Thank you very much. I've tested with the same configuration and Windows clients are now able to obtain leases again.
In docs for the new /app,
It does not explain how to use the "Variables To Use In Environment":
and I'm not sure how to use them but be helpful...
More generally explaining the /app networking options, including dynamic firewall rule actions, likely be useful. Right now docs don't really much UI either using different terms/words.
I m looking and curious by this feature :
*) arm64 - allow enabling receive packet steering on /system/resource/irq/rps menu in order to overcome unbalanced CPU load;
What is Receive Packet Steering (RPS)?
- RPS is a Linux kernel feature that helps distribute incoming network packet processing across multiple CPU cores.
- By default, network interrupts (IRQs) are often handled by a single CPU core, leading to uneven CPU utilization (e.g., one core at 100% while others are idle).
- RPS redirects packet processing to less busy cores, improving performance, especially in high-traffic scenarios.
I've enabled this feature for all interfaces on RB5009, except ether5 used as out of band mgmt.
Don't see changes on rb5009 after 1h with lot of iperf.
maybe not enough loaded or not interesting for this kind of routeur.
On one remote test site, upgrade has failed, no dstnat, wg and vxlan work. Need to reboot, downgrade, restore cfg and upgrade again to rc6... Nothing in log but strange.
I'll let running iperf on both sites, on with rps enabled on 1 site only.
But at this time cpu usage is less than 15% everywhere...
Redistribute routes from RIP into OSPF still doesn’t work for me in theese releases, since 7.20beta2. Last working version is 7.19.6
BGP-VPNv4 is unable to redistribute other BGP multihop (recursive) to other PEs still exist in 7.21.rc6
Ive succesfully used RPS on an AX3 with ether1 (2.5G) to Wifi Traffic.
While doing SMB Transfers from a NAS the throughput increased around 100mbps with RPS enabled on ether1.
Second attempt to test v7.21RC on my RB5009, in rc6 seems all issues with IPSEC is finally gone (started in beta2)
Good job MT!
RPS is good for handling frames to/from NIC. In ROS case that would mean anything which relates (more or less directly) to fast path. But doesn't affect anything when (any kind of) HW offload is active.
As soon as there's other processing involved which includes other ROS processes (e.g. firewalling, wireless driver, bridge filters), then other CPU load-sharing mechanizms kick in (if these exist and are applicable).
I've detected two bugs with this rc6 without rps enabled :
- wireguard peer with 14 alloweds ip : some don't work
- some srcnat ipv4 : don't work
Just rollback all to 7.20.6 without any others changes to get working situation.
For me, it was the opposite; in 7.20.7 I had Wi-Fi AC drops, while in 7.21.rc5 and rc6 they were stable.
arm64 - allow enabling receive packet steering on /system/resource/irq/rps menu in order to overcome unbalanced CPU load;
Why is the only arm64 what about 32-bit?
The menu exists on 7.20.7 as well. It can even be enabled. But don't know if it does something.
Lately MikroTik are differentiating feature sets by architecture, to limit the size of the RouterOS image on devices that have only 16MB of flash and are running out of space.
SMIPS devices get less, ARM64 devices get more. This matches the typical size of flash memory in tiny and older devices, and the increased size of flash in (typically newer) ARM64 devices.
Maybe at some time we will see a “RouterOS lite” and “RouterOS” being released, where for the tiniest devices only “lite” will be available and more recent devices (including ARM32?) have the option to select between them to be able to run exotic features on devices with a small configuration that still fits in the flash, or for the ARM32 devices that do have enough flash (e.g. RB4011)?
At the moment, no.
Retrieving routes via SNMP usually insn’t practical anyway, because:
- when there is more than one identical route, the SNMP walk will encounter non-increasing OIDs
- there is no provision (in the SNMP OID) for retrieving more than one route table
Maybe also why in 7.22beta there is an option now to change L009 from arm to arm64.
Haven't seen it working yet, though...
I do still wonder if arm64 ROS + RPS could “fix” the biggest issue with the wap ax (the weak cpu)
because of the cpu the throughput is very limited.
Why do you think that wAP ax could much benefit of RPS? It has only 2 ports (ether1 and ether2) which can handle lots of traffic without much high-level processing ... at the same time it has only 2 CPU cores between which RPS could share the processing load.
IMO we'll see the biggest benefit of this feature on Ampere machines with large number of (relatively low performance) CPU cores. The number of CPU cores on these machines easily exceed 64 (or even multiple times that) while Routerboard devices are mostly limited to 4 CPU cores or less (with exception of high-end CCR2xxx devices with up to 16 CPU cores).
And a side note to ARM vs ARM64 performance: similar instructions on same CPU will typically perform equally regardless the CPU mode (32-bit vs. 64-bit). In some cases, 32-bit performance can be actually better if there's lots of work with RAM due to smaller native size of certain data types.
The performance difference, which is quite notable in x86/x86-64 world of desktop computers, is largely result of using legacy CPU instructions when OS/apps are running in 32-bit mode due to required/wanted backward compatibility with older x86 processors. The difference can become small/negligible if application is optimized to use modern instructions in 32-bit mode. One example are SIMD instructions which are used as standard in 64-bit mode (because all 64-bit x86 processors have at least basic set of such instructions) while they can be used in 32-bit mode if app is specially tailored to do so but many apps are not built this way (if not for other reasons because resulting executable may bemecome much larger than without these alternate code paths).
I wonder what in particular you expect from wAP ax' CPU? I'm using one as AP and can comfortably pass up to 800Mbps via wireless if station handles that kind of throughput (e.g. Samsung galaxy S24 ... A56 can do around 600Mbps in same conditions ... both numbers are ookla speedtest results with WAN speed of 1Gbps). And wAP's CPU is nowhere near 100% (neither core) during tests.
How did you enable it exactly? I’m afk right now, but I recall not seeing anything changed in WinBox (it’s CLI only?)