Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
ip_vs_sed.c
Go to the documentation of this file.
1 /*
2  * IPVS: Shortest Expected Delay scheduling module
3  *
4  * Authors: Wensong Zhang <[email protected]>
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  *
11  * Changes:
12  *
13  */
14 
15 /*
16  * The SED algorithm attempts to minimize each job's expected delay until
17  * completion. The expected delay that the job will experience is
18  * (Ci + 1) / Ui if sent to the ith server, in which Ci is the number of
19  * jobs on the ith server and Ui is the fixed service rate (weight) of
20  * the ith server. The SED algorithm adopts a greedy policy that each does
21  * what is in its own best interest, i.e. to join the queue which would
22  * minimize its expected delay of completion.
23  *
24  * See the following paper for more information:
25  * A. Weinrib and S. Shenker, Greed is not enough: Adaptive load sharing
26  * in large heterogeneous systems. In Proceedings IEEE INFOCOM'88,
27  * pages 986-994, 1988.
28  *
29  * Thanks must go to Marko Buuri <[email protected]> for talking SED to me.
30  *
31  * The difference between SED and WLC is that SED includes the incoming
32  * job in the cost function (the increment of 1). SED may outperform
33  * WLC, while scheduling big jobs under larger heterogeneous systems
34  * (the server weight varies a lot).
35  *
36  */
37 
38 #define KMSG_COMPONENT "IPVS"
39 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
40 
41 #include <linux/module.h>
42 #include <linux/kernel.h>
43 
44 #include <net/ip_vs.h>
45 
46 
47 static inline unsigned int
48 ip_vs_sed_dest_overhead(struct ip_vs_dest *dest)
49 {
50  /*
51  * We only use the active connection number in the cost
52  * calculation here.
53  */
54  return atomic_read(&dest->activeconns) + 1;
55 }
56 
57 
58 /*
59  * Weighted Least Connection scheduling
60  */
61 static struct ip_vs_dest *
62 ip_vs_sed_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
63 {
64  struct ip_vs_dest *dest, *least;
65  unsigned int loh, doh;
66 
67  IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
68 
69  /*
70  * We calculate the load of each dest server as follows:
71  * (server expected overhead) / dest->weight
72  *
73  * Remember -- no floats in kernel mode!!!
74  * The comparison of h1*w2 > h2*w1 is equivalent to that of
75  * h1/w1 > h2/w2
76  * if every weight is larger than zero.
77  *
78  * The server with weight=0 is quiesced and will not receive any
79  * new connections.
80  */
81 
83  if (!(dest->flags & IP_VS_DEST_F_OVERLOAD) &&
84  atomic_read(&dest->weight) > 0) {
85  least = dest;
86  loh = ip_vs_sed_dest_overhead(least);
87  goto nextstage;
88  }
89  }
90  ip_vs_scheduler_err(svc, "no destination available");
91  return NULL;
92 
93  /*
94  * Find the destination with the least load.
95  */
96  nextstage:
98  if (dest->flags & IP_VS_DEST_F_OVERLOAD)
99  continue;
100  doh = ip_vs_sed_dest_overhead(dest);
101  if (loh * atomic_read(&dest->weight) >
102  doh * atomic_read(&least->weight)) {
103  least = dest;
104  loh = doh;
105  }
106  }
107 
108  IP_VS_DBG_BUF(6, "SED: server %s:%u "
109  "activeconns %d refcnt %d weight %d overhead %d\n",
110  IP_VS_DBG_ADDR(svc->af, &least->addr), ntohs(least->port),
111  atomic_read(&least->activeconns),
112  atomic_read(&least->refcnt),
113  atomic_read(&least->weight), loh);
114 
115  return least;
116 }
117 
118 
119 static struct ip_vs_scheduler ip_vs_sed_scheduler =
120 {
121  .name = "sed",
122  .refcnt = ATOMIC_INIT(0),
123  .module = THIS_MODULE,
124  .n_list = LIST_HEAD_INIT(ip_vs_sed_scheduler.n_list),
125  .schedule = ip_vs_sed_schedule,
126 };
127 
128 
129 static int __init ip_vs_sed_init(void)
130 {
131  return register_ip_vs_scheduler(&ip_vs_sed_scheduler);
132 }
133 
134 static void __exit ip_vs_sed_cleanup(void)
135 {
136  unregister_ip_vs_scheduler(&ip_vs_sed_scheduler);
137 }
138 
139 module_init(ip_vs_sed_init);
140 module_exit(ip_vs_sed_cleanup);
141 MODULE_LICENSE("GPL");