Wireless Pers Commun DOI 10.

1007/s11277-011-0358-8

An Enhanced A-MSDU Frame Aggregation Scheme for 802.11n Wireless Networks

Anwar Saif Mohamed Othman Shamala Subramaniam Nor Asila Wati Abdul Hamid

Springer Science+Business Media, LLC. 2011

Abstract The main goal of the IEEE 802.11n standard is to achieve a minimum throughput of 100 Mbps at the MAC service access point. This high throughput has been achieved via many enhancements in both the physical and MAC layers. A key enhancement at the MAC layer is frame aggregation in which the timing and headers overheads of the legacy MAC are reduced by aggregating multiple frames into a single large frame before being transmitted. Two aggregation schemes have been dened by the 802.11n standard, aggregate MAC service data unit (A-MSDU) and aggregate MAC protocol data unit (A-MPDU). As a consequence of the aggregation, new aggregation headers are introduced and become parts of the transmitted frame. Even though these headers are small compared to the legacy headers they still have a negative impact on the network performance, especially when aggregating frames of small payload. Moreover, the A-MSDU is highly inuenced by the channel condition due mainly to lack of subframes sequence control and retransmission. In this paper, we have proposed an aggregation scheme ( mA-MSDU) that reduces the aggregation headers and implements a retransmission control over the individual subframes at the MSDU level. The analysis and simulations results show the significance of the proposed scheme, specifically for applications that have a small frame size such as VoIP. Keywords Frame aggregation Aggregation headers WLAN A-MSDU Next generation networks 802.11n

A. Saif (B M. Othman S. Subramaniam N. A. W. A. Hamid ) Department of Communication Technology and Network, Faculty of Computer Science and Information Technology, Universiti Putra Malaysia, 43400 UPM, Serdang, Selangor D.E., Malaysia e-mail: anwarsaif.ye@gmail.com M. Othman e-mail: mothman@fsktm.upm.edu.my S. Subramaniam e-mail: shamala@fsktm.upm.edu.my N. A. W. A. Hamid e-mail: asila@fsktm.upm.edu.my

123

A. Saif et al.

1 Introduction IEEE 802.11n [1] is a new standard that introduces many enhancements at the PHY and MAC layers in order to achieve a throughput up to 100 Mbps at the MAC access service point (SAP). The 802.11n data rates are significantly improved through the use of spatial multiplexing using MIMO and 40MHz channels. As a consequence, the MAC functions need to be enhanced in order to get benet from the high data rates provided by PHY layer. The major MAC enhancements are accomplished through the use of frame aggregation and block acknowledgement. Without these MAC enhancements, the xed MAC headers limit the overall throughput even for high data rates. According to the limit theorem dened in [2], overheads consume the channel time and limit the throughput compared with the actual data rate, even when the data rate goes to innite high. The overheads of the MAC layer are due to the nature of the channel access that is dened by the legacy IEEE 802.11 [3] distributed coordination function (DCF). These overheads such as large frame headers, inter-frame spacing, back-off timer, and acknowledgement need to be reduced in order to make use of the high data rates of the PHY layer. To reduce the overheads at the MAC layer, IEEE 80211n introduces many enhancements such as block ACK, reverse direction transmission and frame aggregation. In the frame aggregation, multiple frames are aggregated into a large frame and treated as a single frame during transmission. However, frame aggregation has introduced a new header for each subframe. These headers are required for de-aggregation at the receiver side and xed irrespective of the payload of the aggregated subframes. Moreover, the increasing demand on applications with small frame sizes such as VoIP and online gaming as well as the large distribution of the small frames on the internet [4] made it essential to introduce more optimized frame aggregations in order to full the small frame transmission requirements. The following conventions are used in this paper: MSDU is the received/forwarded packet from/to the LLC layer, subframe is an MSDU attached with the aggregation headers, and superframe is the aggregation frame that is formed from one or more subframes. The rest of the paper is organized as follows: Sect. 2 gives a background on the 802.11n MAC aggregation schemes and enhancements whereas Sect. 3 addresses the related studies. In Sect. 5 we have analysed the header overheads of different aggregation schemes under different aggregation sizes. The description of the proposed scheme along with its header analysis are presented in Sect. 6. The performance evaluation is illustrated in Section 7. Finally the work is concluded in Sect. 9.

2 IEEE 802.11n MAC Enhancements According to the 802.11 DCF protocol, before transmitting a frame the transmitting station keeps monitoring the channel until it detects an idle period of the distributed inter-frame spacing (DIFS). If the idle period is detected, a decrementing random back-off timer will be generated. Transmission will begin when the back-off timer reaches zero and the channel still idle. Many DCF enhancements such as block acknowledgement and frame aggregation have been proposed by the IEEE 802.11n in order to improve the performance and satisfy the requirements of the high-speed wireless networks.

123

An Enhanced A-MSDU Frame Aggregation Scheme

2.1 The Block Acknowledgement The block acknowledgement (BA) was introduced in the 802.11e [5] amendment to improve the MAC efciency by sending a bulk of data frames. This bulk of frames are acknowledged by a single BA frame instead of an ACK frame for each individual data frame. When a station gains a channel access opportunity, it sends multiple data frames separated by a small inter-frame spacing (SIFS). The sent frames are acknowledged by only one BA frame at the end of the transmitted block. The BA frame contains a bitmap of size 128 bytes that indicates the status of each frame of the transmitted block. This type of acknowledgement requires negotiation messages between the sender and receiver for establishing and tearing down the transmission session. The transaction of the negotiation messages reduces the channel efciency, mainly in erroneous channels. Moreover, the length of the bitmap is considered large as the BA is always sent using the basic data rate. IEEE 802.11n enhanced the BA by introducing a new block ACK scheme called compressed block ACK (CBA) by which both the on-air overhead and memory requirements are reduced. In the CBA, the bitmap is reduced to only 8 bytes instead of 128 bytes. The rst bit of the bitmap represents the MPDU with the same sequence number as the start sequence number (SSN) and subsequent bits indicate successive sequence numbers. The bit at index n of the bitmap acknowledges the MPDU with the nth sequence number relative to the SSN. Setting the bit n in the bitmap to 1 indicates that the corresponding MPDU has been received successfully, if not the value will be set to 0. 2.2 Frame Aggregation Frame aggregation, as dened by 802.11n, is a MAC-layer function that combines several MAC frames into a single PHY Layer Convergence Procedure (PLCP) frame for transmission. Two aggregation types have been dened, aggregate MAC service data unit (A- MSDU) and aggregate MAC protocol data unit (A-MPDU). 2.2.1 A-MSDU Aggregation Scheme In this aggregation scheme, several MSDUs destined to the same receiver are concatenated in a single MPDU. This operation is performed on the top of the MAC layer where the coming MSDUs are buffered and then concatenated in order to form the A-MSDU. The aggregation process is ended when the size of the buffered MSDUs reaches the maximum A-MSDU frame size or the delay of the oldest MSDU reaches the maximum delay limit. All MSDUs in one A-MSDU should be of the same trafc class, and the subframe header parameters destination address (DA) and source address (SA) should correspond to the MAC header parameters (RA and TA), multi-casting is not allowed in such aggregation. Figure 1 shows the A-MSDU structure, each A-MSDU subframe consists of a subframe header (DA, SA and length), MSDU and 03 bytes of padding which is necessary to keep the subframe at a multiple of four bytes for de-aggregation at the receiver side. If there is a corruption in any subframe, the whole A-MSDU will be dropped. 2.2.2 A-MPDU Aggregation Scheme The aggregate MPDU combines multiple MPDUs frames in a single PHY protocol data unit (PPDU) frame. Since the A-MPDU is constructed from MPDUs, it is possible to

123

A. Saif et al.
Fr. Ctrl Dur. /ID AD 1 AD 2 AD 3 Seq. Ctrl AD 4 QoS Ctrl HT Ctrl

A-MSDU

FCS

Subframe 1

Subframe 2

.
MSDU 0-2304B

Subframe n

Subframe header 14B

Padding 0-3B

DA 6B

SA 6B

length 2B

Fig. 1 The A-MSDU frame structure

Fig. 2 The A-MPDU frame structure

aggregate frames with different trafc identiers (TIDs). But, the A-MPDU subframes must be addressed to the same receiver address. Moreover, there is no waiting time during the construction of the A-MPDU. The A-MPDU is formed from the already available packets in the buffer. The corruption of any subframe does not require the retransmission of the whole A-MPDU, only the corrupted MPDUs need to be retransmitted. The structure of the A-MPDU is shown in Fig. 2. 3 Related Studies It was obvious that increasing the data rate and changing the coding might increase the performance of wireless networks and that what really happened in IEEE 802.11a. In these networks the performance is enhanced when the data rate is increased to 54 Mbps by using 5 GHz band and orthogonal frequency division multiplexing (OFDM) modulation technology. However, Xiao and Rosdahl [6,2] proved that a theoretical throughput upper limit (TUL) exists for the IEEE 802.11 protocols. The existence of such limit indicates that by simply increasing the data rate without reducing the PHY and MAC overheads is bounded even when the data rate goes to innite high. The authors showed that the MAC inefciency is due to the MAC headers, back-off time, inter-frame spacing and ACKs, mainly when the data rate is high or the frame is small. They came to a conclusion that pursuing higher data rates and reducing overheads are necessary in order to increase the throughput. To overcome the overheads of the IEEE 802.11 MAC, many aggregation mechanisms have been proposed for next generation wireless LANs. Xiao [7] proposed several MAC enhancements via frame aggregation. The author classied the frame aggregation into different

123

An Enhanced A-MSDU Frame Aggregation Scheme

aspects and proposed various aggregation schemes for each aspect. In these schemes, the size of the aggregated frames is limited to only the fragmentation threshold. Packing and concatenation [8] as well as aggregation scheme above the MAC layer [9] are early aggregation attempts but their aggregation headers still considered large for small payloads and the behaviour under erroneous channels has not been addressed. Other aggregation schemes have fragmented the packets before aggregation. In the aggregation and fragment retransmission scheme (AFR) proposed by Tianj et al. [10], multiple packets are fragmented and then aggregated into a single large frame. This scheme provides an error control of the transmitted fragments but introduces more headers and delay. Moreover, large buffers are required to enable the fragmentation/de-fragmentation processes. Adaptation is an effective method to solve the variety of the applications demands and the changing network conditions. Therefore, many adaptive schemes have been proposed. Riggio [11] have proposed an adaptive aggregation and differentiation scheme in which a priority mechanism and scheduling is implemented at the top of the MAC. However, this scheme was intended for the mesh networks, and the adaptation is based on the mesh network parameters. The schemes reserved pool of queues will consume a large memory. Yuxia and Vincent [12] have studied the saturation throughput and delay performance of the 802.11n aggregation techniques under error-prone channels. Based on their observations they have proposed an adaptive algorithm that denes the optimal frame size for the A-MSDU aggregation scheme. The algorithm works as follows: before transmitting an aggregated A-MSDU frame, the sending station will obtain an estimation of the channel bit error rate (BER), calculate the optimal frame for unidirectional and bi-directional respectively, and then construct the aggregated frame with a size that is close to the optimal frame size. Selvam and Srikanth [13] have presented an aggregation scheduler which will adaptively determine the aggregation type based on the size of the aggregation buffer. However, ordering the frames in the sender according to the size will affect the receiver reordering buffer leading to a higher delay. The IEEE 802.11n introduced A-MSDU and A-MPDU schemes as aggregation schemes for the next generation wireless networks. Many researchers have investigated the performance of these two schemes in terms of throughput and delay under different trafc characteristics and different network conditions. Dionysius et al. [14] have investigated the improvement of 802.11n throughput under error free channel. They have shown that, both schemes enhance the throughput with an advantage for the A-MPDU due to its large aggregation size. They clearly demonstrated that small packet size is the key factor that lowers the throughput efciency. Wang and Hung [15] have investigated the performance of IEEE 802.11n MAC protocols. They have demonstrated the inuence of aggregation, block acknowledgement, and reverse direction on the throughput. Ginzburg and Kesselman [16] have presented an analytical framework for estimating the maximum throughput of 802.11n using A-MPDU and A-MSDU aggregation schemes and concluded that the performance of A-MSDU aggregation significantly degrades for high packet error rates and high PHY rates. Kim et al. [17] showed through their analytical model as well as simulation that in unsaturated regions, the aggregation size has a small inuence on the throughput, while in the saturated regions, the throughput increases as the aggregation size increases. They also showed that under error free environment, the A-MSDU outperforms the A-MPDU as the frame aggregation size becomes larger. The previous works have studied the performance of each aggregation scheme and showed the throughput performance gained by aggregating multiple frames into a large frame. None

123

A. Saif et al.

of them studied the impact of the aggregation headers themselves on the aggregation. Moreover, the decent performance of the A-MSDU in clear channels motivates us to enhance its performance in erroneous channel by enabling retransmission control over its subframes. In this paper, we have analysed the overheads introduced by the aggregation headers and showed their impact on the performance especially aggregating small frames. Moreover, an MSDU aggregation scheme with optimized headers and retransmission control over the subframes has been proposed. The evaluation of the proposed scheme has been accomplished through simulations.

4 Motivation During 802.11n development, many proposals have been introduced to overcome the legacy 802.11 timing and headers overheads by means of aggregating multiple frames into a single frame. Only the A-MSDU and A-MPDU aggregations were adopted by the 802.11n working group. The A-MSDU was initially designed to be simple with a small aggregation size. It constructs the A-MSDU frame from the available frames despite their destination addresses. Moreover, to keep the A-MSDU design simple, many functions have been avoided such as multi-casting, broadcasting, multi-trafc frame aggregations, and selective retransmission over the subframes. Due to the lacks of selective retransmission and to avoid high frame loss in erroneous channels, the A-MSDU frame was limited to a maximum of 8 KB. However, A-MSDU aggregation shows an outstanding performance compared to the A-MPDU when they are used in clear channels and under the same aggregation size, especially when the aggregated subframes are small, thanks to the small aggregation headers [12,17]. The decent performance of the light headers A-MSDU in clear channels motives us to study the effect of the aggregation headers themselves, optimizes the current headers, and introduces selective retransmission over the subframes. Theoretically, it is possible to have subframes of different DA and SA in one A-MSDU as long as they all map to the same receiver address. Practically, however, A-MSDU collects frames of the same destination address and wraps them in a single frame ( MPDU) and then transmits the 802.11-wrapped frame [18]. Thus, disabling the multi-destination support from the A-MSDU by removing the repeated addresses with every subframe will contribute to reducing the A-MSDU headers overhead. Moreover, there are many scenarios cases where single-destination aggregation is favourable. These cases include the subframes security encryption, downlink ( AP to STA) frame transfers, and single-hop wireless transmission. With security encryption enabled, each subframe needs to be encrypted using the security association for each individual destination address, which means more overhead is added. However, only one encryption is needed for the whole A-MSDU if the whole constituent subframes are destined to the same destination. Moreover, in the wireless downlink the destination of any frame sent by the access point is a single station. The case also exists when we have a single-hop wireless network, only the frames of the destination hop should be aggregated in one A-MSDU. Thus, sending A-MSDU frame that contains subframes of different destinations in the downlink or in single-hop is a drawback rather than an advantage.

5 Analysis of the Headers Overhead Although frame aggregation significantly reduces the overheads introduced by the legacy MAC layer, there are many other aspects that need to be investigated in order to have a

123

An Enhanced A-MSDU Frame Aggregation Scheme Table 1 The notations used in the analysis Notation SA-MSDU SA-MPDU SAFR SMA-MSDU SMPDU SPPDU n sfhdr sfp sfd sfl sfsa sfda sffcs MAChdr FCS MPDUdel CMNhdr fraghdr fragfcs Desribtion A-MSDU size A-MPDU size AFR size mA-MSDU size MPDU size PPDU size Number of subframes/fragments A-MSDU/ mA-MSDU subframe header Subframe padding Subframe data Subframe length Subframe source address Subframe destination address Subframe check sequence MAC header Normal/HT control Frame check sequence MPDU delimiter The common header Fragment header Fragment check sequence Size(byte) 14/4 0..3 0..2304 2 6 6 2 36/40 4 4 1 8 2

more optimized frame aggregation. These aspects include the header size in comparison to the MSDU size and delay introduced due to unsaturated trafc. To the best of our knowledge, none of the previous works has investigated the inuence of the aggregation hearers themselves on the performance especially when the MSDU tends to be small. In the next section, we have analysed the headers of the 802.11n A-MSDU and A-MPDU as well as the AFR aggregation scheme from the related works. In this analysis, we have shown the relation between the headers and payload size for different aggregation sizes. The analysis is based on the MAC structure of the 802.11n and the compressed block ACK. Since we are concerned about the impact of headers, we have assumed an error free channel during our analysis. We also assumed that the subframes are always tted to a multiple of 4 (0 padding). The MAC header is included in our analysis since it is used in some aggregation schemes for sending the shared headers. The headers of the ACK frame are not addressed here since we are focusing on the headers of the aggregation schemes only. Table 1 shows the notations used in our analysis and their corresponding sizes. 5.1 Analysis of the A-MSDU Scheme Headers The A-MSDU frame consists of n subframes (sf ), each subframe has a header of 14 bytes, variable-size MSDU, and padding. The header consists of the source address destination address and length (L). The MAC header and FCS are appended to the packed A-MSDU and then submitted to the PHY layer for transmission. The MPDU that is constructed from an A-MSDU can be formulated as follows:

123

A. Saif et al.
60

50

MSDU=64B MSDU=128B MSDU=256B MSDU=512B MSDU=1024B MSDU=1500B

40

HDR[%]

30

20

10

0 2 4

16

32

64

No. of subframes
Fig. 3 The headers to the data ratio of the A-MSDU aggregation scheme

SMPDU = MAChdr + SAMSDU + FCS

n

= MAChdr +
i=1

(sfhdr (i) + sfd (i) + sfp (i)) + FCS

According to our assumption that the padding is zero (sfp = 0), then
n

SMPDU = MAChdr + FCS +

i=1

(sfhdr (i) + sfd (i))

Then the headers to the data ratio (HDR) of an A-MSDU carried in an MPDU can be expressed as: HDR = MAChdr + FCS + n sfhdr (i) i=1 n i=1 sfd (i)

Figure 3 shows the headers to the data ratio of different A-MSDUs that have n equal size MSDUs. From Fig. 3, the HDR ratio of the A-MSDU aggregation is inversely proportional to the MSDU size. For small MSDUs of 64,128 and 256 bytes the HDR reaches up to 57, 27, and 13%, respectively for an aggregation size of 2 subframes. The ratio then decreases with the increasing number of subframes. For large MSDUs of 1,024 and 1,500 bytes, the HDR ratio is small and almost constant for any number of subframes in the A-MSDU. In A-MSDU aggregation, the size of the MSDU plays a major rule in the performance since the repeated subframe headers contribute to increasing the header overhead especially for small MSDUs. Upon selecting the A-MSDU to aggregate small MSDUs, we had better aggregate as many MSDUs as possible taking into account the time constrains and delay limits of the participating applications. Thus, an optimized A-MSDU frame aggregation can be obtained if the subframe headers are minimized.

123

An Enhanced A-MSDU Frame Aggregation Scheme

5.2 Analysis of the A-MPDU Scheme Headers The A-MPDU frame consists of n subframes, each subframe consists of MAC header, delimiter of 4 bytes, variable-size MPDU, padding, and frame check sequence. The variable size MPDU can either be MSDU or A-MSDU, here we will assume only MSDUs are included in the A-MPDU frame.
n

SPPDU =
i=1

(MAChdr (i) + sfd (i) + sfp (i) + FCS(i) + MPDUdel (i))

where n is the number of subframes. By assuming that the padding is 0 (sfp = 0), then
n

SPPDU =
i=1

(MAChdr (i) + sfd (i) + FCS(i) + MPDUdel (i))

The HDR ratio of the A-MPDU can be expressed as follows: HDR =

n i=1 (MAChdr (i) + FCS(i) + MPDUdel (i)) n i=1 sfd (i)

Although the A-MPDU aggregation leads to a high throughput due to its capability to aggregate up to 64 subframes or 64 KB of payload, the HDR ratio will be large if the aggregated MSDUs are small. From Fig. 4, we can see that the HDR of the A-MPDU is not affected by the number of aggregated MPDUs, but by the size of the MPDU payload. In the A-MPDU aggregation, the subframe headers are almost constants (MAC headers and FCS) and hardly to be optimized, unless we go for a new MAC design that does not take into account the backward compatibility with the older 802.11 standards, and that what really happened to the 802.11n green eld MAC structure. The eld that we can play around is the delimiter, but the delimiter itself is only 4 bytes and its inuence is very small. Another way to optimize the A-MPDU aggregation headers is to avoid the repetition of a whole MAC header with every MPDU and that would be difcult since the A-MPDU was built to support different functionalities such as the group addressing, multi-TIDs and multi-casting. Thus,
70 60 50 MSDU=64B MSDU=128B MSDU=256B MSDU=512B MSDU=1024B MSDU=1500B

HDR[%]

40 30 20 10 0

0 2 4

16

32

64

No. of subframes
Fig. 4 The headers to the data ratio of the A-MPDU aggregation scheme

123

A. Saif et al.

Fig. 5 The AFR frame structure

A-MPDU with its current structure is accepted to aggregate the large MSDUs but not for small MSDUs. 5.3 Analysis of the AFR Scheme Headers Besides the analysis of the A-MSDU and A-MPDU aggregation schemes, we have analysed the aggregation with the fragment retransmission scheme (AFR) from the related works addressed in Sect. 3. The AFR scheme aggregates MSDUs from the upper layer, but before aggregation, the MSDUs are segmented into small fragments and then transmitted in a single MAC frame. Each fragment is attached with a fragmentation header and a fragment check sequence. Even though this scheme provides error control over the fragments, the headers added to each fragment increase the aggregation overhead. Moreover, fragmentation and de-fragmentation as well as keeping the fragments in sequence are tedious processes that increase the system delay. The structure of the AFR scheme is shown in Fig. 5, in addition to the MAC header and FCS, the AFR has introduced 8 bytes fragmentation header and 2 bytes check sequence for each fragment. SAFR = MAChdr + n (fraghdr + fragd + fragfcs ) + FCS = MAChdr + FCS + n (fraghdr + fragfcs ) + n fragd where n is the number of fragments. The HDR of the AFR scheme can be expressed as follows: MAChdr + FCS + n (fraghdr + fragfcs ) HDR = n fragd If the AFR segmentation threshold is larger than the MSDU size, the MSDU will be treated as a single fragment and then attached with the unnecessary headers. This appears clearly in Fig. 6, where the ratio is kept the same for an MSDU of size 64 under different fragmentation sizes, 64 bytes Fig. 6a, 128 bytes Fig. 6b, and 256 bytes Fig. 6c. The ratio reaches 43% if the aggregation size is 2 and reduces to 20% for an aggregation size of 8. On the other hand, if the segmentation threshold is smaller than the MSDU, the MSDU will be segmented into more than one fragment and then transmitted with the necessary fragmentation headers. In this case, the fragmentation headers are proportional to the actual MSDU size. unlike A-MSDU and A-MPDU, the HDR of the AFR still exists even for large MSDUs duo to the large number of fragments. For large MSDUs of 1,500 bytes the ratio of the headers overhead is kept almost at 14, 7 and 4% for a fragmentation of 64, 128 and

123

An Enhanced A-MSDU Frame Aggregation Scheme Fig. 6 The headers to the data ratio of the AFR aggregation scheme
50 MSDU=64B MSDU=128B MSDU=256B MSDU=512B MSDU=1024B MSDU=1500B

40

HDR[%]

30

20

10

0 2 4

16

32

64

No. of subframes (a) fragment size = 64 bytes

50 MSDU=64B MSDU=128B MSDU=256B MSDU=512B MSDU=1024B MSDU=1500B

40

HDR[%]

30

20

10

0 2 4

16

32

64

No. of subframes (b) fragment size =128 bytes

50 MSDU=64B MSDU=128B MSDU=256B MSDU=512B MSDU=1024B MSDU=1500B

40

HDR[%]

30

20

10

0 2 4

16

32

64

No. of subframes (c) fragment size = 256 bytes

123

A. Saif et al.

256 bytes respectively. More results are depicted on Fig. 6 under different MSDU sizes and different fragmentation sizes. Thus, for small MSDUs and small aggregation size, the ratio is large irrespective of the fragment size and decreases with the increasing of the MSDU size. The lowest ration is obtained when the fragment size is equal to the MSDU size which violates the AFR concept. In the next section, we have proposed a minimized headers MSDU aggregation scheme ( mA-MSDU). In this scheme the subframe headers are reduced and control bits are introduced in order to enables error control over the A-MSDU subframes at the MSDU level.

6 The mA-MSDU Aggregation Scheme Although frame aggregation provides a significant reduction on the legacy 802.11 headers, other headers are added in order to accomplish the aggregation and de-aggregation processes. These headers might become a source of overhead, unless they are optimized and reduced. Due to these headers, one aggregation might be benecial for certain MSDU sizes and becomes a bottleneck for other sizes depending on the aggregation size (number of subframes in the aggregation). From the previous analysis, we can see that A-MSDU aggregation headers are adequate for large MSDUs but they produce large header overheads for small MSDUs due to the attached header with every subframe. The mA-MSDU aggregation scheme minimizes the headers overhead by optimizing the subframes aggregation headers. It aggregates MSDUs that have the same destination address and then maps the MAC headers to subframes headers. It uses a common header for the shared subframess information. Moreover, aggregation at the MSDUs level does not provide error control and retransmission over its subframes. The mA-MSDU have introduced implicit sequence control (ISC) for the subframes based on their index in the mA-MSDU aggregation frame. The sequence control mechanism enables the scheme to retransmit the corrupted subframes in the next subsequent retransmissions, avoids using a sequence number with every subframe, and keeps the subframe ordering at the receiver side. 6.1 mA-MSDU Frame Format During the design of the mA-MSDU we have kept the structure of the actual MAC layer unchanged, we only used the variable payload of the MAC frame to build the aggregation. Figure 7 shows the mA-MSDU structure where the rst byte of the payload is assigned for the common header (CMNhdr ) and the remaining payload is assigned for the subframes. From now on we will use the term superframe for the aggregation frame. The common header is a one byte eld, its rst 6 bits represent the number of subframes in the superframe, thus 64 subframes can be addressed. The Lost packet (lp), is a one bit led with a default value of 0. It is set to 1 if the preceding superframe is dropped at the receiver due to exceeding the retransmission limit or due to lost ACK. The last bit is reserved for future extension. The subframe has a maximum size of 2,310 bytes and consists of three elds: the subframe header (sfhdr ), MSDU, and subframe check sequence (sffcs ). The MSDU has a variable size of up to the maximum 802.11 transmission unit (2,304 bytes). The aggregation size shall not exceed 7935 bytes which is dened for the A-MSDU in 802.11n. The two bytes (sffcs ) is used to check the integrity of the subframes. Upon failure of the subframe integrity check, the individual subframe will be marked for retransmission.

123

An Enhanced A-MSDU Frame Aggregation Scheme

Length (12b) Retry (1b) Flush (1b) Rsvd (2b) FCS (8b) Signature (8b)

No.subframes (6b)

Lost pkt (1b)

Rsvd (1b)

Subframe header (4B)

MSDU (0-2304B)

FCS (2B)

MAC Header

Common header (1B)

Subframe1 (2310B)

Subframe2

..

Subframen

FCS (4B)

Fig. 7 The mA-MSDU frame structure

The subframe header contains the control elds that are necessary for the de-aggregation at the receiver side. The 12 bits eld is used to express the size of the MSDU in the subframe. The retry bit will inform the receiver whether this subframe is a retransmitted subframe or not. If so, this subframe will be checked before being added to the receiver queue (RQ) to avoid the duplication. The ush bit is set when the subframe is a retransmitted subframe and its lifetime has been expired. Upon receiving a subframe with the ush bit of 1, the receiver will ush out the corresponding subframe from the RQ that have a status ag of 0. The one byte FCS is used to check the validity of the subframe header and the signature byte is used to align the de-aggregation in case of corruption in any of the subframes. We have used the same signature as it is dened in the 802.11n specication. 6.2 The mA-MSDU Aggregation Scheme Description The ideas behind the mA-MSDU are to reduce the headers of the subframes and enable subframe retransmission at the MSDU level. The MSDUs that are destined to the same destination will be aggregated and share the addresses of the holder MPDU. Only the number of subframes, length of the MSDU, and some status ags are associated with the MSDU in order to enable the de-aggregation at the receiver side. Only the corrupted subframes will be retransmitted and not the whole superframe as the conventional A-MSDU does. At the sender MAC layer, the received MSDUs from the upper layer are queued in a queue called transmitting queue (TQ). While constructing the superframe, only the MSDUs that have the same destination address as the head of the queue will be associated with the necessary aggregation header and then appended to the superframe. The index of the subframe in the superframe will be considered as a sequence number of that subframe and then the index led in the TQ of the corresponding MSDU will be updated accordingly. The MSDUs in the TQ that are not involved in the current superframe will have an index of 1. Upon receiving the superframe at the receiver side, the de-aggregation process will start. Based on the subframe check sequence, the subframes will be added to the RQ with a status ag of 1 if received successfully or 0 otherwise. If the RQ is full, the remaining subframes will be dropped and considered as if they were received with errors. The bitmap acknowledgement will be constructed according to the status ags of the subframes in the RQ and then sent back to the receiver. If the correctly received subframes are in correct order, their corresponding MSDUs will be forwarded to the upper layer and then removed from the RQ.

123

A. Saif et al.

At the sender side, the TQ will be updated according to the received bitmap. If the bit i in the bitmap is set to 1, the MSDU with index i in the TQ will be considered as received correctly and then removed, otherwise it will be considered as lost and will be retransmitted at the head of the next superframe. The lost superframes will be retransmitted according to the retry limits. If the retry limit is exceeded, the MSDUs in TQ that form the superframe will be dropped and the lost packet ag in the next superframe will be enabled in order to ush out the subframes that are unordered in the RQ. 6.3 Implicit Sequence Control According to the 802.11 standard the sequence number is attached to the frame as a part of the MAC header at the MPDU level. But such sequencing does not exist at the MSDU level. Therefore, to enable the MSDU retransmission we have introduced the implicit sequence control based on the index of the subframe that holds the MSDU in the current superframe. Since we can aggregate up to 64 MSDUs, the implicit sequence number will range from 0 to 63. The subframes will be constructed from the MSDUs in the TQ and then appended to the superframe according to their index. Similarly, the CBA bitmap will be constructed according to the index of the subframe in a way that the ith bit of the bitmap will acknowledge the subframe of the ith index in the superframe. In case if the channel is clear, the subframes will be forwarded to the upper layers whenever received by the receiver and no retransmission is required. But in erroneous channels, retransmission happens frequently and the retransmitted subframes are always put on the top of the next superframe followed by the new subframes. The retransmitted subframes will update the RQ in a way that, if the ith retransmitted subframe is correctly received then the ith entry in the RQ that have a status ag of 0 will be updated and the MSDUs of the ordered subframes in the RQ will be forwarded accordingly. Thus, using the implicit sequence number along with the retransmission of the corrupted subframes at the beginning of the next superframe will ensure the ordered forwarding of the received MSDUs to the upper layers. Figure 8 exhibits an example of the ISC mechanism under an erroneous channel where some subframes are corrupted in a superframe of size 8. The rst superframe has been received with two of its subframes are corrupted, sub2 and sub5, while only sub10 is corrupted in the retransmitted superframe. Based on the CBA, the sender will retransmit the corrupted subframes on the top of the next superframe. If the RQ is null, the received subframes will be appended to the RQ with their corresponding status 1 for correctly the received subframes and 0 for the corrupted ones. The CBA will be constructed accordingly and then send back to the sender. The sender will retransmit the corrupted subframes sub2 and sub5 on the top of the next superframe at index 0 and index 1 respectively, followed by the new subframes sub8 to sub13. Upon receiving the superframe, the rst subframe at index 0 will update the rst subframe in the RQ that have a status ag of 0, sub2 in this case. The second subframe at index 1 will update sub5 in the RQ. The MSDUs of the ordered subframes will be forwarded until a subframe with a status ag of 0 is met, in our case sub2 in the rst transmission and sub10 in the second. 6.4 Analysis of the mA-MSDU Headers Since our aggregation scheme is based on MSDUs, the mA-MSDU analysis resembles the A-MSDU analysis but with different subframe structure. The mA-MSDU subframe consists of a subframe check sequence, a more optimized subframe header and the MSDU itself.

123

An Enhanced A-MSDU Frame Aggregation Scheme

Sender Receiver

0 sub0

1 sub1

2 sub2

3 sub3

4 sub4

5 sub5

6 sub6

7 sub7

Flag

sub0 sub1 sub2 sub3 sub4 sub5

1 1 0 1 1 0 1 1 0 1 1 0 1 1 1 1

index

BAC(11011011)

sub0 sub1 sub2 sub3 sub4 sub5 sub6 sub7

0 1 2 3 4 5 6 7 Update 2nd with Flag=0

BAC(11110111)

0 sub2

1 sub5

2 sub8

3 sub9

4 sub10

5 sub11

6 sub12

7 sub13

sub6 sub7 sub2 sub3 sub4 sub5 sub6 sub7 sub2 sub3 sub4 sub5 sub6 sub7

Update 1st with Flag=0

sub2 sub5 sub8 sub9 sub10 sub11 sub12 sub13

0 1 2 3 4 5 6 7
Removed /Forwarded subframes In Queue subframes Retransmitted subframes

1 1 1 1 0 1 1 1

sub8 sub9 sub10 sub11 sub12 sub13

Fig. 8 The Sequence control and retransmission in the mA-MSDU scheme

A one byte common header is added to control the mA-MSDU de-aggregation at the receiver side.
n

SMPDU = MAChdr + CMNhdr + FCS +

i=1

(sfd (i) + sfhdr (i) + sffcs (i))

Then, the headers to the data ratio can be expressed as follows: H DR = MAChdr + cAgghdr + FCS + n (sfhdr (i) + sffcs (i)) i=1 n i=1 sfd (i)

Where n is the number of aggregated subframes. As illustrated in Fig. 9, the mA-MSDU behaviour is similar to the other aggregation schemes but with a smaller headers to data ratio. To aggregate 8 MSDUs of size 128 bytes each, the mA-MSDU headers to data ratio is 8.5% whereas it is 15, 34 and 16% for A-MSDU, A-MPDU and AFR(fragment size = 64 bytes) respectively, Figs. 3, 4 and 6a.

7 Performance Evaluation Simulation experiments have been conducted in order to evaluate the mA-MSDU performance. In the rst scenario, the headers to the data ratio of the mA-MSDU scheme is investigated and compared to the ratios of the A-MSDU, A-MPDU and AFR schemes. We have

123

Forwarded

1 1

Forwarded

A. Saif et al.
50

40

MSDU=64B MSDU=128B MSDU=256B MSDU=512B MSDU=1024B MSDU=1500B

HDR[%]

30

20

10

0 2 4

16

32

64

No. of subframes
Fig. 9 The headers to the data ratio of the mA-MSDU aggregation scheme Table 2 The simulation parameters Parameter TSIFS TDIFS Tphyhdr Tidle CWmin Data rate Basic rate Value 16 s 34 s 20 s 9 s 16 140 Mbps 54 Mbps

used the point to point simulation scenario 17 of the usage model [19]. The scenario includes a xed high throughput (HT) access point (AP) and a xed HT station (STA), both operating over a 20 MHz channel. We have used constant bit rate (CBR) UDP trafc to generate a variable load by varying the packet size and keeping the packet interval at 80 s. The trafc ows from the STA (source) to the AP (sink), only the MAC acknowledgement is sent in the reverse direction with the assumption that the channel is an error free channel and all the sent frames are received successfully without any retransmission. We run the simulation for 10 s using the ns-2 simulator [20]. Other simulation parameters are listed in Table 2. The simulation runs by xing the MSDU size and then changing the aggregation size (number of subframes) from 2 to 64. The simulation is repeated for different MSDU sizes ranging from 64 to 1,500 bytes, for the AFR scheme, we have set the fragment size to 64 bytes. The simulation results are depicted in Figs. 10 and 11. In the case of small MSDUs 64, 128 and 256 bytes, we have examined the headers to the data ratio of different aggregation sizes starting from 2 up to 64 MSDUs, the results are shown in Fig. 10. The mA-MSDU scheme produces headers reduction of up to 12, 6, and 3% over the A-MSDU scheme for MSDU size of 64, 128 and 256 bytes respectively. This behaviour is natural due to the doubling of the MSDU size while keeping the headers unchanged. Moreover, by xing the aggregation size to 2 subframes and changing the MSDU size to 64, 128, and 256 bytes the reduction in the mA-MSDU headers over A-MPDU reaches 28, 14, and 7% respectively. If the aggregation size is increased to 32, the reduction ratios increase to about 57, 29 and 14%, respectively,. This behaviour exists due to the linear relation between the A-MPDU aggregation size and the necessary aggregation headers.

123

An Enhanced A-MSDU Frame Aggregation Scheme Fig. 10 The headers to the data ratio of aggregating small MSDUs
90 80 70 60 mA-MSDU A-MSDU A-MPDU AFR

HDR[%]

50 40 30 20 10 0

0 2 4

16

32

64

No. of subframes (a) MSDU size = 64 bytes

50 mA-MSDU A-MSDU A-MPDU AFR

40

HDR[%]

30

20

10

0 2 4

16

32

64

No. of subframes (b) MSDU size =128 bytes

30 mA-MSDU A-MSDU A-MPDU AFR

20

HDR[%]
10 0 0 2 4

16

32

64

No. of subframes (c) MSDU size = 256 bytes

123

A. Saif et al. Fig. 11 The headers to the data ratio of aggregating large MSDUs
20 mA-MSDU A-MSDU A-MPDU AFR

HDR[%]

10

0 2 4

16

32

64

No. of subframes (a) MSDU size = 512 bytes

20 mA-MSDU A-MSDU A-MPDU AFR

HDR[%]

10

0 2 4

16

32

64

No. of subframes (b) MSDU size = 1024 bytes

20 mA-MSDU A-MSDU A-MPDU AFR

HDR[%]

10

0 2 4

16

32

64

No. of subframes (c) MSDU size = 1500 bytes

123

An Enhanced A-MSDU Frame Aggregation Scheme

50

40

mA-MSDU AFR Frag=64 AFR Frag=128 AFR Frag=256

HDR[%]

30

20

10

0 2 4

16

32

64

No. of subframes
Fig. 12 The mA-msdu and AFR headers to data ratio for different AFR fragmentation sizes

The AFR scheme with 64 bytes fragment size behaves nearly the same as the mA-MSDU for small MSDUs, but unlike mA-MSDU when increasing the MSDU size. For large MSDUs, the reduction of the HDR ratio in AFR is not decreasing by the same amount as the mA-MSDU does. When the MSDU size is doubled from 64 to 128 bytes at an aggregation size of 8 MSDUs, the AFR headers to data ratio is reduced from 20% to about 16%, whereas it is reduced from 17% to about 9% for mA-MSDU. Thus, mA-MSDU still has the smallest headers to data ratio among all the investigated schemes under the small MSDUs. For large MSDUs 512, 1,024 and 1,500 bytes, Fig. 11, the headers to the data ratio of the mA-MSDU, A-MPDU and A-MSDU schemes is relatively small ranging from 10% down to about 1% with an advantage for the mA-MSDU. Figure 11c shows that If the MSDU size is large (e.g. 1,500 bytes), the header overhead becomes incomparable to the actual MSDU size with a ratio of less than 3%, which can even be neglected. The AFR under 64 bytes fragmentation size still has headers to data ratio above 10% due to the large number of fragments. Figure 12 shows the mA-MSDU headers to data ratio compared to AFR for MSDUs of 256 bytes under different fragmentation sizes. The simulation is performed under fragmentation sizes of 64, 128, and 256 bytes while keeping the MSDU size at 256 bytes. The AFR headers to data ratio decreases as the fragment size increases. The best result for AFR is achieved when the fragment size is equal to the MSDU size which violates the goal of the AFR scheme. The mA-MSDU still scores the lowest headers to data ratio among the different AFR fragments. In the second experiment we have evaluated the throughput performance of the schemes under different channel conditions. The channel error rate is changed from a clear channel of zero bit error rate to a noisy channel of 104 bit error rate. We have set the data rate to 54 Mbps and the simulations time to 50 s. The number of stations is set to only 10 in order to reduce the impact of collision on the throughput. The system throughput is depicted in Fig. 13 for MSDUs of sizes 1,024 and 128 bytes. The gure shows the impact of the channel error on the performance of the aggregation schemes. Under high bit error rates, the mAMSDU, A-MPDU, and AFR schemes still survive due to their subframes/fragments error control capabilities while A-MSDU can barely transmit anything. Furthermore, the gure shows that the large MSDUs are inuenced by the large bit error rates more than the small

123

A. Saif et al.
70 60

mA-MSDU A-MSDU A-MPDU AFR

Throughput[Mbps]

50 40 30 20 10 0 0

10

-6

10

-5

10

-4

Bit Error Rate

(a) MSDU size = 128 bytes

100 90 80

mA-MSDU A-MSDU A-MPDU AFR

Throughput[Mbps]

70 60 50 40 30 20 10 0 0
-6 -5 -4

10

10

10

Bit Error Rate

(b) MSDU size = 1024 bytes

Fig. 13 The throughput versus bit error rate

MSDUs. From Fig. 13a we can infer that the large aggregation headers of the A-MPDU are behind its performance degradation when aggregating small MSDUs even in clear channels. The mA-MSDU outperforms the other schemes in terms of throughout under different bit error rates and different MSDU sizes. Under bit error rate of 105 , the mA-MSDU improvement over A-MSDU, A-MPDU, and AFR reaches 27, 28 and 3% respectively for MSDUs of size of 128 bytes and reaches 56, 6, and 7% for large MSDUs of size 1,024 bytes. These results demonstrate the significance of the mA-MSDU scheme under different channels conditions with different aggregation sizes. A significant improvement becomes clear when the MSDU tends to be small, which makes the mA-MSDU an appropriate aggregation scheme, specifically for the trafc classes that have small packet size such as VoIP.

123

An Enhanced A-MSDU Frame Aggregation Scheme

8 Discussion The frame aggregation is a magnificent method introduced by the 802.11n and other proposals for the next high-speed networks in order to reduce the overall headers of the legacy MAC distributed coordination function. However, due to the diversity of the applications and their trafc characteristics, the aggregation might not be benecial for some types of applications. The Aggregation under some applications might produce large headers and frame delay. From the headers to the data ratio point of view and based on our analysis and simulations, the aggregation of large MSDUs is slightly effected by the headers size and this affect might be neglected for very large MSDUs. However, for small MSDUs the issue is different, the aggregation headers will contribute to increasing the size of the transmitted frame. To aggregate small MSDUs, the selection of the aggregation scheme and the aggregation size need to be subjected to certain criteria depending on the trafc class and delay constraints. Therefore, either we need to develop an aggregation scheme that mainly support small MSDUs or to make the existing schemes adaptive in order to dynamically select the appropriate aggregation parameters based on the MSDU size, delay constraints, and trafc classes. Similar to other aggregation schemes, mA-MSDU aggregation and de-aggregation processes adds some processing overhead, however this overhead is relatively small compared to the headers and timing overheads that are diminished by the aggregation. Since our scheme re-engineers the A-MSDU aggregation, modications to both sender and receiver become essential. In mA-MSDU subframes manipulation based on the received bitmap ACK is performed at the sender side while bitmap construction and addresses mapping are performed at the receiver side. The mA-MSDU scheme has been tested under infrastructure wireless mode and singlehop adhoc networks. Moreover, the scheme in its current state xes the aggregation size to 8 KB and does not optimize the aggregation size based on the network condition. Thus, testing the scheme under multi-hope wireless networks and optimizing the aggregation size are left for future extension. Furturemore, the current mA-MSDU design does not support aggregation of multi-trafcs since all of its subframes share the same trafc identier (TID) of the common MAC header. Additionally, our scheme co-works with the A-MPDU and does not replace it since A-MPDU has more functions than mA-MSDU in its current state in terms of multicasting, broadcasting and multi-trafc support.

9 Conclusion In this work we have shown the impact of the aggregation headers as well as the channel condition on the performance of different aggregation schemes. We have addressed these issues by proposing the mA-MSDU aggregation scheme that introduces an implicit sequence control to enable error control over the aggregated subframes and minimize the aggregation headers. The analysis and simulation results show the benets of the mA-MSDU scheme on aggregating small MSDUs as well as large MSDUs with relatively small attached headers. Furthermore, the implicit sequence control and the retransmission mechanism of the mAMSDU improve the system performance in erroneous channels. The mA-MSDU shows a significant performance when aggregating small MSDUs which makes it an adequate aggregation mechanism for applications that have small frame size such as VoIP. In the future, the proposed scheme will be evaluated under self adapted trafcs such as TCP trafc as well as delay constraint applications such as VoIP and on-line gaming.

123

A. Saif et al. Acknowledgments 05/03/10/1039RU). This work was supported by the Research University Grant Scheme (RUGS Number:

References
1. IEEE P802.11n/D9.0. (Oct 2009). Draft amendment to wireless LAN medium access control (MAC) and physical layer (phy) specications: Enhancements for higher throughput. 2. Xiao, Y., & Rosdahl, J. (2002). Throughput and delay limits of IEEE 802.11. Communications Letters, IEEE, 6(8), 355357. 3. IEEE 802.11 WG. (September 1999). Part 11: Wireless lAN medium access control (MAC) and physical layer (PHY) specications: High-speed physical layer in the 5 Ghz band. IEEE std. 802.11a. 4. Heidemann, J., Sinha, R., & Papadopoulos, C. (October 2005). Internet packet size distributions: some observations. Technical report. 5. IEEE Std. 802.11e WG. (November 2005). Part 11: Wireless LAN medium access control (MAC) and physical layer (PHY) amendment 8: Medium access control (MAC) quality of service enhancements. 6. Yang, X., & Rosdahl, J. (2003). Performance analysis and enhancement for the current and future IEEE 802.11 MAC protocols. SIGMOBILE Mobile Computing Communications Review, 7(2), 619. 7. Xiao, Y. (2005). IEEE 802.11n: Enhancements for higher throughput in wireless LANs. Wireless Communications IEEE, 12(6), 8291. 8. Xiao, Y. (2004). Packing mechanisms for the IEEE 802.11n wireless LANs. In Global Telecommunications Conference, 2004. GLOBECOM 04. IEEE (Vol. 5, pp. 32753279). 9. Youngsoo, K., Choi, S., Jang, K., & Hwang, H. (2004). Throughput enhancement of IEEE 802.11 WLAN via frame aggregation. In Vehicular technology conference, 2004. VTC2004-Fall. 2004 IEEE 60th (Vol. 4, pp. 30303034). 10. Tianji, L., Qiang, N., Malone, D., Leith, D., Xiao, Y., & Thierry, T. (2009). Aggregation with fragment retransmission for very high-speed WLANs. IEEE/ACM Transactions on Networking, 17(2), 591604. 11. Riggio, R., Miorandi, D., De Pellegrini, F., Granelli, F., & Chlamtac, I. (2008). A trafc aggregation and differentiation scheme for enhanced QoS in IEEE 802.11-based wireless mesh networks. Computer Communications, 31(7), 12901300. 12. Yuxia, L., & Wong, V. W. S. (2006). Wsn01-1: Frame aggregation and optimal frame size adaptation for IEEE 802.11n WLANs. In Global telecommunications conference, 2006. GLOBECOM 06. IEEE (pp. 16). 13. Selvam, T., & Srikanth, S. (2010). A frame aggregation scheduler for IEEE 802.11n. In Communications (NCC), 2010 National Conference on (pp. 15). 14. Skordoulis, D., Ni, Q, Chen, H. -H., Stephens, A. P., Changwen, L., & Jamalipour, A. (2008). IEEE 802.11n MAC frame aggregation mechanisms for next-generation high-throughput WLANs. Wireless Communications, IEEE, 15(1), 4047. 15. Wang, C.-Y., & Wei, H.-Y. (2009). IEEE 802.11n MAC enhancement and performance evaluation. Mobile Networks and Applications, 14(6), 760771. 16. Ginzburg, B., & Kesselman, A. (2007). Performance analysis of A-MSDU and A-MPDU aggregation in IEEE 802.11n. In Sarnoff symposium, 2007 IEEE (pp. 15). 17. Kim, B. S., Hwang, H. Y., & Sung, D. K. (2008). Effect of frame aggregation on the throughput performance of IEEE 802.11n. In Wireless communications and networking conference, 2008. WCNC 2008. IEEE (pp. 17401744). 18. Cisco Systems white paper (2009). 802.11n: The standard revealed, cisco systems. 19. Stephens, A. P. et al. (May 2004). IEEE p802.11 Wireless LANs: Usage models. Technical report, IEEE 802.11n working document 802.11-03/802r23. 20. NS2. http://www.isi.edu/nsnam/ns/.

123

An Enhanced A-MSDU Frame Aggregation Scheme

Author Biographies
Anwar Saif received the B.S. degrees in computer engineering from the Technical university of Budapest, Hungary, in 1996 and the M.S. degree in networking and distributed computation from University Putra Malaysia, Malaysia (UPM), in 2008. Currently, he is working towards the Ph.D. degree at UPM. His research interests are VoIP over wireless networks and MAC protocols.

Mohamed Othman received his Ph.D. from the National University of Malaysia with distinction (Best Ph.D. Thesis in 2000 awarded by Sime Darby Malaysia and Malaysian Mathematical Science Society). Now, he is a Professor in Computer Science and Deputy Dean of Faculty of Computer Science and Information Technology, University Putra Malaysia (UPM) and prior to that he was a Deputy Director of Information Development and Communication Center (iDEC) where he was incharge for UMPNet network campus, uSport Wireless Communication project and UPM DataCenter. In 2002 till 2009, he received many gold and silver medal awards for University Research and Development Exhibitions and Malaysia Technologies Exhibition which is at the national level. His main research interests are in the elds of parallel and distributed algorithms, high-speed networking, network design and management (network security, wireless and trafc monitoring) and scientic computing. He is a member of IEEE Computer Society, Malaysian National Computer Confederation and Malaysian Mathematical Society. He already published more than 110 National and International journals and more than 200 proceeding papers. He is also an associate researcher and coordinator of High Speed Machine at the Laboratory of Computational Science and Informatics, Institute of Mathematical Science (INSPEM), University Putra Malaysia.

Shamala Subramaniam received the B.S. degree in Computer Science from University Putra Malaysia (UPM), in 1996, M.S. (UPM), in 1999, Ph.D. (UPM) in 2002. Her research interests are Computer Networks, Simulation and Modeling, Scheduling and Real Time System.

123

A. Saif et al. Nor Asila Wati Abdul Hamid is a senior lecturer at the Department of Communication Technology and Networks, Faculty of Computer Science and Information Technology, Universiti Putra Malaysia, Malaysia. She received her Ph.D. from the University of Adelaide in 2008. Her research interests are in parallel and distributed computing, cluster computing and other applications of high-performance computing. She is also an Associate Researcher and Coordinator of High Speed Machine at Institute for Mathematical Research (INSPEM), Universiti Putra Malaysia.

123