White Papers

Dell HPC NFS Storage Solution High Availability Configurations with Large Capacities
26
Metadata tests were performed using the mdtest benchmark and include file stat, create and delete
operations.
While these benchmarks do not cover every I/O pattern, they help characterize the I/O performance of
the NSS-HA solution.
As mentioned in Section 5.3 bullet (12), performance was evaluated using NFSv3 as well as NFSv4. In
the following section performance results are presented on NFSv3 and a comparison of NFSv3 and
NFSv4 is presented section 6.5.
6.1. InfiniBand sequential writes and reads
This section presents the results of sequential IO tests over the InfiniBand on the 144TB and 288TB
configurations. The IP-over-IB driver supports two modes of operation – reliable connected mode and
unreliable datagram mode. The two modes differ mainly in packets delivery reliability and in the size
of the MTU. This can be changed at runtime by managing the contents of the
/sys/class/net/ib0/mode file. Results in this section are using the default connected mode. As a
follow on to this document, a blog post on www.hpcatdell.com will discuss the performance of the
datagram mode.
From Figure 8 it is seen that the peak throughput for sequential writes is ~1500MB/s. The write
throughput is constrained by the NSS-HA design choices of RAID controller write cache mirroring and
the NFS sync export option. Thus the performance of the 144TB and 288TB configurations is similar.
InfiniBand large sequential write performance Figure 8.
0
200
400
600
800
1000
1200
1400
1600
1 2 4 8 16 32 48 64
Throughput in MB/s
Number of concurrent clients
InfiniBand large sequential write performance
144 TB -- NFSv3 288 TB -- NFSv3