Diapositiva 1

1
APPLICATION PARAMETERS OPTIMIZATION TO GUARANTEE
QoS IN E-HEALTH SERVICES
I.Martínez, J. García, E. Viruete, J.Fernández
Communications Technologies Group (GTC) - Aragon
Institute of Engineering Research (I3A)
imr_at_unizar.es University of Zaragoza. Ada Byron
Building. Campus Río Ebro - c/María de Luna
3, 50018. Zaragoza (Spain)

• Description of the evaluation scenario
• The implementation of a new e-Health service
  requires a QoS technical evaluation to study its
  behaviour under different conditions. The main
  characteristics of this evaluation (see Fig. 1)
  are
• Evaluation scenario communications among health
  professionals of different specialties.
• e-Health services cooperative work, shared
  medical applications, remote diagnosis, etc.
• Type of information multimedia applications
  based on TCP (biomedical information transfers
  and medical files retrieval) and UDP
  (videoconference, on-line biosignals
  transmission).
• Evaluation parameters (see Fig. 2) TCP data
  (SMSS), UDP data (S), UDP packet size (s), burst
  size (MBS), data rate (1/Dt), delay (EED), loss
  rate (PLR) capacity (C), bandwidth (BW).
• Use Cases (UCs)
• UC1. The health professional sends biomedical
  data to the hospital in Store-and-Forward (SF)
  mode. This SF.Data service includes medical tests
  and patient-related information.
• UC2. Including UC1, the health professional also
  queries the hospital database to manage the
  Electronic Patient Report (EPR) through a
  real-time (RT) web connection (RT.EPR).
• UC3. Including UC2, the health professional
  establishes a RT multimedia conference (including
  audio and video) with other specialist to support
  the diagnosis (RT.Media).
• UC4. The health professional often requires
  acquiring and sending specific biomedical signals
  (ECG, ECO, BP, Sp02) to complete the patient
  diagnosis (RT.Bio).

Fig.1 Evaluation scheme and use cases (Ucs) of a
rural e-Health service
Fig.2 Specific RT and SF parameters considered
in this study

• Results. UC1 and UC2 (TCP services)
• SMSS analysis. Using results previously obtained,
  UC1 and UC2 were evaluated with the QoS network
  conditions recommended by ITU PLRlt0.20,
  EEDlt180s. The evaluation includes specific test
  sizes (S) for the main medical practices
  computing radiography, electrocardiography,
  ecography, etc (see Fig.3).
• The results show that T decreases when SMSS
  rises, and in a nonlinear way with the loss
  variability T remains constant with a low PLR
  level (PLRlt0.05), but T notably increases (31)
  when PLR rises. Since large packets fit the link
  capacity better, T decreases with higher SMSS
  values (but the retransmission percentage
  increases, due to the reactivation algorithm
  regardless of packet size).
• EED analysis keeps this compromise (see Fig. 4)
  it is almost constant without PLR (independently
  of SMSS) but rises when SMSS does (due to SS
  algorithm reactivations with lower SMSS values
  that empty the buffer and reduce EED).
• BW analysis. The available BW (ABW) depends on
  PLR level (see Fig. 5) with low SMSS values, BW
  decreases down to ABW1.4Mb/s in the worst case
  (with SMSS512B and PLR0.20).
• In summary, all these tendencies recommend the
  use of low SMSS values in order to avoid using
  specific priority allocation methods and allow
  sharing network resources with UDP services, as
  it is evaluated in the next section.

(a) PLR lt 0.10 (b) PLR lt 0.03 Fig. 3
Transmission time regarding SMSS with constant
PLR
Fig. 4 EED regarding SMSS
with variable PLR
Fig. 5 ABW regarding SMSS with variable PLR
Results. UC3 and UC4 (UDP services) The UC3 and
UC4 specific evaluation parameters are data size
S4k, 2k, 1k, 500 (B), a variable number (N) of
simultaneous user connections (with user rate
r64, 32 (kb/s)), buffer size range of
Q?5,15, and two groups of sizes large packets,
sHi1472, 1380 (B), and small packets,
sLi512, 240 (B).
Table I. Evolution of PLR as a function of Q
with different S, N combinations and two groups
of packet sizes (a) large packets (b) small
packets

• PLR analysis. The combinations that fulfill the
  PLR threshold (see Table I, for sH11472B (left)
  and sL2240B (right) and r164kb/s) are
• - Large packets Q?10 (with N4) and Q?6 (with
  N2).
• - Small packets Q?15 (with N4) and Q?15 (with
  N2).
• EED analysis. The variation of C shows (see
  Fig.6a)-(b) with S14kB and N4) when the QoS
  thresholds are guaranteed
• - Large packets if C?128kb/s, Q?5 if
  C?192kb/s, Q?7
• if C?256kb/s, Q?9 if C?384kb/s, Q?12.
• - Small packets if C?128kb/s, Q?8 if
  C?192kb/s, Q?10
• if C?256kb/s, Q?11 if C?384kb/s, Q?15.
• This analysis is completed with r232kb/s, for
  the same Si variation range. Fig. 6(a)-(c) shows
  a example with S14kB and sH11472B. The new EED
  (r232kb/s, N8) is lower than the previous
  (r164kb/s, N4), even with the same information
  amount. Moreover, it changes the QoS thresholds
  C192kb/s is now useful with Q?9 (previously
  Q?7), and C384kb/s is useful with 12?Q?15
  (previously only Q?12). Thus, low burst sizes
  fulfill better QoS.

(a) sHi altos, r164kb/s (b) sLi bajos,
r164kb/s (c) sHi altos, r232kb/s Fig. 6 EED
with variable si y rk regarding Q, focusing on
the most critical areas closer to the QoS
thresholds (EEDlt200ms)
Conclusions
The results obtained show the best performance
range with small TCP packets and buffer sizes
lower than 15 packets for the UDP services. These
values are recommended in further designs of
e-Health services to guarantee the specific QoS
requirements. Figure at right shows an example of
how the adaptive selection of the buffer size
would be (buffer is initially sized with Q15,
10, or 6 depending on small packets and N4/N2
(step 1a), large packets and N4 (step 1b) or
N2 (step 1c) with C256kb/s. If C192kb/s, it
can be reconfigured to Q12, 8, or 5, (steps
2a, 2b or 2c).