up vote 2 down vote favorite
share [g+] share [fb]

I want to achieve the best performance with my two servers (SAN and ESX). I have good RAID (LSI, SASes), that is showing 1GB/s result, so the bottleneck now is network part since I have only two NICs on each port (Intels). I have also 4 Broadcoms.

If I`ll team 2 Intels with 2 Broadcoms on each server, will I achieve 4 Gbps performance?

I know that 10Gb NICs would suit my needs better but I do not have this option right now.

link|improve this question
What protocol are you using to connect to the "SAN"? iSCSI? – HampusLi Dec 16 '11 at 15:21
Yes, I`m using iSCSI based SAN – James Dec 16 '11 at 16:02
feedback

3 Answers

up vote 5 down vote accepted

Yes, this works fine in ESX/ESXi - we do it on all our servers.

Just make sure that you realize the implications - you'll loose any features that isn't supported on both the NIC's (like certain type of offloading).

link|improve this answer
feedback
up vote 1 down vote

Using bonding in linux this should work flawlessly. And since ESX is pretty much linux-based, i expect it to work as well. Not sure about windows though.

link|improve this answer
Appreciate your answer. Yes, but my SAN server is Windows based. – James Dec 16 '11 at 15:14
Both of them should use IEEE 802.3ad Link Aggregation Control Protocol (LACP) – Mircea Vutcovici Dec 16 '11 at 15:31
3  
ESX is not Linux based. Only the console is based on RHEL or on busybox so that as, mortals, can manage it in a familiar way. – Mircea Vutcovici Dec 16 '11 at 15:39
1  
vmware ESX is Linux-based. Please ignore the nonsense comment above. – adaptr Dec 16 '11 at 15:40
1  
@adaptr please read en.wikipedia.org/wiki/VMware_ESX#Linux_dependencies If ESX was based on Linux they would be forced to release the source code under GPL v2. – Mircea Vutcovici Dec 16 '11 at 15:52
show 3 more comments
feedback
up vote 1 down vote

The problem with that is that teaming or bonding does not increase bandwidth between two stations.

Teaming can increase the overall bandwidth available to a host and - if LACP or the like is supported - on a switch as well, but the consumers of that bandwidth must be multiple stations/hosts/IPs.

The only way to increase bandwidth between host A and host B using multiple network connections is to... use multiple network connections.

You would have to assign each NIC on each end a separate (virtual) IP and route traffic appropriately.

PS. vmware bonding primarily offers physical NIC failover, and connectivity for multiple virtual port groups to the outside in a flexible manner.

Increasing point-to-point bandwidth is not what it does either.

EDITed just in case this wasn't clear: no, connecting 2 systems with multiple NICs in each does NOT increase the bandwidth between them.

link|improve this answer
Thanks. Nut ther main question was if the fact that NICs are from different vendors should affect total performance? – James Dec 16 '11 at 15:48
Since all of that is handled by the ESX (Linux) kernel, there is no reason that it should. – adaptr Dec 16 '11 at 15:49
1  
This isn't entirely correct.. I use 4 NIC's for iSCSI and use round-robin MPIO to balance the load across the nic's. You're right in the way that you can't get a single datastream faster than a single path, but remember that you usually run more than 1 VM on a ESX host. You should edit your answer to reflect this. – pauska Dec 16 '11 at 15:57
pauska: presumably the VM's do not access the storage directly. However using iscsi multipathing (on separate subnets) is a great suggestion. – HampusLi Dec 16 '11 at 16:05
May I wonder why "separate subnets" are necessary? – James Dec 16 '11 at 16:19
show 1 more comment
feedback

Your Answer

 
or
required, but never shown
discard

Not the answer you're looking for? Browse other questions tagged or ask your own question.