Stimmen - 0, Durchschnittliche Bewertung: 0
(
)
)
|
Dieses Gerät hat auch andere Anweisungen:
Anleitung Zusammenfassung
High Availability disk products for the HP e3000
Page 3 of 4
installed at a remote data center. This feature allows the customer to design a disaster tolerant
solution.
Another feature not yet certified is SMMF (Symmetrix Multiple Mirror Facility). SMMF provides
mirroring of data within the same cabinet but to drives connected to other machines. This is much like
the SRDF feature but internal to the cabinet.
The configuration of the EMC Symmetrix 3000 requires a trained EMC technician. Once configured,
the system administrator can use the familiar VOLUTIL utility to create User Volumes or to add
members to existing volume sets.
Disk Subsystem Performance
Scenario For a Successful Implementation
The performance of the disk subsystem depends greatly on a number factors. Current data shows
that 10-20 physical disk I/Os per second per drive is about the limit before performance starts to
degrade because of the I/O bottleneck. The number of physical devices that the adapter can handle
and the amount of memory installed in the HP3000 can also impact performance. The application's
I/O characteristics, serial vs. random, can greatly affect performance due to the pre-fetching
algorithms used by either MPE/iX or by the Symmetrix 3000 processor. All these factors should be
considered before reorganizing the disk subsystem either by moving data from smaller to larger disks
(JBOD or arrays) or combining LDEVs per physical device (Symmetrix).
Below are two case studies where the first case shows a properly configured disk subsystem where a
customer improved performance and the second case shows an improperly configured disk
subsystem that exposed some problems:
Case1 This customer has an extensive amount of batch processing. The user's
previous configuration was below the 10-20 I/Os per second per device
threshold and they had a good understanding of their application's needs (I/O
resources). They effectively used MPE/iX's User Volumes to segregate major
application data onto different volume sets. They also migrated their data from
4Gb stand-alone disk drives (JBOD) to 4Gb Symmetrix drives, without reducing
the number of spindles or LDEVs.
Analysis This was a successful integration of a new disk technology, like the
EMC Symmetrix 3000. The serial nature of the application lends itself to the
caching algorithms done by the Symmetrix processor. The Symmetrix
processor pre-fetches more data than the MPE/iX memory manager, allowing
for a better hit ratio to the Symmetrix cache. A cache miss causes an increase
in the total time it takes to do an I/O, from the start of the request to its
completion, because the I/O request is satisfied by the slower disk mechanism
instead of the much faster cache hit.
Case2 A site containing a large number of 1Gb and 2Gb disks with 3-4 User
Volumes and a 4 member System Volume Set was consolidated into a
Symmetrix cabinet and reconfigured and reloaded onto one System Volume
Set with disk drives partitioned into 4Gb volumes. This user was at the I/O
threshold of 20 I/Os per second per LDEV before the implementation and was
clearly over it after consolidating the data onto the Symmetrix. This customer
also expected a 20 to 40 percent performance improvement because of the
Symmetrix cache but was very disappointed after collecting performance data
on his application's very random I/O characteristics.
Analysis This new configuration reduces the number of LDEVs and also
reduces the number of pathways to the data. By reducing the number of
pathways to the data, MPE/iX takes longer to fetch all the data that is required
for an operation to take place. MPE/iX tries to asynchronously gather this data
http://www.hp.com/cgi-bin/pf-new.cgi?IN=http://jazz.external.hp.com/papers/ha.html
7/18/2008
...Dieses Handbuch ist für folgende Modelle:
Software - MPE/iX 6.5 Operating System (30.4 kb)
Software - MPE/iX 7.0 Operating System (30.4 kb)
Software - MPE/iX 7.5 Operating System (30.4 kb)