diff --git a/db/boinc_db.cpp b/db/boinc_db.cpp
index 77f9c9951f..44c884d782 100644
--- a/db/boinc_db.cpp
+++ b/db/boinc_db.cpp
@@ -901,7 +901,8 @@ void DB_WORKUNIT::db_print(char* buf){
         "fileset_id=%lu, "
         "app_version_id=%ld, "
         "transitioner_flags=%d, "
-        "size_class=%d ",
+        "size_class=%d, "
+        "keywords='%s' ",
         create_time, appid,
         name, xml_doc, batch,
         rsc_fpops_est, rsc_fpops_bound, rsc_memory_bound, rsc_disk_bound,
@@ -921,7 +922,8 @@ void DB_WORKUNIT::db_print(char* buf){
         fileset_id,
         app_version_id,
         transitioner_flags,
-        size_class
+        size_class,
+        keywords
     );
 }
 
@@ -944,7 +946,8 @@ void DB_WORKUNIT::db_print_values(char* buf) {
         "%lu, "
         "%ld, "
         "%d, "
-        "%d)",
+        "%d, "
+        "'%s')",
         create_time, appid,
         name, xml_doc, batch,
         rsc_fpops_est, rsc_fpops_bound,
@@ -965,7 +968,8 @@ void DB_WORKUNIT::db_print_values(char* buf) {
         fileset_id,
         app_version_id,
         transitioner_flags,
-        size_class
+        size_class,
+        keywords
     );
 }
 
@@ -1005,6 +1009,7 @@ void DB_WORKUNIT::db_parse(MYSQL_ROW &r) {
     app_version_id = atol(r[i++]);
     transitioner_flags = atoi(r[i++]);
     size_class = atoi(r[i++]);
+    strcpy2(keywords, r[i++]);
 }
 
 void DB_CREDITED_JOB::db_print(char* buf){
@@ -2020,6 +2025,7 @@ void WORK_ITEM::parse(MYSQL_ROW& r) {
     wu.app_version_id = atol(r[i++]);
     wu.transitioner_flags = atoi(r[i++]);
     wu.size_class = atoi(r[i++]);
+    strcpy2(wu.keywords, r[i++]);
 }
 
 int DB_WORK_ITEM::enumerate(
diff --git a/db/boinc_db.h b/db/boinc_db.h
index 3d016981d1..86ace2a764 100644
--- a/db/boinc_db.h
+++ b/db/boinc_db.h
@@ -18,7 +18,14 @@
 #ifndef _BOINC_DB_
 #define _BOINC_DB_
 
-// Structures corresponding to database records.
+// Structures passed to and from DB queries.
+//
+// Mostly these correspond to DB tables, and inherit structs
+// defined in boinc_db_types.h
+// But some of them - TRANSITIONER_ITEM, STATE_COUNTS, SCHED_RESULT_ITEM, etc. -
+// combine the info from multiple tables (from joins)
+// or have subsets of table data.
+//
 // Some of these types have counterparts in client/types.h,
 // but don't be deceived - client and server have different variants.
 
diff --git a/db/boinc_db_types.h b/db/boinc_db_types.h
index 03795fb585..c129121ec9 100644
--- a/db/boinc_db_types.h
+++ b/db/boinc_db_types.h
@@ -480,9 +480,9 @@ struct WORKUNIT {
     int size_class;
         // -1 means none; encode this here so that transitioner
         // doesn't have to look up app
+    char keywords[256];
+        // keywords, as space-separated integers
 
-    // the following not used in the DB
-    char app_name[256];
     void clear();
     WORKUNIT(){clear();}
 };
diff --git a/db/schema.sql b/db/schema.sql
index 7ca2760ed8..404a633137 100644
--- a/db/schema.sql
+++ b/db/schema.sql
@@ -256,6 +256,7 @@ create table workunit (
     app_version_id          integer         not null,
     transitioner_flags      tinyint         not null,
     size_class              smallint        not null default -1,
+    keywords                varchar(254)    not null,
     primary key (id)
 ) engine=InnoDB;
 
diff --git a/html/ops/db_update.php b/html/ops/db_update.php
index 8a2c960aa8..e0414d7c30 100644
--- a/html/ops/db_update.php
+++ b/html/ops/db_update.php
@@ -1049,6 +1049,12 @@ function update_6_13_2017() {
     ");
 }
 
+function update_7_21_2017() {
+    do_query("alter table workunit
+        add column keywords varchar(254) not null
+    ");
+}
+
 // Updates are done automatically if you use "upgrade".
 //
 // If you need to do updates manually,
@@ -1098,6 +1104,7 @@ $db_updates = array (
     array(27015, "update_2_17_2017"),
     array(27016, "update_3_17_2017"),
     array(27017, "update_6_13_2017"),
+    array(27018, "update_7_21_2017"),
 );
 
 ?>
diff --git a/lib/keyword.cpp b/lib/keyword.cpp
index b08fd8265c..3abe47cbb8 100644
--- a/lib/keyword.cpp
+++ b/lib/keyword.cpp
@@ -18,7 +18,6 @@
 // utility functions for keywords
 
 #include <stdio.h>
-#include <algorithm>
 
 #include "parse.h"
 #include "keyword.h"
@@ -54,15 +53,13 @@ void USER_KEYWORDS::write(FILE* f) {
     fprintf(f, "</user_keywords>\n");
 }
 
-double keyword_score(USER_KEYWORDS& user_keywords, JOB_KEYWORDS& job_keywords) {
-    double score = 0;
-    for (unsigned int i=0; i<job_keywords.ids.size(); i++) {
-        int jk = job_keywords.ids[i];
-        if (std::find(user_keywords.yes.begin(), user_keywords.yes.end(), jk) != user_keywords.yes.end()) {
-            score += 1;
-        } else if (std::find(user_keywords.no.begin(), user_keywords.no.end(), jk) != user_keywords.no.end()) {
-            return -1;
-        }
+void JOB_KEYWORDS::parse_str(char* buf) {
+    char* p = strtok(buf, " ");
+    if (!p) return;
+    ids.push_back(atoi(p));
+    while (true) {
+        p = strtok(NULL, " ");
+        if (!p) break;
+        ids.push_back(atoi(p));
     }
-    return score;
 }
diff --git a/lib/keyword.h b/lib/keyword.h
index 5e4f53b4ad..6532bb42a6 100644
--- a/lib/keyword.h
+++ b/lib/keyword.h
@@ -17,6 +17,9 @@
 
 // utility classes for keywords
 
+#ifndef BOINC_KEYWORD_H
+#define BOINC_KEYWORD_H
+
 #include <vector>
 #include "parse.h"
 
@@ -24,12 +27,12 @@ struct USER_KEYWORDS {
     std::vector<int> yes;
     std::vector<int> no;
     int parse(XML_PARSER&);
-    void clear() {
+    inline void clear() {
         yes.clear();
         no.clear();
     }
     void write(FILE*);
-    bool empty() {
+    inline bool empty() {
         return yes.empty() && no.empty();
     }
 };
@@ -38,6 +41,12 @@ struct JOB_KEYWORDS {
     std::vector<int> ids;
     void parse_str(char*);
         // parse space-separated list
+    inline bool empty() {
+        return ids.empty();
+    }
+    inline void clear() {
+        ids.clear();
+    }
 };
 
-extern double keyword_score(USER_KEYWORDS&, JOB_KEYWORDS&);
+#endif
diff --git a/lib/str_util.h b/lib/str_util.h
index 840b2870b1..91751e3965 100644
--- a/lib/str_util.h
+++ b/lib/str_util.h
@@ -42,6 +42,10 @@ extern char* precision_time_to_string(double);
 extern void secs_to_hmsf(double, char*);
 extern std::string timediff_format(double);
 
+inline bool empty(char* p) {
+    return p[0] == 0;
+}
+
 inline bool ends_with(std::string const& s, std::string const& suffix) {
     return
         s.size()>=suffix.size() &&
diff --git a/sched/Makefile.am b/sched/Makefile.am
index ccb66c2c4a..8de97c9f39 100644
--- a/sched/Makefile.am
+++ b/sched/Makefile.am
@@ -167,6 +167,7 @@ cgi_sources = \
     sched_customize.cpp \
     sched_files.cpp \
     sched_hr.cpp \
+    sched_keyword.cpp \
     sched_limit.cpp \
     sched_locality.cpp \
     sched_main.cpp \
diff --git a/sched/sched_check.cpp b/sched/sched_check.cpp
index a34a6e398e..b1ae195b8d 100644
--- a/sched/sched_check.cpp
+++ b/sched/sched_check.cpp
@@ -375,9 +375,10 @@ int wu_is_infeasible_fast(
 
 // Do checks that require DB access for whether we can send this job,
 // and return:
-// 0 if OK to send
-// 1 if can't send to this host
-// 2 if can't send to ANY host
+// CHECK_OK if OK to send
+// CHECK_NO_HOST if can't send to this host
+// CHECK_NO_ANY if can't send to ANY host
+//  e.g. WU error mask is nonzero
 //
 int slow_check(
     WU_RESULT& wu_result,       // the job cache entry.
@@ -402,7 +403,7 @@ int slow_check(
             log_messages.printf(MSG_CRITICAL,
                 "send_work: can't get result count (%s)\n", boincerror(retval)
             );
-            return 1;
+            return CHECK_NO_HOST;
         } else {
             if (n>0) {
                 if (config.debug_send_job) {
@@ -411,7 +412,7 @@ int slow_check(
                         g_reply->user.id, n, wu.id
                     );
                 }
-                return 1;
+                return CHECK_NO_HOST;
             }
         }
     } else if (config.one_result_per_host_per_wu) {
@@ -426,7 +427,7 @@ int slow_check(
             log_messages.printf(MSG_CRITICAL,
                 "send_work: can't get result count (%s)\n", boincerror(retval)
             );
-            return 1;
+            return CHECK_NO_HOST;
         } else {
             if (n>0) {
                 if (config.debug_send_job) {
@@ -435,7 +436,7 @@ int slow_check(
                         g_reply->host.id, n, wu.id
                     );
                 }
-                return 1;
+                return CHECK_NO_HOST;
             }
         }
     }
@@ -454,13 +455,13 @@ int slow_check(
             log_messages.printf(MSG_CRITICAL,
                 "can't get fields for [WU#%lu]: %s\n", db_wu.id, boincerror(retval)
             );
-            return 1;
+            return CHECK_NO_HOST;
         }
 
         // check wu.error_mask
         //
         if (vals[2] != 0) {
-            return 2;
+            return CHECK_NO_ANY;
         }
 
         if (app_hr_type(*app)) {
@@ -477,7 +478,7 @@ int slow_check(
                 // are processed first.
                 //
                 wu_result.infeasible_count++;
-                return 1;
+                return CHECK_NO_HOST;
             }
         }
         if (app->homogeneous_app_version) {
@@ -491,11 +492,11 @@ int slow_check(
                     );
                 }
                 wu_result.infeasible_count++;
-                return 1;
+                return CHECK_NO_HOST;
             }
         }
     }
-    return 0;
+    return CHECK_OK;
 }
 
 // Check for pathological conditions that mean
diff --git a/sched/sched_check.h b/sched/sched_check.h
index 59fca0f165..bea673c9a2 100644
--- a/sched/sched_check.h
+++ b/sched/sched_check.h
@@ -22,6 +22,8 @@
 #include "sched_shmem.h"
 #include "sched_types.h"
 
+// return values from wu_is_infeasible_fast()
+//
 #define INFEASIBLE_MEM          1
 #define INFEASIBLE_DISK         2
 #define INFEASIBLE_CPU          3
@@ -42,7 +44,15 @@ extern int wu_is_infeasible_fast(
     int res_server_state, int res_priority, double res_report_deadline,
     APP&, BEST_APP_VERSION&
 );
+
+// return values from slow_check()
+//
+#define CHECK_OK        0
+#define CHECK_NO_HOST   1
+#define CHECK_NO_ANY    2
+
 extern int slow_check(WU_RESULT&, APP*, BEST_APP_VERSION*);
+
 extern bool result_still_sendable(DB_RESULT& result, WORKUNIT& wu);
 extern bool app_not_selected(int appid);
  
diff --git a/sched/sched_config.cpp b/sched/sched_config.cpp
index 6a7bb4f69d..7b7c595fbd 100644
--- a/sched/sched_config.cpp
+++ b/sched/sched_config.cpp
@@ -298,6 +298,7 @@ int SCHED_CONFIG::parse(FILE* f) {
         if (xp.parse_double("version_select_random_factor", version_select_random_factor)) continue;
         if (xp.parse_double("maintenance_delay", maintenance_delay)) continue;
         if (xp.parse_bool("credit_by_app", credit_by_app)) continue;
+        if (xp.parse_bool("keyword_sched", keyword_sched)) continue;
 
         //////////// SCHEDULER LOG FLAGS /////////
 
@@ -308,6 +309,7 @@ int SCHED_CONFIG::parse(FILE* f) {
         if (xp.parse_bool("debug_edf_sim_workload", debug_edf_sim_workload)) continue;
         if (xp.parse_bool("debug_fcgi", debug_fcgi)) continue;
         if (xp.parse_bool("debug_handle_results", debug_handle_results)) continue;
+        if (xp.parse_bool("debug_keyword", debug_keyword)) continue;
         if (xp.parse_bool("debug_locality", debug_locality)) continue;
         if (xp.parse_bool("debug_locality_lite", debug_locality_lite)) continue;
         if (xp.parse_bool("debug_prefs", debug_prefs)) continue;
diff --git a/sched/sched_config.h b/sched/sched_config.h
index 49385e6abf..75757db866 100644
--- a/sched/sched_config.h
+++ b/sched/sched_config.h
@@ -180,6 +180,8 @@ struct SCHED_CONFIG {
         // to calculate projected_flops when choosing version.
     bool credit_by_app;
         // store per-app credit info in credit_user and credit_team
+    bool keyword_sched;
+        // score jobs based on keywords
 
     // time intervals
     double maintenance_delay;
@@ -194,6 +196,7 @@ struct SCHED_CONFIG {
     bool debug_fcgi;
     bool debug_client_files;        // stuff related to sticky files on client
     bool debug_handle_results;
+    bool debug_keyword;
     bool debug_locality;            // locality scheduling
     bool debug_locality_lite;       // locality scheduling Lite
     bool debug_prefs;
diff --git a/sched/sched_keyword.cpp b/sched/sched_keyword.cpp
index 9bb4a4ffe1..6ef935dc84 100644
--- a/sched/sched_keyword.cpp
+++ b/sched/sched_keyword.cpp
@@ -16,23 +16,32 @@
 // along with BOINC.  If not, see <http://www.gnu.org/licenses/>.
 
 // code related to keyword-based job scoring
+//
+// A job's keywords are stored in workunit.keywords as a char string.
+// We don't want to parse that every time we score the job,
+// so we maintain a list of JOB_KEYWORDs paralleling the job array.
 
 #include <algorithm>
+#include <iterator>
 
 #include "sched_main.h"
 #include "keyword.h"
 
 JOB_KEYWORDS *job_keywords_array;
 
+// compute the score increment for the given job and user keywords
+// (or -1 if the keywords are incompatible)
+//
 double keyword_score_aux(
-    USER_KEYWORDS& user_keywords, JOB_KEYWORDS& job_keywords
+    USER_KEYWORDS& uks, JOB_KEYWORDS& jks
 ) {
     double score = 0;
-    for (unsigned int i=0; i<job_keywords.ids.size(); i++) {
-        int jk = job_keywords.ids[i];
-        if (std::find(user_keywords.yes.begin(), user_keywords.yes.end(), jk) != user_keywords.yes.end()) {
+
+    for (unsigned int i=0; i<jks.ids.size(); i++) {
+        int jk = jks.ids[i];
+        if (std::find(uks.yes.begin(), uks.yes.end(), jk) != uks.yes.end()) {
             score += 1;
-        } else if (std::find(user_keywords.no.begin(), user_keywords.no.end(), jk) != user_keywords.no.end()) {
+        } else if (std::find(uks.no.begin(), uks.no.end(), jk) != uks.no.end()) {
             return -1;
         }
     }
@@ -76,6 +85,8 @@ void keyword_sched_remove_job(int i) {
     job_keywords_array[i].clear();
 }
 
+// called at CGI start to initialize job keyword array
+//
 void keyword_sched_init() {
     job_keywords_array = new JOB_KEYWORDS[ssp->max_wu_results];
 }
diff --git a/sched/sched_keyword.h b/sched/sched_keyword.h
index ed61fb14d1..5814e575b1 100644
--- a/sched/sched_keyword.h
+++ b/sched/sched_keyword.h
@@ -18,7 +18,11 @@
 #ifndef BOINC_SCHED_KEYWORD_H
 #define BOINC_SCHED_KEYWORD_H
 
+// see sched_keywork.cpp
+
 extern double keyword_score(int);
 extern void keyword_sched_remove_job(int index);
+extern void keyword_sched_init();
+
 
 #endif
diff --git a/sched/sched_main.cpp b/sched/sched_main.cpp
index c5584df507..98efcd09fc 100644
--- a/sched/sched_main.cpp
+++ b/sched/sched_main.cpp
@@ -61,6 +61,7 @@
 #include "handle_request.h"
 #include "sched_config.h"
 #include "sched_files.h"
+#include "sched_keyword.h"
 #include "sched_msgs.h"
 #include "sched_types.h"
 #include "sched_util.h"
@@ -543,6 +544,9 @@ int main(int argc, char** argv) {
         send_message("Server error: can't attach shared memory", config.maintenance_delay);
         goto done;
     }
+    if (config.keyword_sched) {
+        keyword_sched_init();
+    }
 
     if (strlen(config.debug_req_reply_dir)) {
         struct stat statbuf;
diff --git a/sched/sched_score.cpp b/sched/sched_score.cpp
index dea57c0987..91c14697fa 100644
--- a/sched/sched_score.cpp
+++ b/sched/sched_score.cpp
@@ -31,6 +31,7 @@
 #include "sched_check.h"
 #include "sched_config.h"
 #include "sched_hr.h"
+#include "sched_keyword.h"
 #include "sched_main.h"
 #include "sched_msgs.h"
 #include "sched_send.h"
@@ -54,7 +55,8 @@ static int get_size_class(APP& app, double es) {
 // Also do some initial screening,
 // and return false if can't send the job to host
 //
-bool JOB::get_score(WU_RESULT& wu_result) {
+bool JOB::get_score(int array_index) {
+    WU_RESULT& wu_result = ssp->wu_results[array_index];
     score = 0;
 
     if (!app->beta && wu_result.need_reliable) {
@@ -130,6 +132,14 @@ bool JOB::get_score(WU_RESULT& wu_result) {
 
     score += wu_result.res_priority;
 
+    if (config.keyword_sched) {
+        double x = keyword_score(array_index);
+        if (x < 0) {
+            return false;
+        }
+        score += x;
+    }
+
     if (config.debug_send_job) {
         log_messages.printf(MSG_NORMAL,
             "[send_job]: score %f for result %lu\n", score, wu_result.resultid
@@ -215,7 +225,7 @@ void send_work_score_type(int rt) {
 
         job.index = i;
         job.result_id = wu_result.resultid;
-        if (!job.get_score(wu_result)) {
+        if (!job.get_score(i)) {
             if (config.debug_send_job) {
                 log_messages.printf(MSG_NORMAL,
                     "[send_job] [RESULT#%lu] get_score() returned false\n",
@@ -298,11 +308,14 @@ void send_work_score_type(int rt) {
         sema_locked = false;
 
         switch (slow_check(wu_result, job.app, job.bavp)) {
-        case 1:
+        case CHECK_NO_HOST:
             wu_result.state = WR_STATE_PRESENT;
             break;
-        case 2:
+        case CHECK_NO_ANY:
             wu_result.state = WR_STATE_EMPTY;
+            if (config.keyword_sched) {
+                keyword_sched_remove_job(job.index);
+            }
             break;
         default:
             // slow_check() refreshes fields of wu_result.workunit;
@@ -315,6 +328,7 @@ void send_work_score_type(int rt) {
             // (since otherwise feeder might overwrite it)
             //
             wu_result.state = WR_STATE_EMPTY;
+            keyword_sched_remove_job(job.index);
 
             // reread result from DB, make sure it's still unsent
             // TODO: from here to end of add_result_to_reply()
diff --git a/sched/sched_score.h b/sched/sched_score.h
index 4a05857429..693a55ac5d 100644
--- a/sched/sched_score.h
+++ b/sched/sched_score.h
@@ -24,7 +24,7 @@ struct JOB {
     APP* app;
     BEST_APP_VERSION* bavp;
 
-    bool get_score(WU_RESULT&);
+    bool get_score(int);
 };
 
 extern void send_work_score();
diff --git a/sched/sched_types.cpp b/sched/sched_types.cpp
index 0dd654100f..12b26b0899 100644
--- a/sched/sched_types.cpp
+++ b/sched/sched_types.cpp
@@ -1373,6 +1373,7 @@ int HOST::parse(XML_PARSER& xp) {
     p_ncpus = 1;
     double dtemp;
     string stemp;
+    int x;
     while (!xp.get_tag()) {
         if (xp.match_tag("/host_info")) return 0;
         if (xp.parse_int("timezone", timezone)) continue;
@@ -1404,6 +1405,10 @@ int HOST::parse(XML_PARSER& xp) {
             continue;
         }
 
+        // unused fields
+        //
+        if (xp.parse_int("n_usable_coprocs", x)) continue;
+
         // parse deprecated fields to avoid error messages
         //
         if (xp.parse_double("p_calculated", dtemp)) continue;
diff --git a/tools/create_work.cpp b/tools/create_work.cpp
index 028e13bf62..c443b3c984 100644
--- a/tools/create_work.cpp
+++ b/tools/create_work.cpp
@@ -62,6 +62,7 @@ void usage() {
         "   [ -d n ]\n"
         "   [ --delay_bound x ]\n"
         "   [ --hr_class n ]\n"
+        "   [ --keywords 'n1 n2 ...' ]\n"
         "   [ --max_error_results n ]\n"
         "   [ --max_success_results n ]\n"
         "   [ --max_total_results n ]\n"
@@ -336,6 +337,8 @@ int main(int argc, char** argv) {
             verbose = true;
         } else if (arg(argv, i, "continue_on_error")) {
             continue_on_error = true;
+        } else if (arg(argv, i, "keywords")) {
+            strcpy(jd.wu.keywords, argv[++i]);
         } else {
             if (!strncmp("-", argv[i], 1)) {
                 fprintf(stderr, "create_work: bad argument '%s'\n", argv[i]);