diff --git a/README.spdk b/README.spdk
new file mode 100644
index 000000000..736f904f0
--- /dev/null
+++ b/README.spdk
@@ -0,0 +1,37 @@
+========================================================================
+ README for mTCP development with spdk
+========================================================================
+
+1. Why mTCP changes
+
+mTCP is a highly scalable user-level TCP stack for multicore systems.
+It requires per-thread architecture and starts "RunMainLoop" for each
+thread with dead-loop handling.
+
+However, spdk will register poller per thread to handle receiving and
+events and so it requires mtcp to supply a separated recv/send handler.
+It's necessary to split RunMainLoop to two APIs: RunMainLoop and
+mtcp_run_instance. Original mTCP apps still use RunMainLoop and it's not
+affected and spdk will use mtcp_run_instance for its poller.
+
+2. How to enable spdk support
+
+Add option "--enable-spdk" to choose spdk support enable or not.
+Also introduce mtcp_crossbuild.sh for reference.
+
+3. SPDK changes
+
+For SPDK, it needs more changes. 
+
+- Add mtcp sock interface. Current now it only supports posix/vpp sock
+  interface. Just follow the framework to add mtcp sock interface.
+- Change connection schedule mode to support mtcp per-thread arch.
+
+Now upstream is ongoing.
+
+========================================================================
+
+                   Contact: mtcp-user at list.ndsl.kaist.edu
+                             April 2, 2015. 
+                 EunYoung Jeong <notav at ndsl.kaist.edu>
+		 M. Asim Jamshed <ajamshed at ndsl.kaist.edu>
diff --git a/apps/example/Makefile.in b/apps/example/Makefile.in
index 23544f309..e9086bebd 100644
--- a/apps/example/Makefile.in
+++ b/apps/example/Makefile.in
@@ -7,9 +7,13 @@ PS=@PSIO@
 NETMAP=@NETMAP@
 ONVM=@ONVM@
 CFLAGS=@CFLAGS@
+LDFLAGS=@LDFLAGS@
+
+# If ARCH is not defined, retrive from system
+ARCH ?= $(shell uname -m)
 
 # Add arch-specific optimization
-ifeq ($(shell uname -m),x86_64)
+ifeq ($(ARCH),x86_64)
 LIBS += -m64
 endif
 
@@ -44,7 +48,7 @@ endif
 ifeq ($(DPDK),1)
 DPDK_MACHINE_LINKER_FLAGS=$${RTE_SDK}/$${RTE_TARGET}/lib/ldflags.txt
 DPDK_MACHINE_LDFLAGS=$(shell cat ${DPDK_MACHINE_LINKER_FLAGS})
-LIBS += -g -O3 -pthread -lrt -march=native ${MTCP_FLD}/lib/libmtcp.a -lnuma -lmtcp -lpthread -lrt -ldl -lgmp -L${RTE_SDK}/${RTE_TARGET}/lib ${DPDK_MACHINE_LDFLAGS}
+LIBS += -g -O3 -pthread -lrt -march=native ${MTCP_FLD}/lib/libmtcp.a -lnuma -lmtcp -lpthread -lrt -ldl -lgmp -L${RTE_SDK}/${RTE_TARGET}/lib ${DPDK_MACHINE_LDFLAGS} ${LDFLAGS}
 endif
 
 # onvm-specific variables
diff --git a/configure.ac b/configure.ac
index 69f60c584..48d6fa680 100644
--- a/configure.ac
+++ b/configure.ac
@@ -63,6 +63,17 @@ AC_FUNC_MMAP
 AC_CHECK_FUNC([clock_gettime],,AC_MSG_ERROR([librt library is missing]))
 AC_CHECK_FUNCS([bzero getpagesize gettimeofday memmove memset munmap select socket strchr strerror strstr strtol],,AC_MSG_ERROR([glibc library is missing]))
 
+# Reset SPDK to 0
+AC_SUBST(SPDK, 0)
+
+dnl Example of default-disabled feature
+AC_ARG_ENABLE([spdk],
+	AS_HELP_STRING([--enable-spdk], [Enable SPDK Support]))
+
+AS_IF([test "x$enable_spdk" = "xyes"], [
+	    AC_SUBST(SPDK, 1)
+])
+
 # Reset DPDK to 0
 AC_SUBST(DPDK, 0)
 # Reset enforcement value
diff --git a/dpdk-iface-kmod/Makefile b/dpdk-iface-kmod/Makefile
index 93da3030c..a84732157 100644
--- a/dpdk-iface-kmod/Makefile
+++ b/dpdk-iface-kmod/Makefile
@@ -6,9 +6,12 @@ endif
 ifeq ($(RTE_TARGET),)
 $(error "Please define RTE_TARGET environment variable")
 endif
+
+RTE_KERNELDIR ?= /lib/modules/$(shell uname -r)/build/
+
 #-------------------------------------------------------------------------#
 include $(RTE_SDK)/mk/rte.vars.mk
-CC=gcc
+CC=$(CROSS)gcc
 obj-m=dpdk_iface.o
 DPDK_MACHINE_LINKER_FLAGS=$${RTE_SDK}/$${RTE_TARGET}/lib/ldflags.txt
 DPDK_MACHINE_LDFLAGS=$(shell cat ${DPDK_MACHINE_LINKER_FLAGS})
@@ -25,7 +28,7 @@ else
 endif
 #-------------------------------------------------------------------------#
 all: dpdk_iface.c $(appname) $(appname).c
-	make -C /lib/modules/$(shell uname -r)/build/ M=$(PWD) modules
+	make ARCH=${ARCH} CROSS_COMPILE=${CROSS} -C ${RTE_KERNELDIR} M=$(PWD) modules
 
 $(appname): $(appname).c
 	$(MSG) "  CC $<"
@@ -34,7 +37,7 @@ $(appname): $(appname).c
 	-L$(DPDK_LIB) ${DPDK_MACHINE_LDFLAGS} -lpthread
 
 clean:
-	make -C /lib/modules/$(shell uname -r)/build/ M=$(PWD) clean
+	make -C ${RTE_KERNELDIR} M=$(PWD) clean
 	$(MSG) "  CLEAN  $(appname)"
 	$(HIDE) rm -rf *~ *.o *.ko dpdk_iface_main
 
diff --git a/dpdk-iface-kmod/dpdk_iface_main.c b/dpdk-iface-kmod/dpdk_iface_main.c
index 143cea4d1..89f579a90 100644
--- a/dpdk-iface-kmod/dpdk_iface_main.c
+++ b/dpdk-iface-kmod/dpdk_iface_main.c
@@ -22,7 +22,11 @@
 typedef struct {
 	PciDevice pd;
 	struct rte_eth_dev_info dev_details;
+#if RTE_VERSION < RTE_VERSION_NUM(19, 8, 0, 0)
 	struct ether_addr ports_eth_addr;
+#else
+	struct rte_ether_addr ports_eth_addr;
+#endif
 } DevInfo;
 
 static DevInfo di[RTE_MAX_ETHPORTS];	
@@ -260,7 +264,11 @@ main(int argc, char **argv)
 	ret = rte_eal_init(rte_argc, rte_argv);
 
 	/* get total count of detected ethernet ports */
+#if RTE_VERSION < RTE_VERSION_NUM(18, 5, 0, 0)
 	num_devices = rte_eth_dev_count();
+#else
+	num_devices = rte_eth_dev_count_avail();
+#endif
 	if (num_devices == 0) {
 		fprintf(stderr, "No Ethernet port detected!\n");
 		exit(EXIT_FAILURE);
diff --git a/mtcp/src/Makefile.in b/mtcp/src/Makefile.in
index c703ae213..04936f1b9 100644
--- a/mtcp/src/Makefile.in
+++ b/mtcp/src/Makefile.in
@@ -4,6 +4,7 @@
 
 ### TARGET ###
 PS=@PSIO@
+SPDK=@SPDK@
 DPDK=@DPDK@
 ENFORCE_RX_IDLE=@ENFORCE_RX_IDLE@
 NETMAP=@NETMAP@
@@ -19,7 +20,10 @@ MTCP_HDR = mtcp_api.h mtcp_epoll.h
 GCC=@CC@
 
 ### FLAGS ###
-ifeq ($(shell uname -m),x86_64)
+
+# If ARCH is not defined, retrive from system
+ARCH ?= $(shell uname -m)
+ifeq ($(ARCH),x86_64)
 GCC_OPT = -m64
 else
 GCC_OPT =
@@ -50,6 +54,10 @@ GCC_OPT += -DNDEBUG -g -O3 -DNETSTAT -DINFO -DDBGERR -DDBGCERR
 #GCC_OPT += -DNDEBUG -g -DNETSTAT -DINFO -DDBGERR -DDBGCERR
 GCC_OPT += $(DBG_OPT)
 
+ifeq ($(SPDK),1)
+GCC_OPT += -DENABLE_SPDK
+endif
+
 ifeq ($(LRO),1)
 GCC_OPT += -DENABLELRO
 endif
diff --git a/mtcp/src/api.c b/mtcp/src/api.c
index c53714bcf..3429f4476 100644
--- a/mtcp/src/api.c
+++ b/mtcp/src/api.c
@@ -1284,7 +1284,7 @@ mtcp_recv(mctx_t mctx, int sockid, char *buf, size_t len, int flags)
         return ret;
 }
 /*----------------------------------------------------------------------------*/
-inline ssize_t
+ssize_t
 mtcp_read(mctx_t mctx, int sockid, char *buf, size_t len)
 {
 	return mtcp_recv(mctx, sockid, buf, len, 0);
diff --git a/mtcp/src/core.c b/mtcp/src/core.c
index 7ce22c805..93b93634a 100644
--- a/mtcp/src/core.c
+++ b/mtcp/src/core.c
@@ -750,123 +750,140 @@ InterruptApplication(mtcp_manager_t mtcp)
 	}
 }
 /*----------------------------------------------------------------------------*/
+#ifndef ENABLE_SPDK
 static void 
 RunMainLoop(struct mtcp_thread_context *ctx)
 {
 	mtcp_manager_t mtcp = ctx->mtcp_manager;
-	int i;
-	int recv_cnt;
-	int rx_inf, tx_inf;
-	struct timeval cur_ts = {0};
-	uint32_t ts, ts_prev;
-	int thresh;
 
-	gettimeofday(&cur_ts, NULL);
-	TRACE_DBG("CPU %d: mtcp thread running.\n", ctx->cpu);
 
-	ts = ts_prev = 0;
 	while ((!ctx->done || mtcp->flow_cnt) && !ctx->exit) {
-		
-		STAT_COUNT(mtcp->runstat.rounds);
-		recv_cnt = 0;
-			
-		gettimeofday(&cur_ts, NULL);
-		ts = TIMEVAL_TO_TS(&cur_ts);
-		mtcp->cur_ts = ts;
-
-		for (rx_inf = 0; rx_inf < CONFIG.eths_num; rx_inf++) {
-
-			static uint16_t len;
-			static uint8_t *pktbuf;
-			recv_cnt = mtcp->iom->recv_pkts(ctx, rx_inf);
-			STAT_COUNT(mtcp->runstat.rounds_rx_try);
+		mtcp_run_instance(ctx);
+	}
 
-			for (i = 0; i < recv_cnt; i++) {
-				pktbuf = mtcp->iom->get_rptr(mtcp->ctx, rx_inf, i, &len);
-				if (pktbuf != NULL)
-					ProcessPacket(mtcp, rx_inf, ts, pktbuf, len);
-#ifdef NETSTAT
-				else
-					mtcp->nstat.rx_errors[rx_inf]++;
+#if TESTING
+	DestroyRemainingFlows(mtcp);
 #endif
-			}
-		}
-		STAT_COUNT(mtcp->runstat.rounds_rx);
 
-		/* interaction with application */
-		if (mtcp->flow_cnt > 0) {
-			
-			/* check retransmission timeout and timewait expire */
-#if 0
-			thresh = (int)mtcp->flow_cnt / (TS_TO_USEC(PER_STREAM_TCHECK));
-			assert(thresh >= 0);
-			if (thresh == 0)
-				thresh = 1;
-			if (recv_cnt > 0 && thresh > recv_cnt)
-				thresh = recv_cnt;
+	TRACE_DBG("MTCP thread %d out of main loop.\n", ctx->cpu);
+	/* flush logs */
+	flush_log_data(mtcp);
+	TRACE_DBG("MTCP thread %d flushed logs.\n", ctx->cpu);
+	InterruptApplication(mtcp);
+	TRACE_INFO("MTCP thread %d finished.\n", ctx->cpu);
+}
 #endif
-			thresh = CONFIG.max_concurrency;
-
-			/* Eunyoung, you may fix this later 
-			 * if there is no rcv packet, we will send as much as possible
-			 */
-			if (thresh == -1)
-				thresh = CONFIG.max_concurrency;
-
-			CheckRtmTimeout(mtcp, ts, thresh);
-			CheckTimewaitExpire(mtcp, ts, CONFIG.max_concurrency);
-
-			if (CONFIG.tcp_timeout > 0 && ts != ts_prev) {
-				CheckConnectionTimeout(mtcp, ts, thresh);
-			}
-		}
-
-		/* if epoll is in use, flush all the queued events */
-		if (mtcp->ep) {
-			FlushEpollEvents(mtcp, ts);
-		}
-
-		if (mtcp->flow_cnt > 0) {
-			/* hadnle stream queues  */
-			HandleApplicationCalls(mtcp, ts);
-		}
-
-		WritePacketsToChunks(mtcp, ts);
 
-		/* send packets from write buffer */
-		/* send until tx is available */
-		for (tx_inf = 0; tx_inf < CONFIG.eths_num; tx_inf++) {
-			mtcp->iom->send_pkts(ctx, tx_inf);
-		}
+RTE_DEFINE_PER_LCORE(uint32_t , ts_prev) = 0;
 
-		if (ts != ts_prev) {
-			ts_prev = ts;
-			if (ctx->cpu == mtcp_master) {
-				ARPTimer(mtcp, ts);
+int mtcp_run_instance(void *tmp)
+{
+        struct mtcp_thread_context *ctx= (struct mtcp_thread_context *)tmp;
+        mtcp_manager_t mtcp = ctx->mtcp_manager;
+        int i;
+        int recv_cnt;
+        int rx_inf, tx_inf;
+        struct timeval cur_ts = {0};
+        int thresh;
+        uint32_t ts=0;
+
+        if ((!ctx->done || mtcp->flow_cnt) && !ctx->exit) {
+
+                STAT_COUNT(mtcp->runstat.rounds);
+                recv_cnt = 0;
+
+                gettimeofday(&cur_ts, NULL);
+                ts = TIMEVAL_TO_TS(&cur_ts);
+                mtcp->cur_ts = ts;
+
+                for (rx_inf = 0; rx_inf < CONFIG.eths_num; rx_inf++) {
+
+                        static uint16_t len;
+                        static uint8_t *pktbuf;
+                        recv_cnt = mtcp->iom->recv_pkts(ctx, rx_inf);
+                        STAT_COUNT(mtcp->runstat.rounds_rx_try);
+
+                        for (i = 0; i < recv_cnt; i++) {
+                                pktbuf = mtcp->iom->get_rptr(mtcp->ctx, rx_inf, i, &len);
+                                if (pktbuf != NULL)
+                                        ProcessPacket(mtcp, rx_inf, ts, pktbuf, len);
 #ifdef NETSTAT
-				PrintNetworkStats(mtcp, ts);
+                                else
+                                        mtcp->nstat.rx_errors[rx_inf]++;
 #endif
-			}
-		}
+                        }
+                }
+                STAT_COUNT(mtcp->runstat.rounds_rx);
 
-		mtcp->iom->select(ctx);
+                /* interaction with application */
+                if (mtcp->flow_cnt > 0) {
 
-		if (ctx->interrupt) {
-			InterruptApplication(mtcp);
-		}
-	}
+                        /* check retransmission timeout and timewait expire */
+#if 0
+                        thresh = (int)mtcp->flow_cnt / (TS_TO_USEC(PER_STREAM_TCHECK));
+                        assert(thresh >= 0);
+                        if (thresh == 0)
+                                thresh = 1;
+                        if (recv_cnt > 0 && thresh > recv_cnt)
+                                thresh = recv_cnt;
+#endif
+                        thresh = CONFIG.max_concurrency;
+
+                        /* Eunyoung, you may fix this later
+                         * if there is no rcv packet, we will send as much as possible
+                         */
+                        if (thresh == -1)
+                                thresh = CONFIG.max_concurrency;
+
+                        CheckRtmTimeout(mtcp, ts, thresh);
+                        CheckTimewaitExpire(mtcp, ts, CONFIG.max_concurrency);
+
+                        if (CONFIG.tcp_timeout > 0 && ts != RTE_PER_LCORE(ts_prev)) {
+                                CheckConnectionTimeout(mtcp, ts, thresh);
+                        }
+                }
+
+                /* if epoll is in use, flush all the queued events */
+                if (mtcp->ep) {
+                        FlushEpollEvents(mtcp, ts);
+                }
+
+                if (mtcp->flow_cnt > 0) {
+                        /* hadnle stream queues  */
+                        HandleApplicationCalls(mtcp, ts);
+                }
+
+                WritePacketsToChunks(mtcp, ts);
+
+                /* send packets from write buffer */
+                /* send until tx is available */
+                for (tx_inf = 0; tx_inf < CONFIG.eths_num; tx_inf++) {
+                        mtcp->iom->send_pkts(ctx, tx_inf);
+                }
+
+                if (ts != RTE_PER_LCORE(ts_prev)) {
+                        RTE_PER_LCORE(ts_prev) = ts;
+                        if (ctx->cpu == mtcp_master) {
+                                ARPTimer(mtcp, ts);
+#ifndef ENABLE_SPDK
+#ifdef NETSTAT
 
-#if TESTING
-	DestroyRemainingFlows(mtcp);
+                                PrintNetworkStats(mtcp, ts);
+#endif
 #endif
+                        }
+                }
 
-	TRACE_DBG("MTCP thread %d out of main loop.\n", ctx->cpu);
-	/* flush logs */
-	flush_log_data(mtcp);
-	TRACE_DBG("MTCP thread %d flushed logs.\n", ctx->cpu);
-	InterruptApplication(mtcp);
-	TRACE_INFO("MTCP thread %d finished.\n", ctx->cpu);
+                mtcp->iom->select(ctx);
+
+                if (ctx->interrupt) {
+                        InterruptApplication(mtcp);
+                }
+        }
+
+        return 0;
 }
+
 /*----------------------------------------------------------------------------*/
 struct mtcp_sender *
 CreateMTCPSender(int ifidx)
@@ -1075,7 +1092,7 @@ InitializeMTCPManager(struct mtcp_thread_context* ctx)
 	return mtcp;
 }
 /*----------------------------------------------------------------------------*/
-static void *
+void *
 MTCPRunThread(void *arg)
 {
 	mctx_t mctx = (mctx_t)arg;
@@ -1152,7 +1169,7 @@ MTCPRunThread(void *arg)
 	fprintf(stderr, "CPU %d: initialization finished.\n", cpu);
 	
 	sem_post(&g_init_sem[ctx->cpu]);
-
+#ifndef ENABLE_SPDK
 	/* start the main loop */
 	RunMainLoop(ctx);
 
@@ -1164,8 +1181,20 @@ MTCPRunThread(void *arg)
 	DestroyHashtable(g_mtcp[cpu]->listeners);
 	
 	TRACE_DBG("MTCP thread %d finished.\n", ctx->cpu);
-	
 	return 0;
+#else
+	sem_wait(&g_init_sem[ctx->cpu]);
+	sem_destroy(&g_init_sem[ctx->cpu]);
+
+	running[ctx->cpu] = TRUE;
+
+	if (mtcp_master < 0) {
+		mtcp_master = ctx->cpu;
+		TRACE_INFO("CPU %d is now the master thread.\n", mtcp_master);
+	}
+
+	return ctx;
+#endif
 }
 /*----------------------------------------------------------------------------*/
 #ifndef DISABLE_DPDK
@@ -1231,6 +1260,7 @@ mtcp_create_context(int cpu)
 		return NULL;
 	}
 #endif
+#ifndef ENABLE_SPDK
 #ifndef DISABLE_DPDK
 	/* Wake up mTCP threads (wake up I/O threads) */
 	if (current_iomodule_func == &dpdk_module_func) {
@@ -1268,6 +1298,7 @@ mtcp_create_context(int cpu)
 		TRACE_INFO("CPU %d is now the master thread.\n", mtcp_master);
 	}
 
+#endif
 	return mctx;
 }
 /*----------------------------------------------------------------------------*/
diff --git a/mtcp/src/dpdk_module.c b/mtcp/src/dpdk_module.c
index ebf124ecb..0688a2098 100644
--- a/mtcp/src/dpdk_module.c
+++ b/mtcp/src/dpdk_module.c
@@ -87,7 +87,7 @@
 
 #define ETHER_IFG			12
 #define	ETHER_PREAMBLE			8
-#define ETHER_OVR			(ETHER_CRC_LEN + ETHER_PREAMBLE + ETHER_IFG)
+#define ETHER_OVR			(MTCP_ETHER_CRC_LEN + ETHER_PREAMBLE + ETHER_IFG)
 
 static const uint16_t nb_rxd = 		RTE_TEST_RX_DESC_DEFAULT;
 static const uint16_t nb_txd = 		RTE_TEST_TX_DESC_DEFAULT;
@@ -98,7 +98,7 @@ static struct rte_mempool *pktmbuf_pool[MAX_CPUS] = {NULL};
 //#define DEBUG				1
 #ifdef DEBUG
 /* ethernet addresses of ports */
-static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
+static MTCP_ETHER_ADDR ports_eth_addr[RTE_MAX_ETHPORTS];
 #endif
 
 static struct rte_eth_dev_info dev_info[RTE_MAX_ETHPORTS];
@@ -106,7 +106,7 @@ static struct rte_eth_dev_info dev_info[RTE_MAX_ETHPORTS];
 static struct rte_eth_conf port_conf = {
 	.rxmode = {
 		.mq_mode	= 	ETH_MQ_RX_RSS,
-		.max_rx_pkt_len = 	ETHER_MAX_LEN,
+		.max_rx_pkt_len = 	MTCP_ETHER_MAX_LEN,
 #if RTE_VERSION > RTE_VERSION_NUM(17, 8, 0, 0)
 		.offloads	=	(
 #if RTE_VERSION < RTE_VERSION_NUM(18, 5, 0, 0)
@@ -412,7 +412,7 @@ dpdk_get_wptr(struct mtcp_thread_context *ctxt, int ifidx, uint16_t pktsize)
 	m = dpc->wmbufs[ifidx].m_table[len_of_mbuf];
 
 	/* retrieve the right write offset */
-	ptr = (void *)rte_pktmbuf_mtod(m, struct ether_hdr *);
+	ptr = (void *)rte_pktmbuf_mtod(m, MTCP_ETHER_HDR *);
 	m->pkt_len = m->data_len = pktsize;
 	m->nb_segs = 1;
 	m->next = NULL;
@@ -467,16 +467,16 @@ dpdk_recv_pkts(struct mtcp_thread_context *ctxt, int ifidx)
 struct rte_mbuf *
 ip_reassemble(struct dpdk_private_context *dpc, struct rte_mbuf *m)
 {
-	struct ether_hdr *eth_hdr;
+	MTCP_ETHER_HDR *eth_hdr;
 	struct rte_ip_frag_tbl *tbl;
 	struct rte_ip_frag_death_row *dr;
 
 	/* if packet is IPv4 */
 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
-		struct ipv4_hdr *ip_hdr;
+		MTCP_IPV4_HDR *ip_hdr;
 
-		eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
-		ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
+		eth_hdr = rte_pktmbuf_mtod(m, MTCP_ETHER_HDR *);
+		ip_hdr = (MTCP_IPV4_HDR *)(eth_hdr + 1);
 
 		/* if it is a fragmented packet, then try to reassemble. */
 		if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
@@ -702,6 +702,10 @@ dpdk_load_module(void)
 			/* init port */
 			printf("Initializing port %u... ", (unsigned) portid);
 			fflush(stdout);
+
+			if (!strncmp(dev_info[portid].driver_name, "net_bnxt", 8))
+				port_conf.rx_adv_conf.rss_conf.rss_key_len = 40;
+			
 			ret = rte_eth_dev_configure(portid, CONFIG.num_cores, CONFIG.num_cores, &port_conf);
 			if (ret < 0)
 				rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u, cores: %d\n",
@@ -822,7 +826,7 @@ dpdk_dev_ioctl(struct mtcp_thread_context *ctx, int nif, int cmd, void *argp)
 			goto dev_ioctl_err;
 		m = dpc->wmbufs[eidx].m_table[len_of_mbuf - 1];
 		m->ol_flags = PKT_TX_IP_CKSUM | PKT_TX_IPV4;
-		m->l2_len = sizeof(struct ether_hdr);
+		m->l2_len = sizeof(MTCP_ETHER_HDR);
 		m->l3_len = (iph->ihl<<2);
 		break;
 	case PKT_TX_TCP_CSUM:
@@ -831,19 +835,19 @@ dpdk_dev_ioctl(struct mtcp_thread_context *ctx, int nif, int cmd, void *argp)
 		m = dpc->wmbufs[eidx].m_table[len_of_mbuf - 1];
 		tcph = (struct tcphdr *)((unsigned char *)iph + (iph->ihl<<2));
 		m->ol_flags |= PKT_TX_TCP_CKSUM;
-		tcph->check = rte_ipv4_phdr_cksum((struct ipv4_hdr *)iph, m->ol_flags);
+		tcph->check = rte_ipv4_phdr_cksum((MTCP_IPV4_HDR *)iph, m->ol_flags);
 		break;
 #ifdef ENABLELRO
 	case PKT_RX_TCP_LROSEG:
 		m = dpc->cur_rx_m;
 		//if (m->next != NULL)
 		//	rte_prefetch0(rte_pktmbuf_mtod(m->next, void *));
-		iph = rte_pktmbuf_mtod_offset(m, struct iphdr *, sizeof(struct ether_hdr));
+		iph = rte_pktmbuf_mtod_offset(m, struct iphdr *, sizeof(MTCP_ETHER_HDR));
 		tcph = (struct tcphdr *)((u_char *)iph + (iph->ihl << 2));
 		payload = (uint8_t *)tcph + (tcph->doff << 2);
 
 		seg_off = m->data_len -
-			sizeof(struct ether_hdr) - (iph->ihl << 2) -
+			sizeof(MTCP_ETHER_HDR) - (iph->ihl << 2) -
 			(tcph->doff << 2);
 
 		to = (uint8_t *) argp;
@@ -866,13 +870,13 @@ dpdk_dev_ioctl(struct mtcp_thread_context *ctx, int nif, int cmd, void *argp)
 		if ((dev_info[nif].tx_offload_capa & DEV_TX_OFFLOAD_TCP_CKSUM) == 0)
 			goto dev_ioctl_err;
 		m = dpc->wmbufs[eidx].m_table[len_of_mbuf - 1];
-		iph = rte_pktmbuf_mtod_offset(m, struct iphdr *, sizeof(struct ether_hdr));
+		iph = rte_pktmbuf_mtod_offset(m, struct iphdr *, sizeof(MTCP_ETHER_HDR));
 		tcph = (struct tcphdr *)((uint8_t *)iph + (iph->ihl<<2));
-		m->l2_len = sizeof(struct ether_hdr);
+		m->l2_len = sizeof(MTCP_ETHER_HDR);
 		m->l3_len = (iph->ihl<<2);
 		m->l4_len = (tcph->doff<<2);
 		m->ol_flags = PKT_TX_TCP_CKSUM | PKT_TX_IP_CKSUM | PKT_TX_IPV4;
-		tcph->check = rte_ipv4_phdr_cksum((struct ipv4_hdr *)iph, m->ol_flags);
+		tcph->check = rte_ipv4_phdr_cksum((MTCP_IPV4_HDR *)iph, m->ol_flags);
 		break;
 	case PKT_RX_IP_CSUM:
 		if ((dev_info[nif].rx_offload_capa & DEV_RX_OFFLOAD_IPV4_CKSUM) == 0)
diff --git a/mtcp/src/include/mtcp.h b/mtcp/src/include/mtcp.h
index ba07fd773..1e92c3433 100644
--- a/mtcp/src/include/mtcp.h
+++ b/mtcp/src/include/mtcp.h
@@ -39,6 +39,22 @@
 #define ERROR (-1)
 #endif
 
+#if RTE_VERSION < RTE_VERSION_NUM(19, 8, 0, 0)
+typedef struct ether_addr MTCP_ETHER_ADDR;
+typedef struct ipv4_hdr MTCP_IPV4_HDR;
+typedef struct ether_hdr MTCP_ETHER_HDR;
+
+#define MTCP_ETHER_MAX_LEN  ETHER_MAX_LEN
+#define MTCP_ETHER_CRC_LEN  ETHER_CRC_LEN
+#else
+typedef struct rte_ether_addr MTCP_ETHER_ADDR;
+typedef struct rte_ipv4_hdr MTCP_IPV4_HDR;
+typedef struct rte_ether_hdr MTCP_ETHER_HDR;
+
+#define MTCP_ETHER_MAX_LEN  RTE_ETHER_MAX_LEN
+#define MTCP_ETHER_CRC_LEN  RTE_ETHER_CRC_LEN
+#endif
+
 #define ETHERNET_HEADER_LEN		14	// sizeof(struct ethhdr)
 #define IP_HEADER_LEN			20	// sizeof(struct iphdr)
 #define TCP_HEADER_LEN			20	// sizeof(struct tcphdr)
diff --git a/mtcp/src/include/mtcp_api.h b/mtcp/src/include/mtcp_api.h
index 0a1ee40fd..017cb51d7 100644
--- a/mtcp/src/include/mtcp_api.h
+++ b/mtcp/src/include/mtcp_api.h
@@ -46,7 +46,7 @@ int
 mtcp_init(const char *config_file);
 
 void 
-mtcp_destroy();
+mtcp_destroy(void);
 
 int 
 mtcp_getconf(struct mtcp_conf *conf);
@@ -68,6 +68,9 @@ typedef void (*mtcp_sighandler_t)(int);
 mtcp_sighandler_t 
 mtcp_register_signal(int signum, mtcp_sighandler_t handler);
 
+void *
+MTCPRunThread(void *arg);
+
 int 
 mtcp_pipe(mctx_t mctx, int pipeid[2]);
 
@@ -87,6 +90,9 @@ mtcp_setsock_nonblock(mctx_t mctx, int sockid);
 int 
 mtcp_socket_ioctl(mctx_t mctx, int sockid, int request, void *argp);
 
+int
+mtcp_run_instance(void *tmp);
+
 int 
 mtcp_socket(mctx_t mctx, int domain, int type, int protocol);
 
@@ -124,7 +130,7 @@ int
 mtcp_getpeername(mctx_t mctx, int sockid, struct sockaddr *addr,
 		 socklen_t *addrlen);
 
-inline ssize_t
+ssize_t
 mtcp_read(mctx_t mctx, int sockid, char *buf, size_t len);
 
 ssize_t
diff --git a/mtcp/src/io_module.c b/mtcp/src/io_module.c
index 126c94f46..9864ad12a 100644
--- a/mtcp/src/io_module.c
+++ b/mtcp/src/io_module.c
@@ -260,7 +260,11 @@ SetNetEnv(char *dev_name_list, char *port_stat_list)
 		char socket_mem_str[32] = "";
 		// int i;
 		int ret, socket_mem;
+#if RTE_VERSION < RTE_VERSION_NUM(19, 8, 0, 0)
 		static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
+#else
+		static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
+#endif
 
 		/* STEP 1: first determine CPU mask */
 		mpz_init(_cpumask);
@@ -343,8 +347,10 @@ SetNetEnv(char *dev_name_list, char *port_stat_list)
 		/* initialize the dpdk eal env */
 		ret = rte_eal_init(argc, argv);
 		if (ret < 0) {
-			TRACE_ERROR("Invalid EAL args!\n");
-			exit(EXIT_FAILURE);
+			if(rte_errno != EALREADY) {
+				TRACE_ERROR("Invalid EAL args!\n");
+				exit(EXIT_FAILURE);
+			}
 		}
 		/* give me the count of 'detected' ethernet ports */
 #if RTE_VERSION < RTE_VERSION_NUM(18, 5, 0, 0)
diff --git a/mtcp/src/onvm_module.c b/mtcp/src/onvm_module.c
index eed55f263..4d8a8a5ea 100644
--- a/mtcp/src/onvm_module.c
+++ b/mtcp/src/onvm_module.c
@@ -63,7 +63,11 @@ static struct rte_mempool *pktmbuf_pool = NULL;
 //#define DEBUG				1
 #ifdef DEBUG
 /* ethernet addresses of ports */
+#if RTE_VERSION < RTE_VERSION_NUM(19, 8, 0, 0)
 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
+#else
+static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
+#endif
 #endif
 
 static struct rte_eth_dev_info dev_info[RTE_MAX_ETHPORTS];
diff --git a/mtcp_crossbuild.sh b/mtcp_crossbuild.sh
new file mode 100755
index 000000000..64bf43ec0
--- /dev/null
+++ b/mtcp_crossbuild.sh
@@ -0,0 +1,110 @@
+#!/bin/bash
+
+args="$*"
+
+usage ()
+{
+  echo "args=$args"
+  echo
+  echo "`basename $0` -h -c <x86/aarch64> -k <kernel_dir> -i <ext_lib> -d <dpdk_dir> -e"
+  echo
+  echo "Helper script, used to build dpdk."
+  echo
+  echo " -h               Help Usage"
+  echo " -c <x86/aarch64> specify build platform"
+  echo " -k <kernel_dir>  Directory that kernel builds if enable LKM build option"
+  echo " -i <ext_lib>     If needed introduce external lib dependencies"
+  echo " -d <dpdk_dir>    Directory that dpdk builds"
+  echo " -e               Enable SPDK Support"
+  echo
+}
+
+while getopts "hc:k:i:d:e" opt; do
+  case $opt in
+    h)  show_usage=1
+        ;;
+    c)  build_arch="$OPTARG"
+        ;;
+    k)  kernel_build="$OPTARG"
+        ;;
+    i)  ext_lib="$OPTARG"
+        ;;
+    d)  dpdk_build="$OPTARG"
+        ;;
+    e)  spdk_enable=1
+	;;
+    \?)
+      echo "Invalid option: -$OPTARG" >&2
+      show_usage=1
+      ;;
+  esac
+done
+
+if [ "$show_usage" == "1" ]; then
+  usage
+  exit 1
+fi
+
+CUR_PATH=`pwd`
+
+if [ "$build_arch" == "aarch64" ]; then
+  CROSS=aarch64-linux-gnu-
+  RTE_TARGET=arm64-stingray-linuxapp-gcc
+else
+  RTE_TARGET=x86_64-native-linuxapp-gcc
+fi
+
+if [ -z "$kernel_build" ]; then
+  RTE_KERNELDIR=/lib/modules/`uname -r`/build
+else
+  RTE_KERNELDIR=$kernel_build
+fi
+
+if [ -z "$ext_lib" ]; then
+  EXT_LIB_DIR=$CUR_PATH/../ext_lib
+else
+  EXT_LIB_DIR=$ext_lib
+fi
+
+if [ "$spdk_enable" == "1" ]; then
+  SPDK_OPTION="--enable-spdk"
+fi
+
+if [ -z $dpdk_build ]; then
+  RTE_SDK=$CUR_PATH/../dpdk
+else
+  RTE_SDK=$dpdk_build
+fi
+DPDK_BUILD=$RTE_SDK/$RTE_TARGET
+export RTE_SDK RTE_TARGET DPDK_BUILD RTE_KERNELDIR
+
+# check ldflags.txt
+if grep "ldflags.txt" $RTE_SDK/mk/rte.app.mk > /dev/null
+then
+    :
+else
+    sed -i -e 's/O_TO_EXE_STR =/\$(shell if [ \! -d \${RTE_SDK}\/\${RTE_TARGET}\/lib ]\; then mkdir \${RTE_SDK}\/\${RTE_TARGET}\/lib\; fi)\nLINKER_FLAGS = \$(call linkerprefix,\$(LDLIBS))\n\$(shell echo \${LINKER_FLAGS} \> \${RTE_SDK}\/\${RTE_TARGET}\/lib\/ldflags\.txt)\nO_TO_EXE_STR =/g' $RTE_SDK/mk/rte.app.mk
+    echo "Need to rebuild dpdk."
+    exit 1
+fi
+
+echo
+echo "======================================"
+echo "Build_arch  : $build_arch"
+echo "RTE_TARGET  : $RTE_TARGET"
+echo "DPDK Build  : $RTE_SDK"
+echo "SPDK Support: $spdk_enable"
+echo "======================================"
+echo
+
+CUR_PATH=`pwd`
+
+# build kernel module
+cd $CUR_PATH/dpdk-iface-kmod && make CROSS=$CROSS RTE_KERNELDIR=$RTE_KERNELDIR V=1
+
+# build application
+cd $CUR_PATH && autoreconf -f -i && \
+    ./configure --host=aarch64 CC=${CROSS}gcc LD=${CROSS}ld --with-dpdk=$RTE_SDK/$RTE_TARGET --with-dpdk-lib=$RTE_SDK/$RTE_TARGET/lib $SPDK_OPTION CFLAGS="-I$EXT_LIB_DIR/include" LDFLAGS="-L$EXT_LIB_DIR/lib64 -L$EXT_LIB_DIR/lib"
+make -j `grep -c ^processor /proc/cpuinfo` ARCH=arm64 CC=${CROSS}gcc LD=${CROSS}ld V=0
+
+# end of file
diff --git a/util/Makefile.in b/util/Makefile.in
index 0d03bf271..f48fcafe2 100644
--- a/util/Makefile.in
+++ b/util/Makefile.in
@@ -1,7 +1,9 @@
 ### GCC ###
 GCC=@CC@
 
-ifeq ($(shell uname -m),x86_64)
+# If ARCH is not defined, retrive from system
+ARCH ?= $(shell uname -m)
+ifeq ($(ARCH),x86_64)
 GCC_OPT = -m64
 else
 GCC_OPT =