SGW TAC Selection

[TerryAlu]

Instead of selecting SGW with round-robin manner, I make some changes to select an available SGW by TAC of eNodeB.

[acetcom]

We will apply this fancy configuration to NextEPC. But, we are wondering how to define the configuration file.

Here is our proposal.

sgw_selection:  rr ( round-robin), tac(use predefined TAC) , dns( use DNS query based on 3gpp spec)

Now we are a bit busy with other things. Soon we will review this section and apply it.
Thank you very much for giving us this idea,

[TerryAlu]

It sounds great to define the configuration file by sgw_selection, so I change the keyword from tac_selection to sgw_selection. Besides, I correct my indentations at previous commit.

[acetcom]

Ultra late response!. So sorry!

IMHO, there is a little problem in your great work!. You added tac information in gtp_node_t structure. This approach is not allowed. mme_sgw_t structure with including gtp_node_t should be created in MME context.

If you are not changing it, let us modify your source code for creating mme_sgw_t structure.

Thanks a lot!

[TerryAlu]

Creating mme_sgw_t in MME context seems better. Please help me to modify it. Thanks for the suggestions!

[acetcom]

Here is my proposal for mme_sgw_t structure.

#include "gtp/gtp_node.h"

typedef struct _mme_sgw_t {
    gtp_node_t      gnode;

    c_uint16_t        tac[MAX_NUM_OF_TAI];
    c_uint8_t          num_of_tac;
} mme_sgw_t;

typedef struct _mme_context_t {
   mme_sgw_t     *sgw;
}

gtp_node_t has a lnode_t. So you can use linked-list function with mme_sgw_t structure.
And also, please try to reduce the (gtp_node_t *) type-casting to use function call parameter.

Thanks!

[TerryAlu]

I have some questions to use linked-list function with mme_sgw_t.

After I create links between mme_sgw_t structures as the following figure, should I create a variable in mme_context_t to hold the reference to the first mme_sgw_t element? So I can traverse the link from the first every time.

Besides, is it better to change the content of sgw_list from gtp_node_t to mme_sgw_t? If it is true, I need to define a function similar to gtp_add_node in mme_context.h in order to avoid modifying gtp_node.c. But this will cause some copy-and-paste programming.

[acetcom]

Ok! Let me clarify my idea with using source code as follows.

diff --git a/lib/gtp/gtp_node.h b/lib/gtp/gtp_node.h
index b24c3b6e..3b868e79 100644
--- a/lib/gtp/gtp_node.h
+++ b/lib/gtp/gtp_node.h
@@ -14,7 +14,7 @@ extern "C" {
     do { \
         d_assert((__cTX), break, "Null param"); \
         d_assert((__gNODE), break, "Null param"); \
-        (__cTX)->gnode = __gNODE; \
+        (__cTX)->gnode = (gtp_node_t *)__gNODE; \
     } while(0)

 /**
diff --git a/src/mme/mme_context.h b/src/mme/mme_context.h
index 852c082c..c976e6b4 100644
--- a/src/mme/mme_context.h
+++ b/src/mme/mme_context.h
@@ -16,6 +16,7 @@
 #include "s1ap/s1ap_message.h"
 #include "nas/nas_message.h"
 #include "fd/s6a/s6a_message.h"
+#include "gtp/gtp_node.h"

 /* S1AP */
 #include "S1AP_Cause.h"
@@ -34,6 +35,8 @@ extern "C" {

 #define MAX_NUM_OF_BPLMN            6

+typedef struct _mme_sgw_t mme_sgw_t;
+
 typedef struct _enb_ue_t enb_ue_t;
 typedef struct _mme_ue_t mme_ue_t;

@@ -72,7 +75,7 @@ typedef struct _mme_context_t {
     c_sockaddr_t    *gtpc_addr6;    /* MME GTPC IPv6 Address */

     list_t          sgw_list;       /* SGW GTPC Client List */
-    gtp_node_t      *sgw;           /* Iterator for SGW round-robin */
+    mme_sgw_t       *sgw;           /* Iterator for SGW round-robin */

     list_t          pgw_list;       /* PGW GTPC Client List */
     c_sockaddr_t    *pgw_addr;      /* First IPv4 Address Selected */

 #define MAX_NUM_OF_BPLMN            6

+typedef struct _mme_sgw_t mme_sgw_t;
+
 typedef struct _enb_ue_t enb_ue_t;
 typedef struct _mme_ue_t mme_ue_t;

@@ -72,7 +75,7 @@ typedef struct _mme_context_t {
     c_sockaddr_t    *gtpc_addr6;    /* MME GTPC IPv6 Address */

     list_t          sgw_list;       /* SGW GTPC Client List */
-    gtp_node_t      *sgw;           /* Iterator for SGW round-robin */
+    mme_sgw_t       *sgw;           /* Iterator for SGW round-robin */

     list_t          pgw_list;       /* PGW GTPC Client List */
     c_sockaddr_t    *pgw_addr;      /* First IPv4 Address Selected */
@@ -140,6 +143,13 @@ typedef struct _mme_context_t {

 } mme_context_t;

+typedef struct _mme_sgw_t {
+    gtp_node_t      gnode;
+
+    c_uint16_t      tac[MAX_NUM_OF_TAI];
+    c_uint8_t       num_of_tac;
+} mme_sgw_t;
+
 typedef struct _mme_enb_t {
     index_t         index;  /* An index of this node */
     fsm_t           sm;     /* A state machine */

What do you think about it?
Feel free to raise any issue.

Thanks a lot!

[TerryAlu]

I have structures similar to yours in mme_context.h. But, I have a trouble allocating memory to mme_sgw_t structure and making mme_sgw_t linked list. Because I want to use gtp_add_node function, which only works with gtp_node_t.

[acetcom]

There is a problem the way I proposed. I'm so sorry if you took a lot of time to figure this out.
We think we should fix a lot of source code to support tac_select function.

typedef struct _mme_sgw_t {
    lnode_t node ;

    c_uint16_t      tac[MAX_NUM_OF_TAI];
    c_uint8_t       num_of_tac;
    gtp_node_t     *gnode;
} mme_sgw_t;

We would like to place lnode_t node that manages the list in mme_sgw_t and change the gnode to pointer.
Then, gnode add/remove interface should be changed as below.

rv = gtp_create_node(&sgw->gnode,
                                context_self()->parameter.no_ipv4,
                                context_self()->parameter.no_ipv6,
                                context_self()->parameter.prefer_ipv4);
rv = gtp_delete_node(sgw->gnode);

Yes. In doing so, we will not use the gtp_node_t used by SGW/PGW directly, but we will have to construct the following structure.

• On SGW

typedef struct _sgw_mme_t {
    lnode_t node ;

    gtp_node_t     *gnode;
} sgw_mme_t;

typedef struct _sgw_pgw_t {
    lnode_t node ;

    gtp_node_t     *gnode;
} sgw_pgw_t;

• On PGW

typedef struct _pgw_sgw_t {
    lnode_t node ;

    gtp_node_t     *gnode;
} pgw_sgw_t;

Then we need to implement add/remove function to manage the list of these structures in MME/SGW/PGW context.

What do you think about our second proposal? You can fix this whole part. If you do not have time to do that, we will start this re-factoring.

[TerryAlu]

I think the second proposal is practical! But I don't have time to make those changes currently. You can refactor it when you are available.

[acetcom]

To solve this problem, I created some of common gtp_node_t, namely mme_sgw_t and mme_pgw_t. I think it would be more efficient to keep all the rest. It means that I didn't created sgw_mme_t, sgw_pgw_t, and pgw_sgw_t.

And then, I modified the configuration file that you proposed.

#
# <SGW Selection Mode>
#
# o Round-Robin
#   (If you does not set the `selection_mode`, default mode is Round-Robin)
#
#   selection_mode: rr
#   gtpc:
#     addr: 127.0.0.2
#     addr: 127.0.2.2
#     addr: 127.0.4.2
#
# o SGW selection by eNodeB TAC
#
#   selection_mode: tac
#   gtpc:
#     - addr: 127.0.0.2
#       tac: 26000
#     - addr: 127.0.2.2
#       tac: [25000, 27000, 28000]
#

Thanks to you, we have also solved the problems of our structure and we can add one good feature for SGW selection.

I'm really appreciated for your work!

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ TerryAlu
❌ CaiYunZhan

---

CaiYunZhan seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}