Transcript Network-Level Impacts on User

Network-Level Impacts
on User-Level
Web Performance
Carey Williamson
Nayden Markatchev
University of Calgary
July 2003
SPECTS 2003
1
Introduction
“The Web has been both a blessing and a curse.”
-- CLW 2001

Blessing


made the Internet available to the
masses
 shields users from the low-layer
technical details of networking
 provides seamless exchange of
information, in a timeindependent, locationindependent, and platformindependent manner
July 2003
SPECTS 2003
Curse

made the Internet available to the
masses
 placed a lot of stress on the
Internet infrastructure
 traffic volume, sustained growth
 demands on the TCP/IP protocol
suite (i.e.,TCP is not really a good
“fit” for Web traffic demands)
2
Related Work: TCP and the Web







Persistent-connection HTTP [Mogul 1995]
Larger TCP initial window size [Allman et al 1998]
TCP “fast start” to reduce Web transfer latency
[Padmanabhan/Katz 1998]
Parallel (concurrent) TCP connections supported
in most Web browsers today (e.g., 4)
Ensemble-TCP to manage aggregation of TCP
connections to same dest. [Eggert et al 2000]
Rate-based pacing of TCP packets for the Web
[Aggarwal et al 2000] [Ke/Williamson 2000]
Context-aware TCP/IP [Williamson/Wu 2002]
July 2003
SPECTS 2003
3
Motivation

Most of the current Web performance literature
is focused on either:
 Web
caching simulation studies (i.e., with an
application-layer view, focusing on hit ratios, but
ignoring network-level issues and protocol effects); or
 TCP performance studies (i.e., packet-level studies,
but often focusing on throughput for bulk transfers,
rather than response times for (short) Web transfers)

Our Objective: To explore the relationships
between TCP, network-level effects, Web
caching, and user-perceived Web response time
July 2003
SPECTS 2003
4
Research Methodology Overview
Network simulation (ns2)
 Synthetic Web workloads (WebTraff)
 Simple network model:

 two-level
Web proxy caching hierarchy
 settable parameters for link capacity,
propagation delay, cache hit ratio, etc
Packet-level simulation study (TCP Reno)
 Performance metric: object transfer time

July 2003
SPECTS 2003
5
Network Model
Web
Server
C3
C2
d3
Proxy2
d2
Proxy1
C1
d1
Clients
July 2003
SPECTS 2003
6
Network Model
Web
Server
(Hit at Proxy1)
Proxy2
Proxy1
Clients
July 2003
SPECTS 2003
7
Network Model
Web
Server
(Hit at Proxy2)
Proxy2
Proxy1
Clients
July 2003
SPECTS 2003
8
Network Model
Web
Server
(Download from server)
Proxy2
Proxy1
Clients
July 2003
SPECTS 2003
9
Simulation Model Assumptions






Two-level Web proxy caching hierarchy
All Web content is cacheable static content
Data transfers are unidirectional toward the
clients (i.e., we ignore the HTTP request step)
One-way TCP model (i.e., models the data
transfer only, using DATA/ACK; no SYN/FIN)
TCP Reno, with segment size of 512 bytes
Proxy caches behave as store-and-forward
routers (on a per-packet basis)
July 2003
SPECTS 2003
10
Simulation Methodology

Multi-step process:
 Workload
generation using WebTraff (makes a
time-ordered sequence of 5000 Web object
transfer sizes, with desired request arrival rate)
 Modify workload file to randomly associate
transfers with either Proxy1, Proxy2, or Server
based on desired cache hit ratios (HR1, HR2)
 Use the ns2 network simulator to model the
TCP transfers on the desired network model
July 2003
SPECTS 2003
11
Experimental Design
Factors
Levels
Link Capacity C (Mbps)
10, 100, 1000
Propagation Delay d (msec)
1, 5, 10, 30, 60
Request Arrival Rate (req/sec)
10, 20, 40, 80
Child Proxy Hit Ratio HR1
20%, 30%, 40%
Parent Proxy Hit Ratio HR2
7%, 10%, 15%
Network
Parameters
Workload
Parameters
Full-factorial experiment (540 possible combinations)
Performance metric: TCP transfer time for each Web object
download (plotted versus transfer size)
July 2003
SPECTS 2003
12
Web Workload Model



5000 HTTP transfers synthetically generated by the
WebTraff tool [Markatchev/Williamson 2002]
Poisson arrival process assumed for Web requests
Four different request arrival rates considered:




July 2003
Light: 10 req/sec (approx. 0.77 Mbps offered load)
Moderate: 20 req/sec (approx. 1.54 Mbps offered load)
Medium: 40 req/sec (approx. 3.08 Mbps offered load)
Heavy: 80 req/sec (approx. 6.16 Mbps offered load)
SPECTS 2003
13
Web Workload Characteristics
July 2003
SPECTS 2003
14
Baseline Scenario

Link Capacity
 C1

Propagation Delay
 d1

= C2 = C3 = 10 Mbps
= 1 msec; d2 = 5 msec; d3 = 30 msec
Hit Ratios
 HR1

= 40%; HR2 = 15%
Request Arrival Rate
 Light:
July 2003
10 requests/sec
SPECTS 2003
15
Simulation Results (Baseline Scenario)
July 2003
SPECTS 2003
16
Simulation Results (Baseline Scenario)
July 2003
SPECTS 2003
17
Simulation Results (Baseline Scenario)
July 2003
SPECTS 2003
18
Simulation Results (Baseline Scenario)
“slower”
“faster”
July 2003
SPECTS 2003
19
Simulation Results (Baseline Scenario)
Queueing Delays
July 2003
SPECTS 2003
20
Simulation Results (Baseline Scenario)
Packet Losses/Retransmissions
July 2003
SPECTS 2003
21
Results Interpretation

TCP slow start is evident (for large RTT)
 The
“width” of steps increases exponentially
 The vertical separation reflects propagation
delay component of RTT

Queuing delays, packet losses, timeouts,
and retransmissions manifest themselves
as deviations from the normal structure
July 2003
SPECTS 2003
22
Effects of Network Link Capacity

To model current network infrastructures, we
considered four sets of link capacities:
 C1
=10 Mbps, C2 =10 Mbps, C3 =10 Mbps (baseline)
 C1 =100 Mbps, C2 =10 Mbps, C3 =10 Mbps
 C1 =100 Mbps, C2 =100 Mbps, C3 =10 Mbps
 C1 =1000 Mbps, C2 =100 Mbps, C3 =10 Mbps

This models increasingly faster client network
access to the Internet, while the WAN backbone
to the server remains slow (10 Mbps)
July 2003
SPECTS 2003
23
Results for Link Capacity
C1 = 10 Mbps (baseline)
July 2003
SPECTS 2003
24
Results for Link Capacity
C1 = 100 Mbps (upgrade)
July 2003
SPECTS 2003
25
Effect of Propagation Delay

Values for propagation delay
 d1
= 1 msec, d2 = 5 msec, d3 = 30 msec
 d1 = 1 msec, d2 = 5 msec, d3 = 60 msec
 d1 = 1 msec, d3 = 10 msec, d3 = 30 msec
 d1 = 1 msec, d2 = 10 msec, d3 = 60 msec

Representing LAN, MAN, WAN scenarios
July 2003
SPECTS 2003
26
Propagation Delay (d2 = 5 msec)
July 2003
SPECTS 2003
27
Propagation Delay (d2 = 10 msec)
July 2003
SPECTS 2003
28
Effect of Request Arrival Rate

Vary the offered load:
 10
requests/sec
 20 requests/sec
 40 requests/sec
 80 requests/sec

Makes network more and more congested
July 2003
SPECTS 2003
29
Effect of Request Arrival Rate
(Light Offered Load: 10 req/sec)
July 2003
SPECTS 2003
30
Effect of Request Arrival Rate
(Moderate Offered Load: 20 req/sec)
July 2003
SPECTS 2003
31
Effect of Request Arrival Rate
(Medium Offered Load: 40 req/sec)
July 2003
SPECTS 2003
32
Effect of Request Arrival Rate
(Heavy Offered Load: 80 req/sec)
July 2003
SPECTS 2003
33
Effect of Cache Hit Ratio

Vary the Cache Hit Ratio at each of the Web proxy
caches in the simulated network




“Good”: HR1 = 40%, HR2 = 15% (baseline)
“Average”: HR1 = 30%, HR2 = 10%
“Poor”: HR1 = 20%, HR2 = 7%
Assess user-perceived Web response time for fairly
realistic range of possible cache hit ratios, and
consideration of “cache filter effects”
July 2003
SPECTS 2003
34
Effect of Cache Hit Ratio
(“Good” HR1 = 40%; HR2 = 15%)
July 2003
SPECTS 2003
35
Effect of Cache Hit Ratio
(“Average” HR1 = 30%; HR2 = 10%)
July 2003
SPECTS 2003
36
Effect of Cache Hit Ratio
(“Poor” HR1 = 20%; HR2 = 7%)
July 2003
SPECTS 2003
37
Effect of Cache Hit Ratio
July 2003
SPECTS 2003
38
Effect of Cache Management Policy

Suppose that the two caches
are coordinated using a sizebased thresholding policy
One cache for “small” items
One cache for “large” items
Is this a good idea?

Scenario considered:




Child Proxy: items <= 8 KB
 Parent Proxy: items > 8 KB
 Same hit ratios as in baseline
July 2003
SPECTS 2003
39
Cache Management Policies
(Default Policy; C1 = 10 Mbps)
July 2003
SPECTS 2003
40
Cache Management Policies
(Threshold Policy; C1 = 10 Mbps)
July 2003
SPECTS 2003
41
Cache Management Policies
(Default Policy; C1 = 100 Mbps)
July 2003
SPECTS 2003
42
Cache Management Policies
(Threshold Policy; C1 = 100 Mbps)
July 2003
SPECTS 2003
43
Summary of Simulation Results
for Cache Management Policies
July 2003
SPECTS 2003
44
Summary and Conclusions
Packet-level network simulation study of
TCP effects on user-perceived Web perf.
 Link capacity, propagation delay, network
congestion, and TCP protocol behaviors
can all have significant impact on the
user-perceived Web response time
 Relationship between Web cache hit ratio
and user-perceived response time tricky
 Cache management and placement hard!

July 2003
SPECTS 2003
45
The End!

Thanks for your attention!

For more information:
 Email:
July 2003
{nayden,carey}@cpsc.ucalgary.ca
SPECTS 2003
46