolerant and patient of others, especially newcomers. We are all here to share and learn!

    The rules are simple: Be patient, be nice, be helpful or be gone!

Dennis Wood

Senior Member
Tim, regarding link aggregation testing, here are a few comments based on 3 different NAS units, 4 different 802.3ad capable switches, and many tests between dual NIC workstations.

1. Aside from older (buggy) Qnap TS509 firmware, and Nvidia's dual NIC "teaming" option, link aggregation on a given workstation has not increased read or write performance to or from a NAS unit in any of our testing. This is where both the NAS and workstation had dual NICs configured as 802.3ad compliant, on a properly configured 802.3ad switch.

2. In terms of your testing, I'd clear, then monitor the port stats (using the gigabit switch GUI) for each member of the LAG to the NAS unit in question. This is a pretty easy way to see what's going on in terms of RX/TX volume on each port. I have seen differences between switches, and certainly clients in terms of how LAG is handled. Monitoring the switch ports of both NAS and test workstation(s) though tells the complete story of what's going on.

3. LAG does make a big difference if you have two RAID0 workstations hitting the NAS unit where the NAS unit has read or write performance in excess of 100MB/s or so (again, obviously not the case for your test on the 809). For example, we've got a TS-639 Pro here with six 1TB drives in RAID0. Read/write speeds to the box from 1 raid0 workstation hovers at about 94MB/s sustained large file read. From two raid0 workstations (Vista SP2) the aggregate is just under 130MB/s and the port stats on the switch show that each workstation is basically getting its own pipe to the NAS. This NAS in RAID5 mode has pretty slow write speeds..so we use it as a fast backup in RAID0. This is using an HP procurve switch.

In other words, I haven't seen what you've seen on the 809 but I'll run this same test to a GS108T connected to a TS-509 unit in LACP mode. Link aggregation in my opinion is not quite the standard it should be between manufactures of switches, NICS, embedding chip sets etc., and the testing results show this.

Cheers,
Dennis.
 
Last edited:
The ReadyNAS Pro (six 2TB WD Black in XRAID-2) from Netgear in dual workstation tests, with the NAS connected using two ports in 802.3ad mode, shows that it can serve up a large file at approx 80MB/s, while another workstation is wriing to the NAS at about 80MB/s. This is pretty impressive in terms of serving up typical files which for most folks aren't anywhere close to our 36GB test file. While two workstations are reading large files simulaneously, aggregate is lower, about 119MB/s. Why? See below.

Using the TS-639 PRO in load balancing mode, and using six drives in RAID 0 the best we saw was an aggregate of about 100MB/s, when the same two workstations were reading or writing simultaneously. Keep in mind that RAID0 on a NAS across 6 drives is not something you'd typically do. With this NAS, I'd say dual NICs are likely pointless in terms of load balancing. We'll do more tests later on the TS-509 which falls between the aforementioned NAS units in terms of performance. We'll look too at how QNAP is dealing with 802.3ad.

The HP Procurve port stats show that the Netgear ReadyNAS Pro in load balancing mode (802.3ad) is using one port in the LAG primarily for TX, and the other one for RX. The QNAP units seem to do the same thing. This would suggest that dual NIC ports in 802.3ad mode don't help during simultaneous reads from two workstations. This also explains why we're seeing better aggregate performance with the ReadyNAS Pro when one workstation is reading a file while another is writing. Basically one workstation has it's own TX pipe while the other has its own RX pipe, each limited to Gigabit wire speed. This information does have an impact on load testing NAS units as regardless of what the manufacturer will claim for data rates, the most you'll likely every see is 120MB/s of TX or RX traffic at any time when using two ports in 802.3ad load balancing mode. So really, future (faster) NAS units with 2 port 802.3ad, would be limited to 120MB/s RX and 120MB/s TX simultaneously. Netgear does have other trunking options on the NAS which may perform differently.

Anyone for a 4 port 802.3ad NAS?
 
Last edited:
Thanks for the report, Dennis.

I found a couple of things that may help explain what you're seeing.
http://en.wikipedia.org/wiki/Link_aggregation#Order_of_frames

And these two documents linked from the Wikipedia article:
http://support.dell.com/support/edocs/network/LAG1855/LAGConsiderationv0.5.pdf

http://grouper.ieee.org/groups/802/3/hssg/public/apr07/frazier_01_0407.pdf

In particular, 802.3ad does not mandate the distribution algorithm. So another switch could produce different results.

While the distribution algorithm isn't mandated, it is mandated that the algorithm can't change frame ordering in a conversation and can't duplicate frames.
 
You're right Tim..those are excellent reads. If I gather all this correctly, one might surmise that an 802.3ad spec is no guarantee that a switch is actually the best for load distribution. The Dell doc surprised me in terms of the method of load balancing and is information that would be essential to effectively set up a network for best performance.

Knowing that the Draytek routers use an actual load algorithm on their dual wan systems, one wonders then if their 802.3ad implementation would be more technically advanced than the Dell product you linked to. Hmmm.
 

Similar threads

Latest threads

Sign Up For SNBForums Daily Digest

Get an update of what's new every day delivered to your mailbox. Sign up here!
Top