Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
1.. SPDX-License-Identifier: GPL-2.0
2
3==========
4SMC Sysctl
5==========
6
7/proc/sys/net/smc/* Variables
8=============================
9
10autocorking_size - INTEGER
11 Setting SMC auto corking size:
12 SMC auto corking is like TCP auto corking from the application's
13 perspective of view. When applications do consecutive small
14 write()/sendmsg() system calls, we try to coalesce these small writes
15 as much as possible, to lower total amount of CDC and RDMA Write been
16 sent.
17 autocorking_size limits the maximum corked bytes that can be sent to
18 the under device in 1 single sending. If set to 0, the SMC auto corking
19 is disabled.
20 Applications can still use TCP_CORK for optimal behavior when they
21 know how/when to uncork their sockets.
22
23 Default: 64K
24
25smcr_buf_type - INTEGER
26 Controls which type of sndbufs and RMBs to use in later newly created
27 SMC-R link group. Only for SMC-R.
28
29 Default: 0 (physically contiguous sndbufs and RMBs)
30
31 Possible values:
32
33 - 0 - Use physically contiguous buffers
34 - 1 - Use virtually contiguous buffers
35 - 2 - Mixed use of the two types. Try physically contiguous buffers first.
36 If not available, use virtually contiguous buffers then.
37
38smcr_testlink_time - INTEGER
39 How frequently SMC-R link sends out TEST_LINK LLC messages to confirm
40 viability, after the last activity of connections on it. Value 0 means
41 disabling TEST_LINK.
42
43 Default: 30 seconds.
44
45wmem - INTEGER
46 Initial size of send buffer used by SMC sockets.
47
48 The minimum value is 16KiB and there is no hard limit for max value, but
49 only allowed 512KiB for SMC-R and 1MiB for SMC-D.
50
51 Default: 64KiB
52
53rmem - INTEGER
54 Initial size of receive buffer (RMB) used by SMC sockets.
55
56 The minimum value is 16KiB and there is no hard limit for max value, but
57 only allowed 512KiB for SMC-R and 1MiB for SMC-D.
58
59 Default: 64KiB
60
61smcr_max_links_per_lgr - INTEGER
62 Controls the max number of links can be added to a SMC-R link group. Notice that
63 the actual number of the links added to a SMC-R link group depends on the number
64 of RDMA devices exist in the system. The acceptable value ranges from 1 to 2. Only
65 for SMC-R v2.1 and later.
66
67 Default: 2
68
69smcr_max_conns_per_lgr - INTEGER
70 Controls the max number of connections can be added to a SMC-R link group. The
71 acceptable value ranges from 16 to 255. Only for SMC-R v2.1 and later.
72
73 Default: 255
74
75smcr_max_send_wr - INTEGER
76 So-called work request buffers are SMCR link (and RDMA queue pair) level
77 resources necessary for performing RDMA operations. Since up to 255
78 connections can share a link group and thus also a link and the number
79 of the work request buffers is decided when the link is allocated,
80 depending on the workload it can be a bottleneck in a sense that threads
81 have to wait for work request buffers to become available. Before the
82 introduction of this control the maximal number of work request buffers
83 available on the send path used to be hard coded to 16. With this control
84 it becomes configurable. The acceptable range is between 2 and 2048.
85
86 Please be aware that all the buffers need to be allocated as a physically
87 continuous array in which each element is a single buffer and has the size
88 of SMC_WR_BUF_SIZE (48) bytes. If the allocation fails, we keep retrying
89 with half of the buffer count until it is ether successful or (unlikely)
90 we dip below the old hard coded value which is 16 where we give up much
91 like before having this control.
92
93 Default: 16
94
95smcr_max_recv_wr - INTEGER
96 So-called work request buffers are SMCR link (and RDMA queue pair) level
97 resources necessary for performing RDMA operations. Since up to 255
98 connections can share a link group and thus also a link and the number
99 of the work request buffers is decided when the link is allocated,
100 depending on the workload it can be a bottleneck in a sense that threads
101 have to wait for work request buffers to become available. Before the
102 introduction of this control the maximal number of work request buffers
103 available on the receive path used to be hard coded to 16. With this control
104 it becomes configurable. The acceptable range is between 2 and 2048.
105
106 Please be aware that all the buffers need to be allocated as a physically
107 continuous array in which each element is a single buffer and has the size
108 of SMC_WR_BUF_SIZE (48) bytes. If the allocation fails, we keep retrying
109 with half of the buffer count until it is ether successful or (unlikely)
110 we dip below the old hard coded value which is 16 where we give up much
111 like before having this control.
112
113 Default: 48