Question about Best Practice Active-Active

[nova17.de]

Hello,

at the moment I setup my first V7 Active/Active iSCSI cluster.
Here is how I did it... What do you think? Is this best practice? Any errors?


NODE1
2 iSCSI Targets, each with 2 Logical Volumes
Target 1: vmfs01 (2x 600 GB Volume)
Target 2: vmfs02 (2x 2000 GB Volume)

NODE2
2 iSCSI Targets, each with 1 Logical Volume
Target 3: server01 (1x 2500 GB Volume)
Target 4: server02 (1x 1500 GB Volume)

Each with 4 Virtual IP addresses and 3x balance-rr bond for the Storage Replication.


I connect iSCSI Target 1 and 2 to my ESXi Servers and I connect Target 3 and Target 4 to physical Windows 2008 Servers.


Why this way ????

VMware: We only setup 600 GB VMFS Datastores but in this location we have one big fileserver. We don't need much performance, that's the reason why a VM is ok.
iSCSI Target 1 with the 2x 600 GB is the regular VMFS datastore for all Virtual Machines.
iSCSI Target 2 with the 2x 2000 GB is VMFS datastore, too - contain only the data disk of the virtual fileserver.
Benefit: If I have to add more diskspace to the fileserver I can remove iSCSI Target 2 from the cluster without the need to shutdown all Virtual Machines.

Windows Servers: Each server have his own iSCSI Target with only one LUN. So Server 1 never can access the disk of Server 2 by mistake and vice versa.
Second Benefit: If I have to add more diskspace to Server 1 I can keep Server 2 running.



Everythink ok?
Any comments?
Best Practice?



Thanks. :-)

Bye Manuel

[To-M]

I would look to use MPIO for the VIP NICs and not use Bonding. MPIO will give you multiple streams for the I/O's where bonding will give you only 1 stream and you still get path failover with MPIO.


You can directly connect the single designated NIC for the Volume Replication to save on the switch port.

For the VMware I would set the iSCSI Target values that I have listed below and also set in VMware's initiator settings as well. Do this before connecting your VM's. For Windows the defaults
are ok.

1. From the console, press CTRL+ALT+W,
2. Select Tuning options -> iSCSI daemon options -> Target options,
3. Select the target in question,


maxRecvDataSegmentLen=262144
MaxBurstLength= 1048576
Maxxmitdatasegment=262144
FirstBurstLength=65536
DataDigest=None
maxoutstandingr2t=8
InitialR2T=No
ImmediateData=Yes
headerDigest=None
(deprecated)Wthreads=8

[alex1]

nova17.de,

true storage fail over is best achieved with VMWare Storage DRS, the Open-E servers are just big boxes.

the ideal method, is to create identical iscsi luns on each server, so both "nodes" have identical volumes & replicated tasks for active/active.

[nova17.de]

Hello Todd,
I use MPIO for the VIP NICs - I only use a balance-rr Bond for the Storage Replication NICs:

4 NICs with VIPs presenting iSCSI -- VMware/Win2008 connected with MPIO
3 NICs as balance-rr bond for Storage Replication -- all cables connected directly 1:1 without a Switch inbetween.


About your iSCSI Target Options: At the moment we're running ESX4 but we're planning the upgrade to V5.x. Are these settings optimized for VMware in general?



Hello Alex,
yes I use the Open-E only as a big box, same iSCSI configuration on each side.



Thank you.
Bye, Manuel

[To-M]

I would still set these settings for the ESX 4.x as well.

[nova17.de]

Hi Todd,

I changed the values. Can you explain in in 2-3 short words what these settings do?
Or how much this will improve the performance?

Thank you
Manuel