NetBSD-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: Network very very slow... was iSCSI and jumbo frames



RVP a écrit :
> On Fri, 5 Feb 2021, BERTRAND Joël wrote:
> 
>> But even if dd is done over block device, throughput is correct
>> without network.
>>
> 
> What's the network throughput like when a client machine (eg.
> hilbert) reads/writes data on this exported sd1 target over agr0?

	I haven't tried with sd1. But this NetBSD server exports swap
partitions for several disless workstations. For example, /dev/rdk1 as a
100 GB partition on ccd0 exported by istgt to hilbert :

legendre# ccdconfig -g
ccd0            32      0x0     2000408739840   /dev/wd0a /dev/wd1a

istgt config contains :

[Global]
  Comment "Global section"
  NodeBase "iqn.2020-02.fr.systella.legendre.istgt"
  PidFile /var/run/istgt.pid
  AuthFile /usr/pkg/etc/istgt/auth.conf
  MediaDirectory /var/istgt
  LogFacility "local7"
  Timeout 30
  NopInInterval 20
  DiscoveryAuthMethod Auto
  MaxSessions 256
  MaxConnections 64
  MaxR2T 256
  MaxOutstandingR2T 16
  DefaultTime2Wait 2
  DefaultTime2Retain 60
  FirstBurstLength 262144
  MaxBurstLength 1048576
  MaxRecvDataSegmentLength 262144
  InitialR2T Yes
  ImmediateData Yes
  DataPDUInOrder Yes
  DataSequenceInOrder Yes
  ErrorRecoveryLevel 0

[UnitControl]
  Comment "Internal Logical Unit Controller"
  AuthMethod CHAP Mutual
  AuthGroup AuthGroup10000
  Portal UC1 127.0.0.1:3261
  Netmask 127.0.0.1

[PortalGroup1]
  Comment "legendre.systella.fr/agr0"
  Portal DA1 192.168.10.128:3260

[InitiatorGroup1]
  Comment "hilbert.systella.fr"
  InitiatorName "ALL"
  Netmask 192.168.10.103/32

[LogicalUnit1]
  Comment "swap hilbert"
  TargetName hilbert
  TargetAlias "swap hilbert"
  Mapping PortalGroup1 InitiatorGroup1
  AuthMethod Auto
  AuthGroup AuthGroup1
  UseDigest Auto
  UnitType Disk
  LUN0 Storage /dev/rdk1 Auto

	Hilbert runs with 32 GB and a Intel(R) Core(TM) i9-10900F. Operating
system is Linux/Debian (kernel 5.10.0-2-amd64). Legendre runs with
9.1_STABLE (CUSTOM). It runs with a customized kernel as it requires
ALTQ support. No other special option in kernel configuration.

	Last evening, I have rebuild FreeCAD and whole KiCAD with make -j20
install. As seamonkey was running on this workstation (and eats a lot of
memory), system was swapping and I have seen that throughput from and to
rdk1 can raise up to 800 Mbps. Maybe a little more, and legendre's CPU
was not overleaded. Please note that there are in the same time three or
four other diskless workstations (FreeBSD, Linux and another NetBSD) on
the same network that use legendre as boot and nfs server.

> Can you do the read/write NAS dd test on legendre with all
> non-essential daemons (bacula, nfs, squid, postgres, ...) except
> iscsid turned off?

	I have done some time ago, same result. I can open a ssh access if you
want to see legendre configuration.

	Best regards,

	JKB


Home | Main Index | Thread Index | Old Index