2 * IPVS: Shortest Expected Delay scheduling module
4 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
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.
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.
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.
29 * Thanks must go to Marko Buuri <marko@buuri.name> for talking SED to me.
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).
38 #define KMSG_COMPONENT "IPVS"
39 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
41 #include <linux/module.h>
42 #include <linux/kernel.h>
44 #include <net/ip_vs.h>
47 static inline unsigned int
48 ip_vs_sed_dest_overhead(struct ip_vs_dest
*dest
)
51 * We only use the active connection number in the cost
54 return atomic_read(&dest
->activeconns
) + 1;
59 * Weighted Least Connection scheduling
61 static struct ip_vs_dest
*
62 ip_vs_sed_schedule(struct ip_vs_service
*svc
, const struct sk_buff
*skb
)
64 struct ip_vs_dest
*dest
, *least
;
65 unsigned int loh
, doh
;
67 IP_VS_DBG(6, "%s(): Scheduling...\n", __func__
);
70 * We calculate the load of each dest server as follows:
71 * (server expected overhead) / dest->weight
73 * Remember -- no floats in kernel mode!!!
74 * The comparison of h1*w2 > h2*w1 is equivalent to that of
76 * if every weight is larger than zero.
78 * The server with weight=0 is quiesced and will not receive any
82 list_for_each_entry_rcu(dest
, &svc
->destinations
, n_list
) {
83 if (!(dest
->flags
& IP_VS_DEST_F_OVERLOAD
) &&
84 atomic_read(&dest
->weight
) > 0) {
86 loh
= ip_vs_sed_dest_overhead(least
);
90 ip_vs_scheduler_err(svc
, "no destination available");
94 * Find the destination with the least load.
97 list_for_each_entry_continue_rcu(dest
, &svc
->destinations
, n_list
) {
98 if (dest
->flags
& IP_VS_DEST_F_OVERLOAD
)
100 doh
= ip_vs_sed_dest_overhead(dest
);
101 if (loh
* atomic_read(&dest
->weight
) >
102 doh
* atomic_read(&least
->weight
)) {
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
);
119 static struct ip_vs_scheduler ip_vs_sed_scheduler
=
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
,
129 static int __init
ip_vs_sed_init(void)
131 return register_ip_vs_scheduler(&ip_vs_sed_scheduler
);
134 static void __exit
ip_vs_sed_cleanup(void)
136 unregister_ip_vs_scheduler(&ip_vs_sed_scheduler
);
140 module_init(ip_vs_sed_init
);
141 module_exit(ip_vs_sed_cleanup
);
142 MODULE_LICENSE("GPL");