Home > Workload Solutions > High Performance Computing > White Papers > HPC High-Performance Storage Solution for BeeGFS > Parameters
This section shows the tuning parameters that we configured on the BeeGFS testbed system in the Dell Technologies HPC and AI Innovation Lab.
for mdev in /dev/mapper/storage*; do
dev=$(basename $(readlink -f "$mdev"))
echo "$dev"
echo deadline > /sys/block/${dev}/queue/scheduler
echo 2048 > /sys/block/${dev}/queue/nr_requests
echo 4096 > /sys/block/${dev}/queue/read_ahead_kb
echo 256 > /sys/block/${dev}/queue/max_sectors_kb
done
$ chmod +x /etc/rc.local
• Tuned the IO scheduler settings for the metadata block devices on the metadata servers by adding the following lines to /etc/rc.local and make /etc/rc.local executable afterwards:
for mdev in /dev/mapper/storage*; do
dev=$(basename $(readlink -f "$mdev"))
echo "$dev"
echo deadline > /sys/block/${dev}/queue/scheduler
echo 128 > /sys/block/${dev}/queue/nr_requests
echo 128 > /sys/block/${dev}/queue/read_ahead_kb
echo 256 > /sys/block/${dev}/queue/max_sectors_kb
done
$ chmod +x /etc/rc.local
# cat /etc/tmpfiles.d/90-beegfs-hugepages.conf
# Recommended configuration for BeeGFS servers
# Disable transparent hugepages
# Type Path Mode UID GID Age Argument
w /sys/kernel/mm/transparent_hugepage/khugepaged/defrag - - - - 0
w /sys/kernel/mm/transparent_hugepage/defrag - - - - never
w /sys/kernel/mm/transparent_hugepage/enabled - - - - never
# VM ratios recommended for BeeGFS
vm.dirty_background_ratio = 5
vm.dirty_ratio = 20
vm.min_free_kbytes = 262144
vm.vfs_cache_pressure = 50
beegfs-meta.conf
connMaxInternodeNum = 64
tuneNumWorkers = 12
tuneUsePerUserMsgQueues = true # Optional
tuneTargetChooser = roundrobin (benchmarking)
beegfs-storage.conf
connMaxInternodeNum = 24
tuneNumWorkers = 12
tuneUsePerTargetWorkers = true
tuneUsePerUserMsgQueues = true # Optional
tuneBindToNumaZone =
tuneFileReadAheadSize = 0m
beegfs-client.conf
connMaxInternodeNum = 24
connRDMABufNumber = 22
connRDMABufSize = 32768
Note: The tuneTargetChooser parameter was set to roundrobin for the purpose of benchmarking so that the targets are chosen in a deterministic, round-robin fashion. However, in a production system, it is recommended to use the “randomized” algorithm which chooses the targets in a random fashion.