RVP a écrit :
> On Thu, 4 Feb 2021, Michael van Elst wrote:
>
>> r...@sdf.org (RVP) writes:
>>
>>> Look at the `enabled' and `ec_enabled' capabilities. Most of them are
>>> off. Not sure if they make much of a difference...
>>
>> The capabilities that aren't 'enabled' by default aren't tested that
>> much.
>> In my experience they either help somewhat with CPU usage or are broken,
>> but they rarely have an effect on network performance.
>
> Thanks. However, I had expected JUMBO_MTU to be switched on when
> MTU was set to >1500.
Maybe, but I think we can reduce the problem to iscsid (and of course
iscsid network access).
Indeed, when I connect a linux workstation (mtu 1500) to NAS through
this NetBSD server, I obtain more than 900 Mbps between iSCSI Linux
initiator and NAS. If NetBSD network stack was responsible of this low
troughput, I think there is no reason that when NetBSD acts as a simple
router, Network throughtput was greater than when it acts as iSCSI
initiator.
I have seen also that NetBSD iSCSI throughput was always about 10 Mbps.
dd uses cache, thus we thought that throughput was greater for small
transferts, but it is not the cas. Real throughput (on network)
_doesn't_ exceed 10 MBps between NetBSD iSCSI initiator and target.
JKB
