Chapter 2 HPSS Planning
136 September 2002 HPSS Installation Guide
Release 4.5, Revision 2
2.11.1 DCE
Due to several problems observed by HPSS, it is highly recommended that all DCE client
implementationsuse the RPC_SUPPORTED_PROTSEQS=ncadg_ip_udp environment variable.
Frequent timeouts may be observed if this is not done. Each HPSS system should be periodically
checked for invalid/obsolete endpoints. Failure to comply may cause miscellaneous failures to
occur as well as significantly decreasing HPSS performance.
DCE will attempt to utilize every available network interface when attempting RPC
communication. This can cause performance problems when using multi-homed machines. It is
highly recommended that all DCE client and server machines use the
RPC_UNSUPPORTED_NETIFS environment variable to specify those interfaces that shouldnot
be used for RPC communications.
A Global Directory Access Daemon (GDAD) is required in all cooperative DCE Cells. Thegdad
uses information from the Domain Name Service, DNS, to resolve communication requirements
betweenDCE Cells. In DCE 3.1, the dced maintains the executing state of the local gdad. Boththe
AIXand SOLARIS DCE implementations of gdad provide the ability to initiate the gdad with a “–
r filename” flag allowing the specification of a specific “resolv.conf” file to be used bygdad.
2.11.2 Encina
Becauseall HPSS transactions update metadata information, it is essential that SFS be optimized to
access the database in an efficient manner. There are a number of items that can greatly affect
performance of the SFS server and its access to the metadata. The following is a list of these that
should be analyzed:
Number of SFS Threads
Buffer Pool Size
SMP Node/Machine
HPSS/SFS Server Proximity
Storage Media
2.11.2.1 SFS Server Parameters
Numberof SFS Threads. The SFS server allows the administrator to change the number of threads
theserver uses to carry out user requests. For those using a script like /etc/rc.encina to start the SFS
manually, the command line should include an argument to override the default value of 10
processing threads. A more reasonable setting for SFS in an HPSS environment is 200 to 350
processingthreads. The Thread Pool Size can be changed by adding -P 200:2 to the arguments in
/etc/rc.encina. The server should be recycled to start using the new setting. The increase in the
number of processing threads will allow HPSS to carry out more simultaneous operations. The
default of 10 threads will cause HPSS to serialize many of its transaction operations which will
decrease performance.
BufferPool Size. SFS allows the administrator to define how much memory the server is allowed
touse to carry out its operations. The default value is 1000 1KB pages of memory which is too small