NFS multiple TCP sessions per client support

[storageninja]

Hi

Especially for a full NVMe NAS like the F4 SSD with 5Gbps we can benefit from multiple TCP sessions for I/O load balancing. Also for my server connected via LACP, or in a professional setup, the load balancing via NIC's improves too with multiple IP+PORTS hashing.

In the screenshots below you can see that my old Synology with DSM 7.2 does support it. Can I enable it on CLI? I hope you implement it via the GUI too.

Synology 172.16.1.2
TNAS 172.16.1.22

Proxmox mounts

Code: Select all

nfs: TNAS-datastore1
	export /Volume1/TNAS-datastore1
	path /mnt/pve/TNAS-datastore1
	server 172.16.1.22
	content images,iso,vztmpl,rootdir
	options vers=4.1,hard,intr,tcp,noatime,rsize=1048576,wsize=1048576,timeo=600,retrans=5,retry=2,nconnect=4
	preallocation metadata
	prune-backups keep-all=1

Code: Select all

nfs: SYNO-backup
	export /volume1/SYNO-datastore
	path /mnt/pve/SYNO-backup
	server 172.16.1.2
	content backup,images
        options vers=4.1,hard,intr,tcp,noatime,rsize=1048576,wsize=1048576,timeo=600,retrans=5,retry=2,nconnect=4
	preallocation metadata
	prune-backups keep-all=1

Code: Select all

root@pve1:~# ss -tna | grep :2049 | grep ESTAB
ESTAB     0      0                      172.16.1.21:794             172.16.1.2:2049
ESTAB     0      0                      172.16.1.21:827            172.16.1.22:2049
ESTAB     0      0                      172.16.1.21:711             172.16.1.2:2049
ESTAB     0      0                      172.16.1.21:781             172.16.1.2:2049
ESTAB     0      0                      172.16.1.21:929             172.16.1.2:2049

[storageninja]

Code: Select all

root@pve:~# time dd if=/dev/zero of=/mnt/pve/TNAS-datastore1/test1threads bs=1M count=1000
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB, 1000 MiB) copied, 3.73299 s, 281 MB/s

real	0m3.866s
user	0m0.000s
sys	0m0.181s

root@pve:~# time dd if=/dev/zero of=/mnt/pve/TNAS-datastore1/test4threads bs=1M count=1000
1000+0 records in
1000+0 records out
1048576000 bytes (1.0 GB, 1000 MiB) copied, 1.95228 s, 537 MB/s

real	0m2.137s
user	0m0.002s
sys	0m0.173s

You can see the performance penalty of the low sessions directly. My server is attached with a LACP bundle of 2,5Gbps (2x) to the network, the F4 SSD got a 5Gbps port. With the mount set to 4 session the performance increases with 91% and reaching 5Gbps over 2 links.




Current Limitation
Per-client connection pooling - TOS distributes a fixed connection budget across all mounts from the same client instead of honoring per-mount nconnect requests
Enhancement Requests for TOS Development Team:

• Remove Per-Client Connection Limits
  Current: ~8 connections total per client IP
  Requested: Honor individual mount nconnect parameters
  Expected result: 4 mounts × nconnect=4 = 16 connections
  
  Add GUI Configuration Options
  Current: No GUI controls for session trunking

Requested: Add advanced NFS options in web interface:

• Maximum connections per client (configurable)
  Default nconnect behavior
  Performance tuning presets

Enhanced Connection Management
Current: Simple connection distribution algorithm
Requested: Intelligent connection allocation:

• Prioritize active mounts
  Load-based connection scaling
  Per-export connection limits

[Kunu]

You've correctly identified a limitation with the NFS implementation on the TOS device. Your ss command output clearly shows that even with the nconnect=4 option, only a single TCP connection is established to the TNAS, while the Synology correctly utilizes multiple connections.coc mod apk

+1 for this request.

[CursaYang]

Hello, after our testing, TOS 6.0.770 supports multiple TCP connections. What version of the system are you currently running?

[storageninja]

Hello, after our testing, TOS 6.0.770 supports multiple TCP connections. What version of the system are you currently running?

6.0.770 did you run multiple volumes? After my findings it seems there is a total limit in sessions, so with multiple volumes the session count drops because it needs to be shared over all volumes? Could you describe your setup? My was with 2 hypervisors and 3 volumes per hypervisor.