OPen-e iSCSI as primary SAN?

All references that I come across are about iSCSI as a secondary storage. I would welcome opinions & experiences about iSCSI as primary storage.

(We are on the verge of making the jump from 3 standalone servers (2 SQL+1 Email) to an Open-e based iSCSI environment. )

Your thoughts?

We use our open-s DSS a server as primary storage for SQL, Exchange and VMware and also a secondary open-e DSS which is replicated from the primary storage.

We have tested using iscsi the speeds where good (140MBs) but the problem with using sql and exchange is that you need to make sure to delay the starting of the services if the servers are rebooted not too difficult by using a batch file with a ping command to provide the timeout value. Also you cannot use dynamic disks with the microsoft iscsi initator as on reboot the drives sometimes do not come back online, so use them as basic disks (know bug with microsoft)

You can use dynamic disks if you use a third party iscsi initiator such as starwind, but agin serices will need to be delayed. Advantages would be that you could extend/mirror dynamic disks.

Ths best solution we found was using fibre channel 2GB as it addresses all the issues and speeds for combined read/write of 270Mbs, we are busy testing the 4GB and so far average speeds of 570Mb/s

Gavin

Thanks for your input!

I noted that you are mirroring the SAN. We would like to mirror the SAN and NAS volumes locally, and also do replication to a DR site.

Any advice in this regard?

I dont think there is support to replicate one dss server to multiple dss servers, which is what you are trying to achieve.

However as a solution you could use software mirroring on the OS to mirror disks between two opn-e servers and then use volume replication from one of these open-e dss servers to a dss server at a remote site.

**but remember that the disks need to be converted to dynamic for windows mirroring, and the issuses with iscsi and dynamic disks as mentioned previously**

Regards

Gavin

In the above case, would not the 'basic/dynamic disk' be a non-issue because we will be replicating volumes at the block level?

When you originally posted you wanted to mirror the san, and replicate to another DR server therefore I suggested the software mirroring with a replication as the solution for you. (Please remeber you can not replicate a volume on an open-e server to more than one open-e server.)

The block level you talk about is replication and does not have auto failover yet so manual intervention would be needed if the primary open-e server failed, where as the mirroring through the OS allows a open-e server to fail without any disruption to the users.

In both of the situations the dynamic / basic disks would still be an issue if you are using a software isci initiator as we found that on occasions when rebooting the application server the disks would appear as offline and we would have to go into disk manager and re-activate them. We did not loose any data but it could be a pain. To resolve this issue we found that using a qlogic isci hba worked perfectly and also improved performance.

Thanks, for a well informed response.

I would like to keep the OS out of the equation, if possible. I prefer to mirror volumes at the iSCSI level. At this time, I am grappling with the following questions:

Can we replicate volumes at the block level, and keep Windows OS out of the equation?

I am thinking of going with 2Gb fibre between switch/SAN, and 1Gb NICs between switch/Server. Could you recommend a switch that will accommodate such a configuration? The other option is to use bonding on the switch/SAN side.

To be able to mirror at block level you would need a HBA that could perform this and I dont know of any that can do this for iscsi or FC. However there may be a qlogic SAN/switch that will perform this for you but I am not sure.

As for a switch that will provide GB ethernet and FC fabric I dont think there is one and would not recommened a common switch.

We have used block level replication between to DSS servers and if one fails we perform a manual failover (until the new release of DSS that will have auto failover). We use the 2 x brocade 16 port FC switchs between our servers and the storage servers in a mesh configuration so that we have redundency on all aspects from FC Cards to FC switch. We then use a GB cisco switch from our workstations to the servers.

Brocade FC switches are expensive, however qlogic do have some entry level FC switchs/SAN