Hi,
How good does esx 3.0.2 server handel software iSCSI target what i mean is the process time between CPU's/cores?
Lets say if i team 2 Gbit NIC for use of software iSCSI storage connection.
What degree of CPU usage will just goto process the data transfer? will VMKernel split the process time to all cores; if i have multiple CPU cores or will one CPU0 handel the process of datatransfer that VMKernel use?
Best regards Leif
Message was edited by:
leif.hannusas@kentor.se
I'm beginner with the Starwind software solution (www.rocketdivision.com).
I'll be probably later more able to correctly evaluate the cpu resources...
Hi,
In what way will you evaluate the cpu resources usages; on the esx server or on the windows host that provides starwind software?.
// Leif
Hi Lief,
The amount of CPU utilization will depend on the amount and type of traffic sent to the SAN, and to some degree even on the network configuration. For example, if you are using jumbo frames, the CPU will have fewer packets to create and manage for a given transfer. Further, your server's NICs might do more (or less) in terms of tcp/ip offload than others (most 1Gb NICs are offloading something these days, offloading TCPIP checksums seems to be commmon, if memory serves).
I seem to recall that one CPU/core ends up doing the work, regardless of the number of NICs teamed.
It might be interesting if someone determined the MHz / packet used by the iSCSI initiator. The touhg part would then be trying to figure out exactly how many packets your VMs and applications would use. Very tough to get definitive answers on this.
A reasonable benchmark - it take about 1GHz of CPU power to drive 1 gigabit Ethernet port at line rate (125MB/s). Remember, it takes a considerable number of drives to deliver this line rate if your activity is the small-block transfers typical of most servers (Exchange, SQL, File and Print).
I hope this helps...
Be attentive to the fact that, for the moment, the latest release 3.0.2 is absolute not reliable (:http://www.vmware.com/community/thread.jspa?threadID=98251&tstart=90) to make objective iscsi tests...
Hmmm, seems like that thread outlines a configuration problem on an EqualLogic array, rather than a problem with iSCSI on 3.0.2...