From: George S. <geo...@gm...> - 2011-05-27 21:02:33
|
On 27.05.2011 23:10, Ross S. W. Walker wrote: > Lars Ellenberg [mailto:lar...@li...] wrote: >> Also, since the targets don't know of each other, they would happily >> give out concurrent reservations each. > Let me expand on how bad this is by saying if this volume were a MSCS > or a VMware VMFS or a GFS or a OCFS volume it would be quickly > corrupted beyond repair. > > Also, you can have double, triple, quad the throughput on a single > head, with double, tripe or quad NICs each on a separate subnet, > and the initiator making a separate session on each subnet and > multipathing set to round-robin. > Thank you. But, still, can you show examples when this configuration will damage data? I actually expects that iscsi simply acts like 'read, write' layer over block device and (with blockio) is like a simple stateless proxy to disk, Which kind of activity is 'statefull' for iscsi? |