[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference decwet::winnt-clusters

Title:WinNT-Clusters
Notice:Info directories moved to DECWET::SHARE1$:[NT_CLSTR]
Moderator:DECWET::CAPPELLOF
Created:Thu Oct 19 1995
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:863
Total number of notes:3478

729.0. "Poor backup performance" by NETRIX::"[email protected]" (Tsvika Lavie) Sun Apr 06 1997 04:06

Hello,

I've experienced a slow Networker Backup performance when I try to backup
a server that is part of the NT Cluster to the Second Cluster machine that is
the backup server (Holding the Tape device).

I'm using a fast Ethernet line and the backup uses the Nonprivate line and
TCP/IP protocol.

The Backup speed I've seen is about 70 kb/s (very very slow).

Backup from other NT Clients are fairly good (ten times faster).

I'm using Networker Save & Restore v4.3.
NT 3.51.
NT Cluster 1.0 SP1 + Hotfixes.

Does any One using similar Environment and has a different results ????

Regards,

Tsvika Lavie

[Posted by WWW Notes gateway]
T.RTitleUserPersonal
Name		DateLines
729.1low tcp/ip performance on clusterTAVEIS::SVIMon Apr 07 1997 05:4714
    More information:
    
    We tested the ftp performance between the two cluster nodes and found
    that the performance is poor, 8 Kbyte/sec. While, the ftp performance
    between each node to another Alpha/NT node that is not part of the
    cluster configuration is 800 Kbytes/sec.
    
    It seems to be a tcp/ip performance problem between the two cluster
    members. Any ideas ?
    
    Thanks,
    
    Tsvika
729.2check bindings and driverDECWET::LEESWill, NTSG DECwest, SeattleMon Apr 07 1997 15:2211
Check the network bindings in case their order may be relevent here.

Second, disable the CFS driver.  It is our cluster name redirector.  It may be 
causing additional overhead in opening connections.

You have eliminated the machines' hardware as the cause by itself by 
deinstalling clusters and measuring the performance?

Let us know what you find out.

Will
729.3Binding problems !!!???NETRIX::"[email protected]"Tsvika LavieTue Apr 08 1997 03:4126
HI Will

> Check the network bindings in case their order may be relevent here.

You are right the problem looks like a binding problem,

The original set up was as follow:
I manually set up the network binding in order  to enable TCP/IP only over the
external Network card (the one that connected to the LAN),and
enable binding for NETBUEI only over the privet Network (between the two
Cluster member)

To test the binding issue I changed the binding as follow:

I Enabled the TCP/IP binding over the privet Network and made the 
the Backup Server to use the privet line to get to the second member,
by using the HOSTS file.

now the TCP/IP performance looks as expected.

Do you have any explanation for this behavior,

Thanks 
Tsvika Lavie

[Posted by WWW Notes gateway]
729.4Also check IP routing?DECWET::CAPPELLOFMy other brain is a polymerTue Apr 08 1997 20:4521
> manually set up the network binding in order  to enable TCP/IP only over the
>external Network card (the one that connected to the LAN),and
>enable binding for NETBUEI only over the privet Network (between the two
>Cluster member)
    
    That should have worked fine.  I normally run my office cluster that
    way.  (TCP/IP on the main network card, NETBEUI on the "private" card.)
    
    I get about 210 KB/sec on a slow ethernet link using FTP.  Same results
    whether I'm talking to a cluster server or a non-cluster machine.
     
    On ServerA, you might want to try
    
    	tracert ServerB
    
    to see if your messages are taking some unecessary hops to get from one
    machine to the other.
    
    By the way, the previous suggestion to look at CFS won't help here. 
    The Cluster client code doesn't get involved with TCP/IP socket
    connections used by FTP or NetWorker.