BeeGFS 7.3.1 and 7.2.7 Released
The BeeGFS, a software-defined parallel filesystem designed for high-performance computing (HPC) and genomics workloads, has just released two new versions. The 7.3.1 release includes changes to the default mount options and security configuration as well as some other fixes, QoL improvements, along with updated the client module to work with Linux 5.15 to support RHEL 9 and Ubuntu 22.04.
v7.3.1
General Changes/Improvements
- The
native
cache mode is no longer marked as experimental. - TCP and UDP receive buffer sizes can now be configured independently from the RDMA buffer size with the configuration options
connTCPRcvBufSize
andconnUDPRcvBufSize
. The default values of0
will preserve the legacy behavior of setting the buffer size toconnRDMABufNum * connRDMABufSize
. Setting the value to-1
will set the buffer size to system defaults. - Improved performance when selecting a connection from the client connection pool by reordering the connection according to availability.
- Updated the client module to work with Linux 5.15 to support RHEL 9 and Ubuntu 22.04.
- Removed unnecessary memory context switching in standard socket code.
- Some general code cleanup.
Fixes
- RDMA events with event code
RDMA_CM_EVENT_TIMEWAIT_EXIT
will no longer be logged by default. They will still be logged on log levelDEBUG
. - Added
connAuthFile
configuration to client runtime config in procfs to fix an issue wherebeegfs-ctl
with the--mount
flag was not able to connect to other servers whenconnAuthFile
is configured. - Fixed a bug in the RDMA buffer allocation that caused the client module to crash during OPA and RoCE device initialization on newer kernels (RHEL >8.5, Ubuntu 22.04).
- The
--access
flag in thebeegfs-ctl
modes--createdir
and--createfile
will no longer ignore special bits for setuid, setgid and sticky.
For more information, please visit the official blog post.
v7.2.7
General Changes/Improvements
- The client will now use the
nosuid
mount option by default to prevent users from escalating privileges by using setuid binaries. This can be overridden by settingsuid
as a mount option inbeegfs-mounts.conf
- Configuring a
connAuthFile
is now mandatory. Services will no longer start ifconnAuthFile
is not configured or connection authentication is explicitly disabled by setting the new configuration optionconnDisableAuthentication
totrue
. Please refer to our documentation about Authentication for more information about setting up connection authentication.
Fixes
- RDMA events with event code
RDMA_CM_EVENT_TIMEWAIT_EXIT
will no longer be logged by default. They will still be logged on log levelDEBUG
. - Added
connAuthFile
configuration to client runtime config in procfs to fix an issue wherebeegfs-ctl
with the--mount
flag was not able to connect to other servers whenconnAuthFile
is configured. - Fixed a bug in the RDMA buffer allocation that caused the client module to crash during OPA and RoCE device initialization on newer kernels (RHEL >8.5).
- The
--access
flag in thebeegfs-ctl
modes--createdir
and--createfile
will no longer ignore special bits for setuid, setgid and sticky.
For more information, please visit the official blog post.
Robust HPC – BeeGFS Authorized Reseller in Malaysia
Robust HPC Sdn Bhd is the authorized reseller of BeeGFS in Malaysia. You can purchase a BeeGFS software license with us, or buy the pre-configured turn-key appliance from us. Technical support services are available with the choice of 8×5 or 24×7.
The following diagram shows the reference architecture of BeeGFS Storage based on Dell EMC servers and NVIDIA (Mellanox) InfiniBand network. The sequential read and write performance hit 152.9GB/s and 139.9GB/s respectively in a 14U configuration. 联系我们 to obtain the details.
BeeGFS can also run on any x86 servers such as SuperMicro, Huawei and other brands. If you need any help in selecting the best hardware, feel free to engage us for consulting services.
发表评论