diff --git a/README.en.md b/README.en.md
index 1e22c10..cb32787 100644
--- a/README.en.md
+++ b/README.en.md
@@ -147,6 +147,7 @@ Both the use and configuration of the protocol is very simple, in most cases, af
 - [asio-kcp](https://github.com/libinzhangyuan/asio_kcp): Use the complete UDP network library of KCP, complete implementation of UDP-based link state management, session control and KCP protocol scheduling, etc.
 - [kcp-java](https://github.com/hkspirt/kcp-java)：Implementation of Java version of KCP protocol.
 - [kcp-netty](https://github.com/szhnet/kcp-netty)：Java implementation of KCP based on Netty.
+- [java-kcp](https://github.com/l42111996/java-Kcp): JAVA version KCP, based on netty implementation (including fec function)
 - [kcp-go](https://github.com/xtaci/kcp-go): High-security GO language implementation of kcp, including simple implementation of UDP session management, as a base library for subsequent development.
 - [kcp-csharp](https://github.com/limpo1989/kcp-csharp): The csharp migration of kcp, containing the session management, which can access the above kcp-go server.
 - [kcp-rs](https://github.com/en/kcp-rs): The rust migration of KCP
@@ -176,13 +177,21 @@ Thanks to [zhangyuan](https://github.com/libinzhangyuan) the author of [asio-kcp
 
 For specifics please refer to: [Reliable Udp Benchmark](https://github.com/libinzhangyuan/reliable_udp_bench_mark) and [KCP-Benchmark](https://github.com/skywind3000/kcp/wiki/KCP-Benchmark), for more guidance to the hesitant users.
 
+MMO Engine [SpatialOS](https://improbable.io/spatialOS) has a benchmark report on KCP/TCP/RakNet:
+
+![](https://raw.githubusercontent.com/skywind3000/kcp/master/images/spatialos-50.png)
+
+for more details, please see the report itself:
+
+- [Kcp a new low latency secure network stack](https://improbable.io/blog/kcp-a-new-low-latency-secure-network-stack)
+
 # KCP is used by
 
 See [Success Stories](https://github.com/skywind3000/kcp/wiki/Success-Stories).
 
 # Donation
 
-![欢迎使用支付宝对该项目进行捐赠](https://raw.githubusercontent.com/skywind3000/kcp/master/donation.png)
+![欢迎使用支付宝对该项目进行捐赠](https://raw.githubusercontent.com/skywind3000/kcp/master/images/donation.png)
 
 Donation is welcome by using alipay, the money will be used to improve the protocol and documentation.
 
diff --git a/README.md b/README.md
index 10ecbb0..0e742c2 100644
--- a/README.md
+++ b/README.md
@@ -1,8 +1,8 @@
 KCP - A Fast and Reliable ARQ Protocol
 ======================================
 
-[![Powered][2]][1] [![Build Status][4]][5]
-[![Backers on Open Collective](https://opencollective.com/kcp/backers/badge.svg)](#backers)
+[![Powered][2]][1] [![Build Status][4]][5]
+[![Backers on Open Collective](https://opencollective.com/kcp/backers/badge.svg)](#backers)
  [![Sponsors on Open Collective](https://opencollective.com/kcp/sponsors/badge.svg)](#sponsors) 
 
 [1]: https://github.com/skywind3000/kcp
@@ -146,6 +146,7 @@ TCP是为流量设计的（每秒内可以传输多少KB的数据），讲究的
 - [asio-kcp](https://github.com/libinzhangyuan/asio_kcp): 使用 KCP的完整 UDP网络库，完整实现了基于 UDP的链接状态管理，会话控制，KCP协议调度等
 - [kcp-java](https://github.com/hkspirt/kcp-java)：Java版本 KCP协议实现。
 - [kcp-netty](https://github.com/szhnet/kcp-netty)：kcp的Java语言实现，基于netty。
+- [java-kcp](https://github.com/l42111996/java-Kcp): JAVA版本KCP,基于netty实现(包含fec功能)
 - [kcp-go](https://github.com/xtaci/kcp-go): 高安全性的kcp的 GO语言实现，包含 UDP会话管理的简单实现，可以作为后续开发的基础库。 
 - [kcp-csharp](https://github.com/limpo1989/kcp-csharp): kcp的 csharp移植，同时包含一份回话管理，可以连接上面kcp-go的服务端。
 - [kcp-csharp](https://github.com/KumoKyaku/KCP): 新版本 Kcp的 csharp移植。线程安全，运行时无alloc，对gc无压力。
@@ -169,6 +170,7 @@ TCP是为流量设计的（每秒内可以传输多少KB的数据），讲究的
 - [CC](http://cc.163.com/)：网易 CC 使用 kcp 加速视频推流，有效提高流畅性
 - [BOBO](http://bobo.163.com/)：网易 BOBO 使用 kcp 加速主播推流
 - [云帆加速](http://www.yfcloud.com/)：使用 KCP 加速文件传输和视频推流，优化了台湾主播推流的流畅度
+- [SpatialOS](https://improbable.io/spatialOS): 大型多人分布式游戏服务端引擎，BigWorld 的后继者，使用 KCP 加速数据传输。
 
 欢迎告知更多案例
 
@@ -190,12 +192,21 @@ TCP是为流量设计的（每秒内可以传输多少KB的数据），讲究的
 
 具体见：[横向比较](https://github.com/libinzhangyuan/reliable_udp_bench_mark) 和 [评测数据](https://github.com/skywind3000/kcp/wiki/KCP-Benchmark)，为犹豫选择的人提供了更多指引。
 
+大型多人游戏服务端引擎 [SpatialOS](https://improbable.io/spatialOS) 在集成 KCP 协议后做了同 TCP/RakNet 的评测：
 
+![](https://raw.githubusercontent.com/skywind3000/kcp/master/images/spatialos-50.png)
 
+对比了在服务端刷新率为 60 Hz 同时维护 50 个角色时的响应时间，详细对比报告见：
+
+- [Kcp a new low latency secure network stack](https://improbable.io/blog/kcp-a-new-low-latency-secure-network-stack)
+
+# 项目历史
+
+See [Success Stories](https://github.com/skywind3000/kcp/wiki/Success-Stories).
 
 # 欢迎捐赠
 
-![欢迎使用支付宝对该项目进行捐赠](https://raw.githubusercontent.com/skywind3000/kcp/master/donation.png)
+![欢迎使用支付宝对该项目进行捐赠](https://raw.githubusercontent.com/skywind3000/kcp/master/images/donation.png)
 
 欢迎使用支付宝手扫描上面的二维码，对该项目进行捐赠。捐赠款项将用于持续优化 KCP协议以及完善文档。
 
diff --git a/ikcp.c b/ikcp.c
index bb1a0c6..2334fd6 100644
--- a/ikcp.c
+++ b/ikcp.c
@@ -43,6 +43,7 @@ const IUINT32 IKCP_THRESH_INIT = 2;
 const IUINT32 IKCP_THRESH_MIN = 2;
 const IUINT32 IKCP_PROBE_INIT = 7000;		// 7 secs to probe window size
 const IUINT32 IKCP_PROBE_LIMIT = 120000;	// up to 120 secs to probe window
+const IUINT32 IKCP_FASTACK_LIMIT = 5;		// max times to trigger fastack
 
 
 //---------------------------------------------------------------------
@@ -66,7 +67,7 @@ static inline const char *ikcp_decode8u(const char *p, unsigned char *c)
 /* encode 16 bits unsigned int (lsb) */
 static inline char *ikcp_encode16u(char *p, unsigned short w)
 {
-#if IWORDS_BIG_ENDIAN
+#if IWORDS_BIG_ENDIAN || IWORDS_MUST_ALIGN
 	*(unsigned char*)(p + 0) = (w & 255);
 	*(unsigned char*)(p + 1) = (w >> 8);
 #else
@@ -79,7 +80,7 @@ static inline char *ikcp_encode16u(char *p, unsigned short w)
 /* decode 16 bits unsigned int (lsb) */
 static inline const char *ikcp_decode16u(const char *p, unsigned short *w)
 {
-#if IWORDS_BIG_ENDIAN
+#if IWORDS_BIG_ENDIAN || IWORDS_MUST_ALIGN
 	*w = *(const unsigned char*)(p + 1);
 	*w = *(const unsigned char*)(p + 0) + (*w << 8);
 #else
@@ -92,7 +93,7 @@ static inline const char *ikcp_decode16u(const char *p, unsigned short *w)
 /* encode 32 bits unsigned int (lsb) */
 static inline char *ikcp_encode32u(char *p, IUINT32 l)
 {
-#if IWORDS_BIG_ENDIAN
+#if IWORDS_BIG_ENDIAN || IWORDS_MUST_ALIGN
 	*(unsigned char*)(p + 0) = (unsigned char)((l >>  0) & 0xff);
 	*(unsigned char*)(p + 1) = (unsigned char)((l >>  8) & 0xff);
 	*(unsigned char*)(p + 2) = (unsigned char)((l >> 16) & 0xff);
@@ -107,7 +108,7 @@ static inline char *ikcp_encode32u(char *p, IUINT32 l)
 /* decode 32 bits unsigned int (lsb) */
 static inline const char *ikcp_decode32u(const char *p, IUINT32 *l)
 {
-#if IWORDS_BIG_ENDIAN
+#if IWORDS_BIG_ENDIAN || IWORDS_MUST_ALIGN
 	*l = *(const unsigned char*)(p + 3);
 	*l = *(const unsigned char*)(p + 2) + (*l << 8);
 	*l = *(const unsigned char*)(p + 1) + (*l << 8);
@@ -283,6 +284,7 @@ ikcpcb* ikcp_create(IUINT32 conv, void *user)
 	kcp->logmask = 0;
 	kcp->ssthresh = IKCP_THRESH_INIT;
 	kcp->fastresend = 0;
+	kcp->fastlimit = IKCP_FASTACK_LIMIT;
 	kcp->nocwnd = 0;
 	kcp->xmit = 0;
 	kcp->dead_link = IKCP_DEADLINK;
@@ -604,7 +606,7 @@ static void ikcp_parse_una(ikcpcb *kcp, IUINT32 una)
 	}
 }
 
-static void ikcp_parse_fastack(ikcpcb *kcp, IUINT32 sn)
+static void ikcp_parse_fastack(ikcpcb *kcp, IUINT32 sn, IUINT32 ts)
 {
 	struct IQUEUEHEAD *p, *next;
 
@@ -618,7 +620,12 @@ static void ikcp_parse_fastack(ikcpcb *kcp, IUINT32 sn)
 			break;
 		}
 		else if (sn != seg->sn) {
+		#ifndef IKCP_FASTACK_CONSERVE
 			seg->fastack++;
+		#else
+			if (_itimediff(ts, seg->ts) >= 0)
+				seg->fastack++;
+		#endif
 		}
 	}
 }
@@ -741,8 +748,8 @@ void ikcp_parse_data(ikcpcb *kcp, IKCPSEG *newseg)
 //---------------------------------------------------------------------
 int ikcp_input(ikcpcb *kcp, const char *data, long size)
 {
-	IUINT32 una = kcp->snd_una;
-	IUINT32 maxack = 0;
+	IUINT32 prev_una = kcp->snd_una;
+	IUINT32 maxack = 0, latest_ts = 0;
 	int flag = 0;
 
 	if (ikcp_canlog(kcp, IKCP_LOG_INPUT)) {
@@ -791,9 +798,18 @@ int ikcp_input(ikcpcb *kcp, const char *data, long size)
 			if (flag == 0) {
 				flag = 1;
 				maxack = sn;
+				latest_ts = ts;
 			}	else {
 				if (_itimediff(sn, maxack) > 0) {
+				#ifndef IKCP_FASTACK_CONSERVE
 					maxack = sn;
+					latest_ts = ts;
+				#else
+					if (_itimediff(ts, latest_ts) > 0) {
+						maxack = sn;
+						latest_ts = ts;
+					}
+				#endif
 				}
 			}
 			if (ikcp_canlog(kcp, IKCP_LOG_IN_ACK)) {
@@ -853,10 +869,10 @@ int ikcp_input(ikcpcb *kcp, const char *data, long size)
 	}
 
 	if (flag != 0) {
-		ikcp_parse_fastack(kcp, maxack);
+		ikcp_parse_fastack(kcp, maxack, latest_ts);
 	}
 
-	if (_itimediff(kcp->snd_una, una) > 0) {
+	if (_itimediff(kcp->snd_una, prev_una) > 0) {
 		if (kcp->cwnd < kcp->rmt_wnd) {
 			IUINT32 mss = kcp->mss;
 			if (kcp->cwnd < kcp->ssthresh) {
@@ -1048,11 +1064,14 @@ void ikcp_flush(ikcpcb *kcp)
 			lost = 1;
 		}
 		else if (segment->fastack >= resent) {
-			needsend = 1;
-			segment->xmit++;
-			segment->fastack = 0;
-			segment->resendts = current + segment->rto;
-			change++;
+			if ((int)segment->xmit <= kcp->fastlimit || 
+				kcp->fastlimit <= 0) {
+				needsend = 1;
+				segment->xmit++;
+				segment->fastack = 0;
+				segment->resendts = current + segment->rto;
+				change++;
+			}
 		}
 
 		if (needsend) {
diff --git a/ikcp.h b/ikcp.h
index 866d188..e525105 100644
--- a/ikcp.h
+++ b/ikcp.h
@@ -225,7 +225,7 @@ typedef struct IQUEUEHEAD iqueue_head;
 
 
 //---------------------------------------------------------------------
-// WORD ORDER
+// BYTE ORDER & ALIGNMENT
 //---------------------------------------------------------------------
 #ifndef IWORDS_BIG_ENDIAN
     #ifdef _BIG_ENDIAN_
@@ -248,6 +248,17 @@ typedef struct IQUEUEHEAD iqueue_head;
     #endif
 #endif
 
+#ifndef IWORDS_MUST_ALIGN
+	#if defined(__i386__) || defined(__i386) || defined(_i386_)
+		#define IWORDS_MUST_ALIGN 0
+	#elif defined(_M_IX86) || defined(_X86_) || defined(__x86_64__)
+		#define IWORDS_MUST_ALIGN 0
+	#elif defined(__amd64) || defined(__amd64__)
+		#define IWORDS_MUST_ALIGN 0
+	#else
+		#define IWORDS_MUST_ALIGN 1
+	#endif
+#endif
 
 
 //=====================================================================
@@ -298,6 +309,7 @@ struct IKCPCB
 	void *user;
 	char *buffer;
 	int fastresend;
+	int fastlimit;
 	int nocwnd, stream;
 	int logmask;
 	int (*output)(const char *buf, int len, struct IKCPCB *kcp, void *user);
diff --git a/donation.png b/images/donation.png
similarity index 100%
rename from donation.png
rename to images/donation.png
diff --git a/images/spatialos-25.png b/images/spatialos-25.png
new file mode 100644
index 0000000..eab42ed
Binary files /dev/null and b/images/spatialos-25.png differ
diff --git a/images/spatialos-50.png b/images/spatialos-50.png
new file mode 100644
index 0000000..3f63732
Binary files /dev/null and b/images/spatialos-50.png differ