view Documentation/networking/gen_stats.txt @ 897:329ea0ccb344

balloon: try harder to balloon up under memory pressure.

Currently if the balloon driver is unable to increase the guest's
reservation it assumes the failure was due to reaching its full
allocation, gives up on the ballooning operation and records the limit
it reached as the "hard limit". The driver will not try again until
the target is set again (even to the same value).

However it is possible that ballooning has in fact failed due to
memory pressure in the host and therefore it is desirable to keep
attempting to reach the target in case memory becomes available. The
most likely scenario is that some guests are ballooning down while
others are ballooning up and therefore there is temporary memory
pressure while things stabilise. You would not expect a well behaved
toolstack to ask a domain to balloon to more than its allocation nor
would you expect it to deliberately over-commit memory by setting
balloon targets which exceed the total host memory.

This patch drops the concept of a hard limit and causes the balloon
driver to retry increasing the reservation on a timer in the same
manner as when decreasing the reservation.

Also if we partially succeed in increasing the reservation
(i.e. receive less pages than we asked for) then we may as well keep
those pages rather than returning them to Xen.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Fri Jun 05 14:01:20 2009 +0100 (2009-06-05)
parents 831230e53067
line source
1 Generic networking statistics for netlink users
2 ======================================================================
4 Statistic counters are grouped into structs:
6 Struct TLV type Description
7 ----------------------------------------------------------------------
8 gnet_stats_basic TCA_STATS_BASIC Basic statistics
9 gnet_stats_rate_est TCA_STATS_RATE_EST Rate estimator
10 gnet_stats_queue TCA_STATS_QUEUE Queue statistics
11 none TCA_STATS_APP Application specific
14 Collecting:
15 -----------
17 Declare the statistic structs you need:
18 struct mystruct {
19 struct gnet_stats_basic bstats;
20 struct gnet_stats_queue qstats;
21 ...
22 };
24 Update statistics:
25 mystruct->tstats.packet++;
26 mystruct->qstats.backlog += skb->pkt_len;
29 Export to userspace (Dump):
30 ---------------------------
32 my_dumping_routine(struct sk_buff *skb, ...)
33 {
34 struct gnet_dump dump;
36 if (gnet_stats_start_copy(skb, TCA_STATS2, &mystruct->lock, &dump) < 0)
37 goto rtattr_failure;
39 if (gnet_stats_copy_basic(&dump, &mystruct->bstats) < 0 ||
40 gnet_stats_copy_queue(&dump, &mystruct->qstats) < 0 ||
41 gnet_stats_copy_app(&dump, &xstats, sizeof(xstats)) < 0)
42 goto rtattr_failure;
44 if (gnet_stats_finish_copy(&dump) < 0)
45 goto rtattr_failure;
46 ...
47 }
49 TCA_STATS/TCA_XSTATS backward compatibility:
50 --------------------------------------------
52 Prior users of struct tc_stats and xstats can maintain backward
53 compatibility by calling the compat wrappers to keep providing the
54 existing TLV types.
56 my_dumping_routine(struct sk_buff *skb, ...)
57 {
58 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
59 TCA_XSTATS, &mystruct->lock, &dump) < 0)
60 goto rtattr_failure;
61 ...
62 }
64 A struct tc_stats will be filled out during gnet_stats_copy_* calls
65 and appended to the skb. TCA_XSTATS is provided if gnet_stats_copy_app
66 was called.
69 Locking:
70 --------
72 Locks are taken before writing and released once all statistics have
73 been written. Locks are always released in case of an error. You
74 are responsible for making sure that the lock is initialized.
77 Rate Estimator:
78 --------------
80 0) Prepare an estimator attribute. Most likely this would be in user
81 space. The value of this TLV should contain a tc_estimator structure.
82 As usual, such a TLV nees to be 32 bit aligned and therefore the
83 length needs to be appropriately set etc. The estimator interval
84 and ewma log need to be converted to the appropriate values.
85 tc_estimator.c::tc_setup_estimator() is advisable to be used as the
86 conversion routine. It does a few clever things. It takes a time
87 interval in microsecs, a time constant also in microsecs and a struct
88 tc_estimator to be populated. The returned tc_estimator can be
89 transported to the kernel. Transfer such a structure in a TLV of type
90 TCA_RATE to your code in the kernel.
92 In the kernel when setting up:
93 1) make sure you have basic stats and rate stats setup first.
94 2) make sure you have initialized stats lock that is used to setup such
95 stats.
96 3) Now initialize a new estimator:
98 int ret = gen_new_estimator(my_basicstats,my_rate_est_stats,
99 mystats_lock, attr_with_tcestimator_struct);
101 if ret == 0
102 success
103 else
104 failed
106 From now on, everytime you dump my_rate_est_stats it will contain
107 uptodate info.
109 Once you are done, call gen_kill_estimator(my_basicstats,
110 my_rate_est_stats) Make sure that my_basicstats and my_rate_est_stats
111 are still valid (i.e still exist) at the time of making this call.
114 Authors:
115 --------
116 Thomas Graf <tgraf@suug.ch>
117 Jamal Hadi Salim <hadi@cyberus.ca>