kristof wrote: > Hi Mike thanks for the link. > > So if I want to start using comstar, I will have to create a few > hundred/thousands target groups (depending on the max number of targets I > would like to support). If I don't do so and I need a few extra groups later > on I will have to disconnect all my iscsi connected clients. > > This looks a little strange for me, since it's very hard to say how many > targets I will need. >
Based on Sumit's previous statements it sounds like the original assumption that went into the design was that the set of target ports would be relatively static. Certainly from an FC perspective it will be -- each FC target port represents a physical FC port so the set of ports will not change often and when it does the system will likely be coming down anyway for hardware reconfiguration. iSCSI creates a slightly different situation since targets can be dynamically created and destroyed via itadm. From that standpoint I think there is a legitimate argument for a capability to modify target groups online (and indeed there is already an RFE for that feature). But... I feel like we still have a disconnect here because this doesn't seem to me like a significant limitation and others clearly believe this is a serious restriction. So I'd like to understand the use cases better. The idea of provisioning targets doesn't really fit into the intended Comstar iSCSI usage model. A target is just an access point, completely independent of any backing store. Even though targets can be created at will via itadm, I would expect an administrator to create targets to represent specific aspects of their server configuration. For example I typically create one target per NIC interface and then bind a different NIC to each target using target portal groups. This is certainly not the only legitimate mapping -- it's up to the administrator to decide what makes sense -- but the underlying expectation is that the system configuration drives the set of iSCSI target ports. From this standpoint an iSCSI target port is very much like an FC target port and therefore it doesn't seem like frequent target group changes would be required. So with that in mind and realizing that I don't have insight into the way you are using target ports, can you tell me more about how target ports map into your environment? In other words, as you add virtual machines to your cloud environment what drives the creation of additional target ports? -Peter _______________________________________________ storage-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/storage-discuss
