Cornelia - thank you for your reply.

I am working from the Redbook: Virtualization Cookbook ... Volume 1 and
am finding typos in it. In this case the cio_ignore command was botched
up and included in the description before a set of steps to configure
hipersockets on RH. it should have been step number 1.

So, I can report partial success, i now have hipersocket network up and
the guests can ping each other.

But it is deathly slow - nothing hiper about it.

Here are the tuning 'knobs' that i am aware of:

[1] MTU size
my LAN is set up with 64K:
DEFINE LAN ZVTSMLAN OWNERID SYSTEM TYPE HIPER IP MFS 64K

and the guest has a NICDEF to match:
  NICDEF 0204 TYPE HIPER DEVICES 3 LAN SYSTEM ZVTSMLAN

oddly, when i look at the network connections on linux, it shows the
MTU as 57344 (?):

# ip a
.
.
.
3: enccw0.0.0204: <BROADCAST,MULTICAST,NOARP,UP,LOWER_UP> mtu 57344
qdisc pfifo_fast state UP group default qlen 1000
    link/ether 02:00:00:00:00:5f brd ff:ff:ff:ff:ff:ff
    inet 192.0.1.32/24 brd 192.0.1.255 scope global noprefixroute
enccw0.0.0204
       valid_lft forever preferred_lft forever


[2] QIOASSIST setting
i have QIOASSIST ON system wide and have set in on specifically for the
two guests i am testing with - Z4 is a VSE guest, LNXADMIN is ClefOS
7.4 guest:

q qioassist z4
ALL USERS SET - ON
 
USER      SETTING   STATUS
Z4        ON        ACTIVE
Ready; T=0.01/0.01 09:37:01
q qioassist lnxadmin
ALL USERS SET - ON
 
USER      SETTING   STATUS
LNXADMIN  ON        ACTIVE


[3] networking buffers
should i mess with changing buffers on ClefOS ?  
(i read somewhere that the default was change from 16 to 128)

# lsqeth enccw0.0.0204
Device name                      : enccw0.0.0204
---------------------------------------------------------------------
        card_type                : GuestLAN: SYSTEM ZVTSMLAN (Type:
HIPERS)
        cdev0                    : 0.0.0204
        cdev1                    : 0.0.0205
        cdev2                    : 0.0.0206
        chpid                    : 01
        online                   : 1
        portname                 : no portname required
        portno                   : 0
        route4                   : no
        route6                   : no
        state                    : UP (LAN ONLINE)
        priority_queueing        : always queue 2
        fake_broadcast           : 0
        buffer_count             : 128
        layer2                   : 0
        isolation                : none
        sniffer                  : 0


Are there other tuning knobs? 
How can I figure out why this connection is slow?

-- 
Phillip Gramly
Systems Programmer
Communications Data Group
Champaign, Illinois


On Mon, 2018-07-16 at 03:27 -0500, Cornelia Huck wrote:
> I'm seconding the suggestion of using chzdev, but some additional
> notes
> from me:
----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390
----------------------------------------------------------------------
For more information on Linux on System z, visit
http://wiki.linuxvm.org/

Reply via email to