The overhead is real, but its cost is not clear. The difference between the performance of direct FCP -vs- EDEV has not been fully measured (or at least not widely reported.) I'm still looking for numbers.
Which is faster? For a guest to perform QDIO or for CP to handle it? Traditionally, we have largely agreed that VM (CP) should handle the heavy lifting. Taking off my virtualization purist hat, I say let the guests do DIAG 250 and let CP do the real input/output. -- R; <>< On Tue, Dec 16, 2008 at 11:42 AM, Kris Buelens <[email protected]> wrote: > But, using simulated FBA minidisks incur higher CP overheads that > letting Linux use the FCP channels. > > 2008/12/16 Alan Altmark <[email protected]> >> >> On Tuesday, 12/16/2008 at 11:07 EST, Lee Stewart >> <[email protected]> wrote: >> > I speak EDEVs, but for various reasons EDEV is not a choice for this >> > customer. They only have SAN disks - raw LUNs. (The VM system itself >> > is installed on the few ECKD volumes -- all the Linuxen are on SAN >> only.) >> >> Sorry, Lee; I misread your original post where you asked how RACF >> *protects* FCP disks. It doesn't. CP doesn't allow ESMs to control >> ATTACH/DEDICATE. (It's on the to-do list.) >> >> If they want RACF control over what guests can see the LUN, then they have >> to use EDEVs, since that's the only way to get minidisks, where the ESMs >> hold sway. >> >> Alan Altmark >> z/VM Development >> IBM Endicott > > > > -- > Kris Buelens, > IBM Belgium, VM customer support >
