Re: [Xen-devel] [RFC PATCH] xen-block: introduces extra request to pass-through SCSI commands

[Ian Jackson <Ian.Jackson@xxxxxxxxxxxxx>]

Bob Liu writes ("[RFC PATCH] xen-block: introduces extra request to 
pass-through SCSI commands"):
> 4) Alternative approach: Using PVSCSI instead:
>  * Doubt PVSCSI can support as many type of backend storage devices
>    as Xen-block.

I don't understand why this would be the case.  Your "extra requests"
are SCSI CDBs, so your proposal only works with SCSI targets.

>  * Much longer path:
>    ioctl() -> SCSI upper layer -> Middle layer -> PVSCSI-frontend -> 
> PVSCSI-backend -> Target framework(LIO?) ->

The pvscsi pieces here is precisely the multiplexer that you really
want, for this all to work correctly.

If you want your guests to be able to drive this thing properly, you
need to make the thing look like a SCSI target anyway.  So effectively
you've reinvented PVSCSI.

When you say "longer path" do you really mean that the performance is
poor ?  I don't think this is due to the number of layers here.  The
layers are quite thin.

If there is a performance problem with PVSCSI then perhaps that should
be investigated.

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel