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

Conference spezko::cluster

Title:+ OpenVMS Clusters - The best clusters in the world! +
Notice:This conference is COMPANY CONFIDENTIAL. See #1.3
Moderator:PROXY::MOORE
Created:Fri Aug 26 1988
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:5320
Total number of notes:23384

5238.0. "HSD10 slows down CLUSTER/NETWORK HSD05 not" by HAMSC3::JPULLER (Jens Puller MCS Hamburg ) Thu Feb 27 1997 08:24

	Hello,
    
    	I would like suggestions to a problem which I posted already 
    	in the HSD05 conference. My original question is below.
    
    	A customer observes very high network load since he installed
    	HSD10 in a cluster. I switched off lock remastering by PE1 = -1
    	There is a quorum disk , but not on the HSD10's. One very 
    	strange observation is  that the high network load keeps
    	happen even after replacing back the older HSD05. The network
    	load will decrease slowly after a week it is back to normal
    	( as it was before using HSD10's ).
    
    	Help is highly appreciated!
    
    	Jens 
    
    
         <<< SSDEVO::DISK$NOTES:[NOTES$LIBRARY]HSD05_PRODUCT.NOTE;1 >>>
                      -< HSD05/HSD10 Product Conference >-
================================================================================
Note 234.0           HSD10 slows down the Cluster;HSD05 not            3 replies
HAMSC3::JPULLER "Jens Puller MCS Hamburg "           60 lines   7-FEB-1997 10:04
--------------------------------------------------------------------------------

        The configuration                         


	Ethern.						Ethern.
	   |	                                            |
         -------       VMS 5.5-2               	         -------
       	| VAX	|      DSSI CLUSTER         	      	| VAX	|
       	| 4000-	|	                		| 4000-	|	
	|  300	|      37 Satellites CAD & Medusa DB 	|  300	|
	|	|     	                        	|	|
         ------                                  	 -------
    BUS0   | | BUS 1                    	   BUS 0   |  |  BUS 1
	   | |____________________________________________ |  |
           |     |      |       |          |          	      |
	   |_____|______|_______|__________|__________________|
                 |      |       |          |      	   |
                 |      |       |          |      	   |
               ------ ------  ------ 	---------	---------
               |RF73| |RF73|  |RF73|  	| HSD10	| 	| HSD10	|
               ______ ______  ______  	|	|	|     	|
			       Q-disk	     	| RZ28	|	| RZ28	|
				     	|	|	|	|
				     	| RZ28	|	| RZ28	|
				     	|	|	|	|
				     	| RZ28	|	| RZ28	|
				     	|	|	|	|
					| RZ28	|	| RZ28	|
					---------	---------

	this Cluster worked quite well with HSD05 installed.
	The Customer change to HSD10 to get the feature of the 
	mini merge.

	Since there are HSD10's installed the sattelite users 
	complain about bad performance.

	The network groups reports  very high load on the Ethernet.
	They use 3COM bridges and hubs. We can consider the network 
	as one segment for this cluster. 
	Indead most of the traffic is SCA protocoll staff ( 6007 )
	Since they run the HSD10 the network is 60% saturared. 
	This is not when using HSD05's.

	Analysing the system shows a very high usage of the interrupt
	stack. I had to tune pool parameters to avoid system buffer 
	shortage. Pool looks fine now but hasn't change the overall
	performance problem. Also the very high interruptstack utilisation
	is still at 60% not allways but very often during that time
	the network load is also very high. From observation  I would 
	say whenever the network is that much loaded the interrupt
	stack shows a usage of more than 80%.

	Any ideas are highly appriciated

	Jens Puller Customer Support Engineer
	MCS in Hamburg Germany

	
    
    
                                           
T.RTitleUserPersonal
Name
DateLines
5238.1VAXLAVC04_U2055 installed?CSC32::S_DANNENLive long and slobberThu Feb 27 1997 13:014
    Make sure that the customer has the VAXLAVC04_U2055 installed. This
    eliminates SCA "babling".
    
    steve
5238.2UTRTSC::thecow.uto.dec.com::JurVanDerBurgChange mode to Panic!Fri Feb 28 1997 06:555
Setting PE1 to -1 only works if the latest locking patches for V5.5-2 are 
applied. V5.5-2 out of the box does not use this parameter.

Jur.

5238.3VAXLAVC04_U2055 was already inHAMSC3::JPULLERJens Puller MCS Hamburg Fri Feb 28 1997 11:1914
    
    	Thanks for the quick reply,
    
    	VAXLAVC04_U2055 had been already installed. We know that the 
    	HSD10 is capable of twice the tranfer rate. Due to this fact we 
    	probably reach the capacity of the 10mbs of the ethernet. On the 
    	other hand the networkers of that customer reported high traffic 
    	during night ( no one is working on the workstations ). 
    	We get for a period of about 1 minute or more high load over 
    	40% to 60% then we have  again quiet time several minutes. 
    	Nearly all traffic is SCA this only if HSD10's are used.  
    
    
    	Jens
5238.4is VAXSHAD09_U2055 ok?HAMSC3::JPULLERJens Puller MCS Hamburg Fri Feb 28 1997 12:303
    .-2
    
    VAXSHAD09_U2055 was installed.  Did I miss one ?