Assigning a device to a specific interface

[peekay]

I have used SanMelody in the past, which allowed a specific "device" (LUNs I think in the case of SanMelody) to be assigned to a specific interface (or NIC). This allowed a form of bandwidth spreading since access to the "shares" was spread over several NICs.

Does StarWind Ent. 4 support this type of functionality? I am trying to connect an ESX 3.5U4 server to a Win2008 (x64) server running StarWind x64 and would like to find the best way of "spreading" the iSCSI traffic over several NICs. I would like to avoid using teaming and MPIO since I am not confident the ESX will support it.

Thanks!

[bboule]

Hello,

The functionality you are describing is not supported in the current StarWind release. With the current release, you can load balance by using MPIO (Round Robin), or you can load balance manually across ESX servers by connecting via different NICs to the storage from different ESX Hosts.

The functionality you are asking about, assigning a device to a specific NIC in the storage box, is currently in development and is slated for a future release of StarWind. In the meantime, please feel free to drop me a line directly if I can provide any additional assistance.

Best Regards,

Bob

Bob Boule
Sr. Solution Engineer
StarWind Software
617-449-7717 x 104
Bob.Boule@StarWindSoftware.com

Skype: rboule
MSN: bobboule@msn.com

[anton (staff)]

You can "bind" multiple instances of StarWind iSCSI target services to the different network interfaces. So say you have NIC1 assigned 192.168.1.1 and NIC2 assigned 10.0.0.1 and TWO StarWindService.exe, one listening to the NIC1 and other listening to the NIC2. From the management console point of view it's going to be two different StarWind target machines (however using one single hardware server).

[peekay]

First, how do you run two Startwind services on the same server? Second and performance wise, are there any benefits to doing this on one server. The server I have has a quad Xeon and 8GB of RAM so horsepower should not be an issue there but what about StarWind itself? Any downside?

I have managed to use two ethernet channels by configuring two ports on the StarWind server to use different subnets and configuring the single iSCSi switch to have two port-based VLANs so I can use failover on ESX. By configuring the ESX "preferred" route on the iSCSI targets, I can balance target traffic between the two iSCSI channels but also having the failover if one channel fails.

Which configuration do you believe is better? Is the LUN-to-port assignment still being considered?

[Constantin (staff)]

You cannnot run two copies of StarWind on 1 box.

[peekay]

I am confused. What was Anton referring to in the earlier post then?

[tulsaconnect]

Yes, I would like to know the answer to this question as well... so.. bump.

[Constantin (staff)]

You can configure ACL per target (choose target and go to Access right tab) or per NIC for this choose server, go to configuration tab and choose Network to choose with what NICs should work StarWind or Management Interface to choose IPs you can use for local/remote management.

[peekay]

Constantin... this approach does not allow true LUN control as the LUNs will still appear on all NICs. However, I will experiment and let you know what I find.
Thanks!

[Constantin (staff)]

Please, keep me updated about your findings.

[peekay]

Constantin,

Unfortunately, I may not get a chance to test since the system is in production. However, using access control does not prevent the shares from being presented on all NICs at the same time. So for this to work on a single iSCSI switch (our configuration), each request from the ESX side would hit all StartWind NICs, have to be access control filtered, then pass on the correct NIC. Does this not impose a large overhead on each iSCSI request? With LUN mapping to a NIC, that LUN only presents itself on the specified NICs, keeping the traffic to that NIC only. Am I missing anything?

In our case, we use the single switch "quasi" failover configuation using port-based VLANs and use "preferred" paths on ESX. This is probably faster than using access control but I would love to hear your take on it.

[Constantin (staff)]

I`ve attached a screenshot where you can choose bind interface for target.

[peekay]

Woah!!!! Why did you not show this in the first place!!!! This is very interesting as it allows you to bind a specific target to a specific IQN/IP and to a specific NIC! Am I missing anything here? This basically means on a single iSCSI switch, I can do away with the quasi-failover on ESX and simply bind each target to a specific ESX IQN/IP and a specific StarWind NIC.

So in theory, here is what I could do:
1. All NICs on same subnet (ESX and StarWind)
2. To spread iSCSI load, bind given targets to different NIC or NICs on StarWind
- HOWEVER, I need to have each UNIQUE target appear only once on the NICs otherwise ESX takes its own path to a given target
3. ESX connects to each LUN as required, with only "assigned" StarWind LUNs appearing on each of the ESX NICs/HBAs.

So if I have two iSCSI HBA channels and one ESX software iSCSI channel then I could spread 5 StarWind LUNS using the access control mechanism over those three ESX channels. All without using VLANs on the iSCSI switch. If some day I change to a 10Gig NIC with NO switch, all I do is change the access control rules to the single NIC and voila!

Any thoughts on this? Did I miss anything?

[Constantin (staff)]

Yes, you are right here
I think this should work, but I don`t if it would be good solution.

[peekay]

Why do you not think it would be a good solution given a single Starwind server?