Re: [PATCH RFC net-next] net/smc: transition to RDMA core CQ pooling

[Mahanta Jambigi]

On 24/02/26 7:49 am, D. Wythe wrote:
> On Fri, Feb 13, 2026 at 04:53:28PM +0530, Mahanta Jambigi wrote:
>>
>>
>> On 09/02/26 1:23 pm, D. Wythe wrote:
>>> On Fri, Feb 06, 2026 at 04:58:23PM +0530, Mahanta Jambigi wrote:
>>>>
>>>>
>>>> On 02/02/26 3:18 pm, D. Wythe wrote:
>>>>> The current SMC-R implementation relies on global per-device CQs
>>>>> and manual polling within tasklets, which introduces severe
>>>>> scalability bottlenecks due to global lock contention and tasklet
>>>>> scheduling overhead, resulting in poor performance as concurrency
>>>>> increases.
>>>>>
>>>>> Refactor the completion handling to utilize the ib_cqe API and
>>>>> standard RDMA core CQ pooling. This transition provides several key
>>>>> advantages:
>>>>>
>>>>> 1. Multi-CQ: Shift from a single shared per-device CQ to multiple
>>>>> link-specific CQs via the CQ pool. This allows completion processing
>>>>> to be parallelized across multiple CPU cores, effectively eliminating
>>>>> the global CQ bottleneck.
>>>>>
>>>>> 2. Leverage DIM: Utilizing the standard CQ pool with IB_POLL_SOFTIRQ
>>>>> enables Dynamic Interrupt Moderation from the RDMA core, optimizing
>>>>> interrupt frequency and reducing CPU load under high pressure.
>>>>>
>>>>> 3. O(1) Context Retrieval: Replaces the expensive wr_id based lookup
>>>>> logic (e.g., smc_wr_tx_find_pending_index) with direct context retrieval
>>>>> using container_of() on the embedded ib_cqe.
>>>>>
>>>>> 4. Code Simplification: This refactoring results in a reduction of
>>>>> ~150 lines of code. It removes redundant sequence tracking, complex lookup
>>>>> helpers, and manual CQ management, significantly improving maintainability.
>>>>>
>>>>> Performance Test: redis-benchmark with max 32 connections per QP
>>>>> Data format: Requests Per Second (RPS), Percentage in brackets
>>>>> represents the gain/loss compared to TCP.
>>>>>
>>>>> | Clients | TCP      | SMC (original)      | SMC (cq_pool)       |
>>>>> |---------|----------|---------------------|---------------------|
>>>>> | c = 1   | 24449    | 31172  (+27%)       | 34039  (+39%)       |
>>>>> | c = 2   | 46420    | 53216  (+14%)       | 64391  (+38%)       |
>>>>> | c = 16  | 159673   | 83668  (-48%)  <--  | 216947 (+36%)       |
>>>>> | c = 32  | 164956   | 97631  (-41%)  <--  | 249376 (+51%)       |
>>>>> | c = 64  | 166322   | 118192 (-29%)  <--  | 249488 (+50%)       |
>>>>> | c = 128 | 167700   | 121497 (-27%)  <--  | 249480 (+48%)       |
>>>>> | c = 256 | 175021   | 146109 (-16%)  <--  | 240384 (+37%)       |
>>>>> | c = 512 | 168987   | 101479 (-40%)  <--  | 226634 (+34%)       |
>>>>>
>>>>> The results demonstrate that this optimization effectively resolves the
>>>>> scalability bottleneck, with RPS increasing by over 110% at c=64
>>>>> compared to the original implementation.
>>>>
>>>> I applied your patch to the latest kernel(6.19-rc8) & saw below
>>>> Performance results:
>>>>
>>>> 1) In my evaluation, I ran several *uperf* based workloads using a
>>>> request/response (RR) pattern, and I observed performance *degradation*
>>>> ranging from *4%* to *59%*, depending on the specific read/write sizes
>>>> used. For example, with a TCP RR workload using 50 parallel clients
>>>> (nprocs=50) sending a 200‑byte request and reading a 1000‑byte response
>>>> over a 60‑second run, I measured approximately 59% degradation compared
>>>> to SMC‑R original performance.
>>>>
>>>
>>> The only setting I changed was net.smc.smcr_max_conns_per_lgr = 32, all
>>> other parameters were left at their default values. redis-benchmark is a
>>> classic Request/Response (RR) workload, which contradicts your test
>>> results. Since I'm unable to reproduce your results, it would be
>>> very helpful if you could share the specific test configuration for my
>>> analysis.
>>
>> I used a simple client–server setup connected via 25 Gb/s RoCE_Express2
>> adapters on the same LAN(connection established via SMC-R v1). After
>> running the commands shown below, I observed a performance degradation
>> of up to 59%.
>>
>> Server: smc_run uperf -s
>> Client: smc_run uperf -m rr1c-200x1000-50.xml
>>
>> cat rr1c-200x1000-50.xml
>>
>> <?xml version="1.0"?>
>> <profile name="TCP_RR">
>> 	<group nprocs="50">
>> 		<transaction iterations="1">
>> 			<flowop type="connect" options="remotehost=server_ip protocol=tcp
>> tcp_nodelay" />
>> 		</transaction>
>> 		<transaction duration="60">
>> 			<flowop type="write" options="size=200"/>
>> 			<flowop type="read" options="size=1000"/>
>> 		</transaction>
>> 		<transaction iterations="1">
>> 			<flowop type="disconnect" />
>> 		</transaction>
>> 	</group>
>> </profile>
> 
> Using the exact same XML profile you provided, I tested this on a 25Gb
> NIC. I observed no degradation. Instead, performance improved
> significantly:
> 
> Original: ~1.08 Gb/s
> Patched: ~5.1 Gb/s
> 
> I suspect the 59% drop might be due to connections falling back to TCP.
> Could you check smcss -a during your test to see if the traffic is
> actually running over SMC-R?

I have checked this. The connection was successful using *SMCR* Mode
itself. Also I have confirmed this via 'smcr -d stats' command which
shows 0 count for TCP fallback.

> 
>>
>> I installed redis-server on the server machine & redis-benchmark on the
>> client machine & I was able to establish the SMC-R using below commands.
>> If you could help me with the exact commands you used to measure the
>> redis-benchmark performance, I can try the same on my setup.
>>
>> Server: smc_run redis-server --port <port_num> --save "" --appendonly no
>>   --protected-mode no --bind 0.0.0.0
>> Client: smc_run redis-benchmark -h <server_ip> -p <port_num> -n 10000 -c
>> 50 -t ping_inline,ping_bulk -q
> 
> Here are the exact commands and scripts I used for the
> redis-benchmark:
> 
> Server: smc_run redis-server --protected-mode no --save
> 
> Client: smc_run redis-benchmark -h <server_ip> -n 5000000 -t set --threads 3
> -c <conn_num>
> 
> D. Wythe