White Papers

Dell HPC NFS Storage Solution - High Availability (NSS6.0-HA) Configuration with Dell PowerEdge 13
th
Generation Servers
16
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 IPoIB 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 clients and the file system. Component failures were introduced on
the server while the clients were actively reading and writing data from/to the file system.
In all scenarios, the client processes completed the read and write operations successfully. As
expected, the client processes take longer to complete if the process was actively accessing data
during a failover event. During the failover period, when the data share is temporarily unavailable,
the client processes were in an uninterruptible sleep state.
Depending on the characteristics of the client processes, they can be expected to either 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 after the failover is completed.
For read and write operations during the failover case, data correctness was successfully verified
using the checkstream utility.
The information about IOzone, checkstream, and dd can be found in Appendix A.
5. NSS6.0-HA I/O Performance
This section presents the results of the I/O performance tests for the current NSS-HA solution. All
performance tests were conducted in a failure-free scenario to measure the maximum capability of the
solution. The tests focused on two types of I/O patterns: large sequential reads and writes, and small
random reads and writes.
A 480TB configuration was benchmarked with IPoIB cluster network connectivity. The 64-node compute
cluster described in section 4.2 was used to generate workload for the benchmarking tests. Each test
was run over a range of clients to test the scalability of the solution.
The IOzone tool was used in this study. IOzone was used for the sequential and random tests. For
sequential tests, a request size of 1024KiB was used. The total amount of data transferred was 256GiB
to ensure that the NFS server cache was saturated. Random tests used a 4KiB request size and each
client read and wrote a 4GiB file. Refer to Appendix A for the complete commands used in the tests.
5.1. IPoIB sequential writes and reads
In the sequential write and read tests, the I/O access patterns are N-to-N, i.e., each client reads and
writes to its own file. Iozone was executed in clustered mode and one thread was launched on each
compute node. As the total transferred data size was kept constant at 256 GiB, the file size per client
varied accordingly for each test case. For example, 256 GiB file was read or written in 1-client test
case, 128 GiB file was read or written per client node in 2-client test case.
Figure 3 shows the sequential write and read performance. The figure shows the aggregate throughput
that can be achieved when a number of clients are simultaneously writing or reading from the storage
over the InfiniBand fabric.