> From: Brad Knowles [mailto:[email protected]] > > I strongly suspect that this performance between ESX clients is bus and > software limited, and that IB, FC, or 10GigE is not going to be able to move > any faster than what you can do over the bus of the machine.
What bus are you talking about? To me, it seems, this whole thing should be ramspeed and cpu limited only. Granted, there's a lot of overhead ... If client 1 wants to read a file, it creates a file open request, which goes to nfs client, which encapsulates in tcp, which goes to the virtual network interface, thinks about mac address, instructs virtual hardware to send a packet, which ESX then thinks about a virtual switch, and passes the packet to some other virtual hardware ... etc etc etc. And finally, client 2 reads the disk, and sends the response packet through virtual hardware. In my test, there was no disk involved. Just a process generating a lot of "a" characters, sending across the virtual network, and another process dumping to /dev/null. End result 2.5 Gbit/sec. Of course, there is another way to give ZFS control of disks for some other machine ... If solaris is the primary OS, and the other machines are VM guests inside solaris. But the only way I'm aware to virtualize windows/linux inside solaris is virtualbox, which is not amazingly attractive. _______________________________________________ Tech mailing list [email protected] https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech This list provided by the League of Professional System Administrators http://lopsa.org/
