The dCache SRM server requires a PostgreSQL database for request persistency and link group space accounting. It is recommended to run SRM on a dedicated node, preferably with a PostgreSQL database running on a separate host. The SRM server has to have sufficient memory to hold request queues and CPU resources mainly for GSI key pair generation. The database host has to have sufficient I/O and memory resources for optimal database performance. As a reference Fermilab US-CMS T1 site uses two dual quad core Intel(R) Xeon(R) CPU E5430 @ 2.66GHz and 32 GB RAM nodes for SRM and SRM database respectively.

Latest Scientific Linux or RHEL would do.

The kernel.shmmax=1073741824 and kernel.shmall=1073741824 kernel parameters should be set for a 4GB RAM Machine. This can be accomplished by running:

[root] # echo 'kernel.shmmax=1073741824' >>  /etc/sysctl.conf
[root] # echo 'kernel.shmall=1073741824' >>  /etc/sysctl.conf
[root] # /bin/sysctl -p

The exact content of US-CMS T1 SRM sysctl.conf is:

kernel.core_uses_pid = 1
kernel.sysrq = 1
kernel.panic = 60
fs.file-max = 131072
net.ipv4.ip_forward = 0
vm.vfs_cache_pressure = 10000
# Keep this amount of memory free for emergency, IRQ and atomic allocations.
vm.min_free_kbytes = 65535
# Network tune parameters
net.ipv4.tcp_timestamps = 0
net.ipv4.tcp_sack = 0
net.ipv4.tcp_window_scaling = 1
kernel.shmmax=1073741824
kernel.shmall=1073741824