White Papers

Dell HPC NFS Storage Solution High Availability Configurations with Large Capacities
25
Impact to clients
Clients mount the NFS file system exported by the server using the HA service IP. This IP is
associated with either an InfiniBand or a 10 Gigabit Ethernet network interface on the NFS server.
To measure any impact on the client, the dd utility and the iozone benchmark were used to read
and write large files between the client and the file system. Component failures were introduced
on the server while the client was actively reading and writing data from the file system.
In all scenarios it was observed that the client processes complete the read and write operations
successfully. As expected, the client processes take longer to complete if the process is actively
accessing data during a failover event. During the failover period when the data share is
temporarily unavailable, the client process was observed to be in an uninterruptible sleep state.
Depending on the characteristics of the client process it can be expected to abort or sleep while
the NFS share is temporarily unavailable during the failover process. Any data that has already
been written to the file system will be available. The cluster configuration includes several design
choices to protect data during a failover scenario as describe in Section 5.3.
For read and write operations during the failover case, data correctness was successfully verified
using the checkstream utility.
Details on the tools used are provided in Appendix B: Benchmarks and test tools.
Performance tests 6.
This section presents the results of the performance related tests conducted on the NSS-HA solution.
The method used for performance analysis is similar to previous versions of the solution and is
described in detail here. The larger capacities and updated performance characteristics of this version
of the solution are some of the big differentiators for this release of the solution.
All performance tests were done in a failure free scenario to measure the maximum capability of the
solution. Analysis focused on three types of IO patterns: large sequential reads and writes, small
random reads and writes, and metadata operation related tests.
Both the 10 Gigabit Ethernet and the IP-over-InfiniBand (IPoIB) cases were benchmarked. The 64-node
compute cluster described in Section 5.2 was used to provide IO load to the NSS-HA solution. Each test
was run over a range of clients to test the scalability of the solution.
Both the 144TB and 288TB configurations were benchmarked. The 288TB solution provides double the
capacity and utilizes twice the disk spindles as the 144TB solution as describes in Section 4.1, Table 2.
Recall from Section 4.2 that a 288TB configuration has two 144TB portions concatenated together. For
tests on the 288TB configuration, care was taken to exercise both parts of the file system by ensuring
multiple concurrent client streams access different parts of the file system.
The iozone and mdtest utilities were used in this study. Details on the benchmarks are provided in
Appendix B: Benchmarks and test tools.
Iozone was used for the sequential and random tests. For sequential tests, a request size of 1024KB was
used. The total amount of data transferred was 128GB to ensure that the NFS server cache was
saturated. Random tests used a 4KB request size and each client read and wrote a 2GB file.