[lustre-discuss] issue: strange pauses between writes, but not everywhere

[Peter Grandi]

>>> On Tue, 28 Oct 2025 20:01:41 +0000, Peter Grandi via
>>> lustre-discuss <lustre-discuss at lists.lustre.org> said:

> * [...] Reading is quite good for all clients. But since
>   yesterday early afternoon inexplicably the clients .13-36
>   have a maximum average write speed of around 35-40MB/s; but
>   if I mount 'temp01' on any of the Lustre servers (and I
>   usually have it mounted on the MDT .40) write rates are as
>   good as before. [...]

Also a whole set of workstations with ULTS 22 + Lustre 2.15 are
not affected. I have been looking at various aspects of LNET and
for example when I do 'lnetctl export' 'tx_min_credits' on some
systems are negative so I guess I can increment them, but on one
of the affected systems I looked at the tx_max_credits was 8 and
the 'tx_min_credits' were all positive.

One curious side node is that I get significantly lower write
rates when with 'lnetctl set discovery 0' than when it is
enabled. Both the Lustre servers and clients I have mentioned
have 4 interfaces but only one has an IP address and is
connected to a switch, so multirail should not have any impact,
as far as I can imagine.