topic Re: teaming throughput help needed in Networking

teaming throughput help needed

donkarnash_1 — Wed, 18 Oct 2006 01:27:00 GMT

Hello,
I have ML570 G3 server and a netgear gigabyte unmanaged switch. The NICS are Gigabyte NICS and I have teamed together a pair of NIC's for the live (public) network and one for the private network (192.168.23.X). The Live network is for the public while the Backup network is connected to the Netgear switch......
please see the attached doc. Thank you

Re: teaming throughput help needed

rick jones — Wed, 18 Oct 2006 19:23:19 GMT

A couple things - if the teaming is using RoundRobin packet scheduling it could be causing TCP to make spurrious retransmissions.

If it isn't using RR scheduling, the traffic of a single flow will be no more than what one link in the aggregate could hold.

Filetransfer via SMB etc is often request/response, which means that bitrates are not always everything.

A single lost TCP segment can ruin your entire thoughput.

What do unmanaged siwtches do when asked to aggregate links? I'd have thought that they might not do good things in that situation.

I assume you mean giga_bit_ links and not gigabyte links as a "gigabyte" link would be a 10 gigabit link and I doubt many people would spring for four of those puppies these days :)

Re: teaming throughput help needed

donkarnash_1 — Thu, 19 Oct 2006 02:23:36 GMT

Ricky,
Firstly, yes I wanted to say giga_bit_links so sorry about that. Very honesly I did not understand your reply. Sorry....but the switch is Netgear GS516T 16-port Copper Gigabit Switch.
How should I configure the NICS ??

Re: teaming throughput help needed

rick jones — Thu, 19 Oct 2006 11:42:26 GMT

I would start by eliminating variables. First, make sure the Netgear switch supports LACP, which I assume is the protocol Windows will used when you select "Switch-assisted Load Balancing" If not LACP then Cisco FastEtherchannel or whatnot.

Next, try the backup network _without_ teaming/bonding/trunking going. Perhaps with something like netperf TCP_STREAM to avoid filesystem issues.

If/when trunking is working, a single TCP connection (aka flow) will go no faster than it could over a single link. Depending on the platform, the "packet schedulers" will pick a link based either on the MAC address(es) or IP addresses or TCP port numbers. They are geared towards increasing the performance of an _aggregate_ of flows - ie multiple connections.

If your backup software only does one TCP connection, the teaming won't do very much at all to increase performance.

Re: teaming throughput help needed

Connery — Thu, 19 Oct 2006 20:55:36 GMT

Don,
Your problem is that an unmanaged switch does not support the "Switch-assisted Load Balancing" (SLB) team type. We called it "switch-assisted" to infer that the switch must be configured for static port trunking/channeling. Since your switch is unmanaged, you can't configure it to work with SLB.

For the best throughput to an unmanaged switch, you should choose Transmit Load Balancing.

I wrote an 80+ page whitepaper that provides lots of information about team types and switch configuration guidelines. Link is here:
ftp://ftp.compaq.com/pub/products/servers/networking/TeamingWP.pdf

Best regards,
-sean