You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
My Setup
I am running a BeeGFS v8 client on a relatively old machine.
It's a Dell T3500 Precision workstation with a dual-port 10Gbpbs Intel X540-AT2 NIC. I have 10 such machines.
Currently I have both ports on the NIC bonded using balance-xor mode and layer3+4 hashing policy.
The client is connecting to a server machine over an Ethernet switch.
The server machine is exposing a single BeeGFS target on a very fast RAID-6 volume, from a single IP address.
On the client machine I am running my own program that does file writes and reads to /mnt/beegfs folder, using 32MB buffer,15 threads, and direct IO.
I am expecting ~2GB/s throughput on both writes and reads, as the limit should be the network speed, the server RAID-6 is very fast can do 20GB/s reads and 16GB/s writes.
Question #1:
Is it possible to break the bond on the dual-port NIC, assign 2 different IP addresses to each port that are on the same subnet as the server management IP, and configure beegfs-client to utilize both ports in a load-balanced way. Basically, I would like to mount two different local folders to the same server target, such that each folder mount is using a different NIC port, then run two instances of my program concurrently to achieve 2GB/s aggregate throughput?
Question #2:
Across my 10 Dell client machines I am seeing vary wide range of throughput. Only 1 out of 10 machines is able to get close to the expected 2GB/s read and write performance. The others range between 1-1.8 GB/s. Another strange behavior I see is some client machines can attain 2GB/s write speed, but read speed is far lower, closer to 1GB/s. Any suggestions anyone can offer as to what might be causing such a variance in behavior? These Dell machines are fairly similar. All using the same NIC, some have different class CPU with fewer cores and less RAM. Core count ranges between 8-12. RAM varies between 8-12GB.
The most unexpected is if a client machine can write fast, why can't it read just as fast?
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
My Setup
I am running a BeeGFS v8 client on a relatively old machine.
It's a Dell T3500 Precision workstation with a dual-port 10Gbpbs Intel X540-AT2 NIC. I have 10 such machines.
Currently I have both ports on the NIC bonded using balance-xor mode and layer3+4 hashing policy.
The client is connecting to a server machine over an Ethernet switch.
The server machine is exposing a single BeeGFS target on a very fast RAID-6 volume, from a single IP address.
On the client machine I am running my own program that does file writes and reads to /mnt/beegfs folder, using 32MB buffer,15 threads, and direct IO.
I am expecting ~2GB/s throughput on both writes and reads, as the limit should be the network speed, the server RAID-6 is very fast can do 20GB/s reads and 16GB/s writes.
Question #1:
Is it possible to break the bond on the dual-port NIC, assign 2 different IP addresses to each port that are on the same subnet as the server management IP, and configure beegfs-client to utilize both ports in a load-balanced way. Basically, I would like to mount two different local folders to the same server target, such that each folder mount is using a different NIC port, then run two instances of my program concurrently to achieve 2GB/s aggregate throughput?
Question #2:
Across my 10 Dell client machines I am seeing vary wide range of throughput. Only 1 out of 10 machines is able to get close to the expected 2GB/s read and write performance. The others range between 1-1.8 GB/s. Another strange behavior I see is some client machines can attain 2GB/s write speed, but read speed is far lower, closer to 1GB/s. Any suggestions anyone can offer as to what might be causing such a variance in behavior? These Dell machines are fairly similar. All using the same NIC, some have different class CPU with fewer cores and less RAM. Core count ranges between 8-12. RAM varies between 8-12GB.
The most unexpected is if a client machine can write fast, why can't it read just as fast?
Beta Was this translation helpful? Give feedback.
All reactions