Based on kernel version 3.9. Page generated on 2013-05-02 23:11 EST.
1 Scaling in the Linux Networking Stack 2 3 4 Introduction 5 ============ 6 7 This document describes a set of complementary techniques in the Linux 8 networking stack to increase parallelism and improve performance for 9 multi-processor systems. 10 11 The following technologies are described: 12 13 RSS: Receive Side Scaling 14 RPS: Receive Packet Steering 15 RFS: Receive Flow Steering 16 Accelerated Receive Flow Steering 17 XPS: Transmit Packet Steering 18 19 20 RSS: Receive Side Scaling 21 ========================= 22 23 Contemporary NICs support multiple receive and transmit descriptor queues 24 (multi-queue). On reception, a NIC can send different packets to different 25 queues to distribute processing among CPUs. The NIC distributes packets by 26 applying a filter to each packet that assigns it to one of a small number 27 of logical flows. Packets for each flow are steered to a separate receive 28 queue, which in turn can be processed by separate CPUs. This mechanism is 29 generally known as âReceive-side Scalingâ (RSS). The goal of RSS and 30 the other scaling techniques is to increase performance uniformly. 31 Multi-queue distribution can also be used for traffic prioritization, but 32 that is not the focus of these techniques. 33 34 The filter used in RSS is typically a hash function over the network 35 and/or transport layer headers-- for example, a 4-tuple hash over 36 IP addresses and TCP ports of a packet. The most common hardware 37 implementation of RSS uses a 128-entry indirection table where each entry 38 stores a queue number. The receive queue for a packet is determined 39 by masking out the low order seven bits of the computed hash for the 40 packet (usually a Toeplitz hash), taking this number as a key into the 41 indirection table and reading the corresponding value. 42 43 Some advanced NICs allow steering packets to queues based on 44 programmable filters. For example, webserver bound TCP port 80 packets 45 can be directed to their own receive queue. Such ân-tupleâ filters can 46 be configured from ethtool (--config-ntuple). 47 48 ==== RSS Configuration 49 50 The driver for a multi-queue capable NIC typically provides a kernel 51 module parameter for specifying the number of hardware queues to 52 configure. In the bnx2x driver, for instance, this parameter is called 53 num_queues. A typical RSS configuration would be to have one receive queue 54 for each CPU if the device supports enough queues, or otherwise at least 55 one for each memory domain, where a memory domain is a set of CPUs that 56 share a particular memory level (L1, L2, NUMA node, etc.). 57 58 The indirection table of an RSS device, which resolves a queue by masked 59 hash, is usually programmed by the driver at initialization. The 60 default mapping is to distribute the queues evenly in the table, but the 61 indirection table can be retrieved and modified at runtime using ethtool 62 commands (--show-rxfh-indir and --set-rxfh-indir). Modifying the 63 indirection table could be done to give different queues different 64 relative weights. 65 66 == RSS IRQ Configuration 67 68 Each receive queue has a separate IRQ associated with it. The NIC triggers 69 this to notify a CPU when new packets arrive on the given queue. The 70 signaling path for PCIe devices uses message signaled interrupts (MSI-X), 71 that can route each interrupt to a particular CPU. The active mapping 72 of queues to IRQs can be determined from /proc/interrupts. By default, 73 an IRQ may be handled on any CPU. Because a non-negligible part of packet 74 processing takes place in receive interrupt handling, it is advantageous 75 to spread receive interrupts between CPUs. To manually adjust the IRQ 76 affinity of each interrupt see Documentation/IRQ-affinity.txt. Some systems 77 will be running irqbalance, a daemon that dynamically optimizes IRQ 78 assignments and as a result may override any manual settings. 79 80 == Suggested Configuration 81 82 RSS should be enabled when latency is a concern or whenever receive 83 interrupt processing forms a bottleneck. Spreading load between CPUs 84 decreases queue length. For low latency networking, the optimal setting 85 is to allocate as many queues as there are CPUs in the system (or the 86 NIC maximum, if lower). The most efficient high-rate configuration 87 is likely the one with the smallest number of receive queues where no 88 receive queue overflows due to a saturated CPU, because in default 89 mode with interrupt coalescing enabled, the aggregate number of 90 interrupts (and thus work) grows with each additional queue. 91 92 Per-cpu load can be observed using the mpstat utility, but note that on 93 processors with hyperthreading (HT), each hyperthread is represented as 94 a separate CPU. For interrupt handling, HT has shown no benefit in 95 initial tests, so limit the number of queues to the number of CPU cores 96 in the system. 97 98 99 RPS: Receive Packet Steering 100 ============================ 101 102 Receive Packet Steering (RPS) is logically a software implementation of 103 RSS. Being in software, it is necessarily called later in the datapath. 104 Whereas RSS selects the queue and hence CPU that will run the hardware 105 interrupt handler, RPS selects the CPU to perform protocol processing 106 above the interrupt handler. This is accomplished by placing the packet 107 on the desired CPUâs backlog queue and waking up the CPU for processing. 108 RPS has some advantages over RSS: 1) it can be used with any NIC, 109 2) software filters can easily be added to hash over new protocols, 110 3) it does not increase hardware device interrupt rate (although it does 111 introduce inter-processor interrupts (IPIs)). 112 113 RPS is called during bottom half of the receive interrupt handler, when 114 a driver sends a packet up the network stack with netif_rx() or 115 netif_receive_skb(). These call the get_rps_cpu() function, which 116 selects the queue that should process a packet. 117 118 The first step in determining the target CPU for RPS is to calculate a 119 flow hash over the packetâs addresses or ports (2-tuple or 4-tuple hash 120 depending on the protocol). This serves as a consistent hash of the 121 associated flow of the packet. The hash is either provided by hardware 122 or will be computed in the stack. Capable hardware can pass the hash in 123 the receive descriptor for the packet; this would usually be the same 124 hash used for RSS (e.g. computed Toeplitz hash). The hash is saved in 125 skb->rx_hash and can be used elsewhere in the stack as a hash of the 126 packetâs flow. 127 128 Each receive hardware queue has an associated list of CPUs to which 129 RPS may enqueue packets for processing. For each received packet, 130 an index into the list is computed from the flow hash modulo the size 131 of the list. The indexed CPU is the target for processing the packet, 132 and the packet is queued to the tail of that CPUâs backlog queue. At 133 the end of the bottom half routine, IPIs are sent to any CPUs for which 134 packets have been queued to their backlog queue. The IPI wakes backlog 135 processing on the remote CPU, and any queued packets are then processed 136 up the networking stack. 137 138 ==== RPS Configuration 139 140 RPS requires a kernel compiled with the CONFIG_RPS kconfig symbol (on 141 by default for SMP). Even when compiled in, RPS remains disabled until 142 explicitly configured. The list of CPUs to which RPS may forward traffic 143 can be configured for each receive queue using a sysfs file entry: 144 145 /sys/class/net/<dev>/queues/rx-<n>/rps_cpus 146 147 This file implements a bitmap of CPUs. RPS is disabled when it is zero 148 (the default), in which case packets are processed on the interrupting 149 CPU. Documentation/IRQ-affinity.txt explains how CPUs are assigned to 150 the bitmap. 151 152 == Suggested Configuration 153 154 For a single queue device, a typical RPS configuration would be to set 155 the rps_cpus to the CPUs in the same memory domain of the interrupting 156 CPU. If NUMA locality is not an issue, this could also be all CPUs in 157 the system. At high interrupt rate, it might be wise to exclude the 158 interrupting CPU from the map since that already performs much work. 159 160 For a multi-queue system, if RSS is configured so that a hardware 161 receive queue is mapped to each CPU, then RPS is probably redundant 162 and unnecessary. If there are fewer hardware queues than CPUs, then 163 RPS might be beneficial if the rps_cpus for each queue are the ones that 164 share the same memory domain as the interrupting CPU for that queue. 165 166 167 RFS: Receive Flow Steering 168 ========================== 169 170 While RPS steers packets solely based on hash, and thus generally 171 provides good load distribution, it does not take into account 172 application locality. This is accomplished by Receive Flow Steering 173 (RFS). The goal of RFS is to increase datacache hitrate by steering 174 kernel processing of packets to the CPU where the application thread 175 consuming the packet is running. RFS relies on the same RPS mechanisms 176 to enqueue packets onto the backlog of another CPU and to wake up that 177 CPU. 178 179 In RFS, packets are not forwarded directly by the value of their hash, 180 but the hash is used as index into a flow lookup table. This table maps 181 flows to the CPUs where those flows are being processed. The flow hash 182 (see RPS section above) is used to calculate the index into this table. 183 The CPU recorded in each entry is the one which last processed the flow. 184 If an entry does not hold a valid CPU, then packets mapped to that entry 185 are steered using plain RPS. Multiple table entries may point to the 186 same CPU. Indeed, with many flows and few CPUs, it is very likely that 187 a single application thread handles flows with many different flow hashes. 188 189 rps_sock_flow_table is a global flow table that contains the *desired* CPU 190 for flows: the CPU that is currently processing the flow in userspace. 191 Each table value is a CPU index that is updated during calls to recvmsg 192 and sendmsg (specifically, inet_recvmsg(), inet_sendmsg(), inet_sendpage() 193 and tcp_splice_read()). 194 195 When the scheduler moves a thread to a new CPU while it has outstanding 196 receive packets on the old CPU, packets may arrive out of order. To 197 avoid this, RFS uses a second flow table to track outstanding packets 198 for each flow: rps_dev_flow_table is a table specific to each hardware 199 receive queue of each device. Each table value stores a CPU index and a 200 counter. The CPU index represents the *current* CPU onto which packets 201 for this flow are enqueued for further kernel processing. Ideally, kernel 202 and userspace processing occur on the same CPU, and hence the CPU index 203 in both tables is identical. This is likely false if the scheduler has 204 recently migrated a userspace thread while the kernel still has packets 205 enqueued for kernel processing on the old CPU. 206 207 The counter in rps_dev_flow_table values records the length of the current 208 CPU's backlog when a packet in this flow was last enqueued. Each backlog 209 queue has a head counter that is incremented on dequeue. A tail counter 210 is computed as head counter + queue length. In other words, the counter 211 in rps_dev_flow[i] records the last element in flow i that has 212 been enqueued onto the currently designated CPU for flow i (of course, 213 entry i is actually selected by hash and multiple flows may hash to the 214 same entry i). 215 216 And now the trick for avoiding out of order packets: when selecting the 217 CPU for packet processing (from get_rps_cpu()) the rps_sock_flow table 218 and the rps_dev_flow table of the queue that the packet was received on 219 are compared. If the desired CPU for the flow (found in the 220 rps_sock_flow table) matches the current CPU (found in the rps_dev_flow 221 table), the packet is enqueued onto that CPUâs backlog. If they differ, 222 the current CPU is updated to match the desired CPU if one of the 223 following is true: 224 225 - The current CPU's queue head counter >= the recorded tail counter 226 value in rps_dev_flow[i] 227 - The current CPU is unset (equal to RPS_NO_CPU) 228 - The current CPU is offline 229 230 After this check, the packet is sent to the (possibly updated) current 231 CPU. These rules aim to ensure that a flow only moves to a new CPU when 232 there are no packets outstanding on the old CPU, as the outstanding 233 packets could arrive later than those about to be processed on the new 234 CPU. 235 236 ==== RFS Configuration 237 238 RFS is only available if the kconfig symbol CONFIG_RPS is enabled (on 239 by default for SMP). The functionality remains disabled until explicitly 240 configured. The number of entries in the global flow table is set through: 241 242 /proc/sys/net/core/rps_sock_flow_entries 243 244 The number of entries in the per-queue flow table are set through: 245 246 /sys/class/net/<dev>/queues/rx-<n>/rps_flow_cnt 247 248 == Suggested Configuration 249 250 Both of these need to be set before RFS is enabled for a receive queue. 251 Values for both are rounded up to the nearest power of two. The 252 suggested flow count depends on the expected number of active connections 253 at any given time, which may be significantly less than the number of open 254 connections. We have found that a value of 32768 for rps_sock_flow_entries 255 works fairly well on a moderately loaded server. 256 257 For a single queue device, the rps_flow_cnt value for the single queue 258 would normally be configured to the same value as rps_sock_flow_entries. 259 For a multi-queue device, the rps_flow_cnt for each queue might be 260 configured as rps_sock_flow_entries / N, where N is the number of 261 queues. So for instance, if rps_sock_flow_entries is set to 32768 and there 262 are 16 configured receive queues, rps_flow_cnt for each queue might be 263 configured as 2048. 264 265 266 Accelerated RFS 267 =============== 268 269 Accelerated RFS is to RFS what RSS is to RPS: a hardware-accelerated load 270 balancing mechanism that uses soft state to steer flows based on where 271 the application thread consuming the packets of each flow is running. 272 Accelerated RFS should perform better than RFS since packets are sent 273 directly to a CPU local to the thread consuming the data. The target CPU 274 will either be the same CPU where the application runs, or at least a CPU 275 which is local to the application threadâs CPU in the cache hierarchy. 276 277 To enable accelerated RFS, the networking stack calls the 278 ndo_rx_flow_steer driver function to communicate the desired hardware 279 queue for packets matching a particular flow. The network stack 280 automatically calls this function every time a flow entry in 281 rps_dev_flow_table is updated. The driver in turn uses a device specific 282 method to program the NIC to steer the packets. 283 284 The hardware queue for a flow is derived from the CPU recorded in 285 rps_dev_flow_table. The stack consults a CPU to hardware queue map which 286 is maintained by the NIC driver. This is an auto-generated reverse map of 287 the IRQ affinity table shown by /proc/interrupts. Drivers can use 288 functions in the cpu_rmap (âCPU affinity reverse mapâ) kernel library 289 to populate the map. For each CPU, the corresponding queue in the map is 290 set to be one whose processing CPU is closest in cache locality. 291 292 ==== Accelerated RFS Configuration 293 294 Accelerated RFS is only available if the kernel is compiled with 295 CONFIG_RFS_ACCEL and support is provided by the NIC device and driver. 296 It also requires that ntuple filtering is enabled via ethtool. The map 297 of CPU to queues is automatically deduced from the IRQ affinities 298 configured for each receive queue by the driver, so no additional 299 configuration should be necessary. 300 301 == Suggested Configuration 302 303 This technique should be enabled whenever one wants to use RFS and the 304 NIC supports hardware acceleration. 305 306 XPS: Transmit Packet Steering 307 ============================= 308 309 Transmit Packet Steering is a mechanism for intelligently selecting 310 which transmit queue to use when transmitting a packet on a multi-queue 311 device. To accomplish this, a mapping from CPU to hardware queue(s) is 312 recorded. The goal of this mapping is usually to assign queues 313 exclusively to a subset of CPUs, where the transmit completions for 314 these queues are processed on a CPU within this set. This choice 315 provides two benefits. First, contention on the device queue lock is 316 significantly reduced since fewer CPUs contend for the same queue 317 (contention can be eliminated completely if each CPU has its own 318 transmit queue). Secondly, cache miss rate on transmit completion is 319 reduced, in particular for data cache lines that hold the sk_buff 320 structures. 321 322 XPS is configured per transmit queue by setting a bitmap of CPUs that 323 may use that queue to transmit. The reverse mapping, from CPUs to 324 transmit queues, is computed and maintained for each network device. 325 When transmitting the first packet in a flow, the function 326 get_xps_queue() is called to select a queue. This function uses the ID 327 of the running CPU as a key into the CPU-to-queue lookup table. If the 328 ID matches a single queue, that is used for transmission. If multiple 329 queues match, one is selected by using the flow hash to compute an index 330 into the set. 331 332 The queue chosen for transmitting a particular flow is saved in the 333 corresponding socket structure for the flow (e.g. a TCP connection). 334 This transmit queue is used for subsequent packets sent on the flow to 335 prevent out of order (ooo) packets. The choice also amortizes the cost 336 of calling get_xps_queues() over all packets in the flow. To avoid 337 ooo packets, the queue for a flow can subsequently only be changed if 338 skb->ooo_okay is set for a packet in the flow. This flag indicates that 339 there are no outstanding packets in the flow, so the transmit queue can 340 change without the risk of generating out of order packets. The 341 transport layer is responsible for setting ooo_okay appropriately. TCP, 342 for instance, sets the flag when all data for a connection has been 343 acknowledged. 344 345 ==== XPS Configuration 346 347 XPS is only available if the kconfig symbol CONFIG_XPS is enabled (on by 348 default for SMP). The functionality remains disabled until explicitly 349 configured. To enable XPS, the bitmap of CPUs that may use a transmit 350 queue is configured using the sysfs file entry: 351 352 /sys/class/net/<dev>/queues/tx-<n>/xps_cpus 353 354 == Suggested Configuration 355 356 For a network device with a single transmission queue, XPS configuration 357 has no effect, since there is no choice in this case. In a multi-queue 358 system, XPS is preferably configured so that each CPU maps onto one queue. 359 If there are as many queues as there are CPUs in the system, then each 360 queue can also map onto one CPU, resulting in exclusive pairings that 361 experience no contention. If there are fewer queues than CPUs, then the 362 best CPUs to share a given queue are probably those that share the cache 363 with the CPU that processes transmit completions for that queue 364 (transmit interrupts). 365 366 367 Further Information 368 =================== 369 RPS and RFS were introduced in kernel 2.6.35. XPS was incorporated into 370 2.6.38. Original patches were submitted by Tom Herbert 371 (firstname.lastname@example.org) 372 373 Accelerated RFS was introduced in 2.6.35. Original patches were 374 submitted by Ben Hutchings (email@example.com) 375 376 Authors: 377 Tom Herbert (firstname.lastname@example.org) 378 Willem de Bruijn (email@example.com)