Conference ssdevo::hsj40_product

1211.0. "HSJ50 and preferred path problems." by KERNEL::BROOKER () Sun Feb 16 1997 00:16

	I'm dealing with a Customer who is in the process of commissioning
	two 8400's and four pairs of HSJ50's spilt across two sites and 
	connected by fibre.  He initially rang in claiming very poor shadow
	copy performance between sites and upon further investigation some 
	of the units (all units are 3 member, RZ29B, stripesets) showed as
	follows:-

WSHJ05>sho d2130
MSCP unit                                    Uses
--------------------------------------------------------------

  D2130                                      S2_T130
        Switches:
          RUN                    NOWRITE_PROTECT        READ_CACHE
          WRITEBACK_CACHE
          MAXIMUM_CACHED_TRANSFER_SIZE = 127
        State:
          ONLINE to the other controller
          No exclusive access
          PREFERRED_PATH = THIS_CONTROLLER
        Size: 25134084 blocks

	Where the preferred path was one controller but the unit was online
	to the other.  How is this possible?

	Also in an attempt to analyse the performance I displayed the Cache
	statistics and got the following:-

                           0.0% Idle     3252 KB/S      239 Rq/S
Unit  ASWC  KB/S  Rd% Wr% Cm% HT% PH% MS% Purge BlChd BlHit Unit  ASWC  KB/S  Rd

D2100 o^ b    836   0  44  55  84   0  16  7493  7493  7493
D2110 x^ b      0   0   0   0   0   0   0     0     0     0
D2120 o^ b    633   0   4  95   5   0  95   635   635   635
D2130 o^ b    932   0  44  55  82   0  18  8352  8352  8382
D2140 o^ b    850   0  44  55  84   0  16  7620  7620  7620
D2150 a^ b      0   0   0   0   0   0   0     0     0     0
D2250 a^ b      0   0   0   0   0   0   0     0     0     0
D2350 o^ b      0   0   0   0   0   0   0     0     0     0
D2400 a^ b      0   0   0   0   0   0   0     0     0     0
D2410 a^ b      0   0   0   0   0   0   0     0     0     0


	What is the relevance of the Cm% stat? and what initiates these
	compares?  Am I also to understand that, in the case of D2100 for
	example, 44% of the Writes are cached and in which case why is the
	hit rate 84%.


	It's been a long night so just one more quick question, the Customer 
	says he is using HSJ assist but I don't know what this is or how it
	works.  From his explanation and whilst trying not to appear too dumb,
	I gather it's a kind of script driven "clone" utiltity although the
	Customer says if cloning is not possible it will use VMS backup instead. 
 	So my question is, how can you tell which method is being employed?

	Any help gratefully accepted, but for the moment I'm off to bed.

	Cheers. 

    There seem to be 2 aspects to this problem:-

       Units  which  are  preferred  to  one  controller  APPEAR  to  be 
       accessed through the other controller (and  the  customer  claims 
       that  some  event  [as  yet unclarified] has caused this failover 
       without controller failure!!)

       VMS hosts not directly connected to the storage  controllers  (in 
       this  case,  hosts  at  the  remote  end  of a FDDI ring in a BRS 
       Cluster) cannot take advantage of HSJ assist (DCD Function)  such 
       that  mounting  a  drive into an existing shadowset causes a full 
       copy when an assisted copy would  have  been  perfectly  feasible 
       (and  planned  for!!).  In short, the host that actually GETS the 
       Copier lock doesn't have direct access to the storage  and  can't 
       `see' the DCD capability.

    Any comments??

    I heard via the grapevine about some MSCP drivers that are not
    available as a patch in Tima.....? Specifically for FDDI BRS clusters
    ...and MSCP serve in these BRS clusters....
    Chris, or anyone, could you please comment?
    
    Regards Ivan