Hp StorageWorks Scalable File Share User Manual Page 78

  • Download
  • Add to my manuals
  • Print
  • Page
    / 84
  • Table of contents
  • BOOKMARKS
  • Rated. / 5. Based on customer reviews
Page view 77
network buffering parameters were set as described in the documentation for the configured
network controller.
A.8.2 Single Stream Throughput
Throughput is limited by the characteristics of the single client. In this particular case, performance
with more than one stripe is mainly limited by the network connection. Figure A-15 shows the
effect of striping on the operation of a single client.
Read performance is adversely affected by striping across OSSs due to contention at the inbound
client port. Several senders are attempting to transmit data at 10 Gb/s each, and the single receiver
can only take data in at 10 Gb/s total. This indicates that the best single stream read performance
is obtained with any particular client accessing a single stripe of a single file (i.e. a single OSS)
at any given time.
Write performance peaks with the client (or client transmission) capabilities when two stripes
are used. Note that the reverse of the read contention issue mentioned above can be inferred.
Several clients attempting to write simultaneously to the same OSS will cause contention at the
inbound port of the OSS (assuming all connections support the same data rate). This write
contention does not appear in single client testing.
Figure A-15 Stripe Count Versus Total Throughput (MB/s)
Multiple clients testing, as well as single client read testing, is complicated by an artifact of the
use of trunking between the blade switches and the 8212zl. The path that a packet takes through
the trunk (i.e. the specific link that the packet traverses) is determined by its source and destination
addresses. This means that every packet from a specific source to a specific destination (on the
other side of the trunk) always travels through a single specific link of the trunk. Therefore, traffic
involving source/destination pairs that route through a particular trunk link will contend for the
bandwidth of that link, not for the aggregate bandwidth of the trunk. The effect can be seen in
the comparison of throughput of one client to that of two clients in Figure A-16. As the number
of clients increases, the traffic is more likely to be spread over the trunk links and utilizes the
aggregate bandwidth of the trunk more effectively.
78 HP SFS G3 Performance
Page view 77
1 2 ... 73 74 75 76 77 78 79 80 81 82 83 84

Comments to this Manuals

No comments