Re: Observations on LG_NAS, mvgbe0 issue?

[Robert Swindells <rjs%fdy2.co.uk@localhost>]

Reinoud Zandijk wrote:
>On Tue, Jul 02, 2013 at 11:49:43AM +0100, Robert Swindells wrote:
>> Reinoud Zandijk <reinoud%NetBSD.org@localhost> wrote:
>> >If I test it from another machine and read the same set from NFS i get
>> >quite different values! The rate drops to 11 Mb/sec and the system
>> >time is around 30% but the interrupt time is 60-70%!
>> 
>> How fast is the NFS server ?
>
>>Well this NAS is also my NFS server :) the client is an i7 amd machine
>>so thats not really a factor to account for in this.

Sorry, I read it that you were writing to the LG_NAS not reading from it.

>> >Am I right to conclude that the Marvell Gbit ethernet controller
>> >mvgbe0 is NOT using DMA? or is it using it but also crappy in design?
>> 
>> Not really, the ethernet controller holds pointers to buffers in memory
>> and raises an interrupt either when they have all been filled or when a
>> tuneable timer expires.
>> 
>> I don't know if we have done very much testing on how many packets it
>> can batch together in one interrupt.
>
>I doubt its batching at all creating so many interrupts and consuming
>so much interrupt time. I also tested the machine on a 100 mbit
>network but the result is the same as the network is quite quiet
>normally other than some spurious http and mail work next to some mild
>NFS.

What are the current values for the mvgbe sysctls ?

Robert Swindells