Whole Page Performance

1
Whole Page Performance

• Leeann Bent and Geoffrey M. Voelker
• University of California, San Diego

2
Whole Page Performance?

• Extensive previous work on how specific
  techniques affect individual object download.
• Caching, Prefetching, CDNs, DNS caching.
• However, user downloads pages of objects.
• Not clear how individual object performance maps
  onto whole page performance
• Goal Study whole page performance
• Extent to which different optimizations are used
• Effect on downloading whole pages of objects

3
Related Work

• Krishnamurthy and Wills99 look at
• Parallel (HTTP1.0), persistent and pipelined
  connections.
• In addition to caching, range requests, and
  content placed on different servers.
• Top-level pages of popular sites.
• Focus on pages where all optimizations used.
• Our Study
• Follow on, with a different perspective.
• Use real user workloads.
• All pages, not just top level pages on popular
  servers
• Not all pages use optimizations
• Base page embedded objects.
• Connection optimizations CDNs DNS.

4
Overview

• Introduction
• Methodology
• Results
• Conclusion

5
Methodology Overview

• Use Medusa to
• Record everyday browsing from six users over four
  days.
• Replay traces toggling performance options
• Parallel Connections
• Using CDNs
• Complete DNS caching
• Persistent Connections
• Compute download costs for whole pages

6
The Medusa Proxy

• Internet

7
Page Download Time

• Page download time
• Time required to download base page and all
  embedded objects.
• Reflects user-perceived web performance
• Calculated using object download time.
• Determine object download time from just after
  DNS lookup to connection close or full object
  return (persistent).
• Incorporate original recorded DNS times where
  appropriate.

8
Example
Individual Object Times
Page Download Times
854 ms
Serial
Parallel (2 conns)
259 ms
580ms
580 ms
9
Traces

• Six users April 27 - 30 (Sat. - Tues.).
• Originally 22,228 objects and 1,455 pages.
• Remove error pages.
• Replay data gathered May 6-7 (Mon - Tues) June
  22-27 (Sat. Thurs.).
• Minimize warming effects by taking median of 5
  consecutive page downloads.

10
Optimization Combinations

• Parallel Connections (1)
• Medusa tracks number of concurrent connections
  used during trace.
• Used to replay parallel download.
• CDN Usage (2)
• When no CDN usage, remove CDN references.
• Replace with references to origin servers.
• When CDN usage enabled, traces left intact.
• DNS Caching (3)
• Simulate ideal DNS caching by excluding DNS time.
  
• Normal DNS add original DNS lookup times from
  trace.
• Persistent Connections (4)
• Use whichever protocol (1.0/1.1) recorded in
  original trace.

11
Overview

• Introduction
• Methodology
• Results
• Conclusion

12
Whole Page Optimizations

• Parallel gives large improvement.

• CDN improvement small.
• 2.5

• DNS improvement consistent.
• 7.4
• 6.7

• Persistent connections not as helpful as
  expected
• 1.5

13
Overall Trace Conclusions

• Parallelism has the greatest effect.
• Parallelism used aggressively on all pages.
• All other options provide incremental benefits.
• Does not mean other optimizations dont work.
• Some overheads may be relatively small.
• Average over all pages.
• Not all pages implement all optimizations.
• We dont simulate more aggressive use of options
  than found in original trace.
• A closer look

14
Ideal DNS Caching

• Average DNS costs
• Per object 7.1 ms
• Per page 529 ms
• DNS improvement moderate across the board.
• 5 14 improvement across all pages.
• Provides moderate benefit to all pages.
• Not all objects require full DNS lookups
• Already effective DNS caching in traces

15
Objects Per Page

• We would expect some other optimizations to have
  a greater effect (e.g. persistent connections).
• Looking at all pages in trace doesnt tell the
  whole story.
• Less opportunity for connection optimizations on
  small pages.
• Page with one object counts as much as a page
  with 152 objects.
• Optimizations more effective on a page with 152
  objects.
• Separate out effects of optimizations in pages
  with different numbers of objects
• Median number of objects per page is 5.
• Average number of objects per page is 15.

16
Page Breakdown

• 1-5 objects
• 1 21
• 2-5 63

• 6 objects improvements.
• 6-15 157
• 16 183

• Persistent
• 1.95
• 18.5

17
Page Breakdown Conclusions

• Performance optimizations dependent on number of
  objects per page.
• Optimizations more effective when more objects
  per page.
• Especially connection optimizations.
• Single object pages see moderate improvement.
• Can usually only benefit from DNS caching and
  CDNs.
• Persistent benefit only if on same server as
  previous page.
• And 26 of pages had one object

18
Persistent Connections

• Still dont see a whole lot of improvement for
  persistent connections.
• Expected to see more benefit for 16 objects.
• Not all pages use persistent connections.
• 20 of pages in our trace use them (229 pages).
• 2211 objects or 16.1.
• 9.65 objects per page.
• Look at only pages that contain persistent
  connections.

19
Persistent Connections

• Persistent connections useful if
• Many objects downloaded over persistent
  connections in the original trace.
• Objects downloaded from few servers.
• For pages lt 6 objects
• 2 out of 3 downloaded with persistent
  connections.
• Average page size 3.
• On average, 1.32 persistent objects per server.
• For pages gt 16 objects
• Average 18 objects with persistent connections.
• On average, 3.92 persistent objects per server.

20
Mostly Persistent Pages

• Know what it takes to see persistent
  optimization improvement
• Look at large pages where persistent connections
  used extensively (gt50 of objects).

• Pages that can benefit, do
• 6 objects improve 33-50.

21
CDN

• Previous study showed CDNs highly effective for
  individual objects. Koletsou01
• What is effect on whole page performance?
• Few pages with explicit Akamai-hosted objects.
• 48 pages or 5.2 of pages.
• 216 objects or 1.6 of total downloaded objects.
• Average of 4.5 CDN objects per page.
• Looked at CDN only page improvements
• CDNs improve CDN containing pages 6 - 30.

22
Conclusions

• Parallel connections have greatest impact.
• Universally applicable and easy to implement.
• Other options give incremental performance across
  all pages.
• Some optimizations provide consistent, but
  moderate, improvement across all pages.
• Some optimizations are not implemented on all
  pages.
• Provide benefit when used extensively.

23
Conclusions

• Can we draw correlation between object and
  real-world whole page performance?
• Depends.
• Not all optimizations widely used.
• When optimizations are used to full advantage,
  they are effective.

24
Medusa Available

• http//ramp.ucsd.edu/lbent/Medusa/index.html

25
The End
26
Medusa Proxy Functionality

• Trace and Replay
• Record requests and replay.
• Parallel connections.
• Persistent connections.
• Transformation
• CDN/no CDN replay.
• Performance Measurement
• Request latency.
• DNS overhead.
• Optimization options
• Use parallel connections.
• Use persistent connections.
• HTTP 1.0 and HTTP 1.1.
• Always attempt, never attempt, mirror trace
  attempt.

27
Page Delimitation

• Determining pages
• Necessary for
• Calculating total page costs.
• Limiting optimizations to within one page.
• Parallel Connections.
• Can analyze page and draw object dependencies.
• High overhead
• May impact user
• Use inter-object times in the original trace
  data.
• Use 2 second inter-object times.

28
Akamaized URLs

• Akamai accounts for 85-98 of CDN hosted objects
  ref.
• Will not account for sites completely hosted on
  Akamai hosts.
• Filter
• http//a1964.g.akamai.net/f/1964/2730/1h/app.whenu
  .com/image.gif
• http//app.whenu.com/image.gif

29
Interleaved Requests

• Requests may get interleaved when recorded in
  parallel mode and replayed in serial mode
• E.G.
• Connection 0 requests www.cnn.com,
  www.cnn.com/style.css.
• Connection 1 requests ar.atwola.com.
• Requests may be ordered in trace as
• www.cnn.com, ar.atwola.com, www.cnn.com/style.css.
  
• Negates benefit of parallel connections.

30
Page Characterization Objects per Page
31
Object Types

• Identified object type by clues in URL
• 80 of URLs images (.gif, .jpg).
• 5.6 html file (.htm, .html).
• 3.8 cgi, perl or javascript (?,.pl, .class).
• 3.3 javascript (.js).
• 3.6 unidentified (no suffix, pdf, txt, etc).

32
Persistent Connection/Brower

• Persistent connections appear correlated with
  browser
• IE - 12 pgs, 15.8 objs.
• Netscape - 19.5 pgs,10.0 objs.
• Omniweb - 66.0 pgs, 72.4 objs.
• Mozilla 5.0/Gecko - 95.8 pgs, 91.3 objs.

33
Persistent Connection Pages

• Still not as improved as expected
• Better than for only large pages
• Serial 7.28 vs. 1.98
• Parallel 24.03 vs.18.5
• Medians dont show improvements in all cases.

34
Mostly Persistent Pages
35
Persistent Connections per Page
36
Same as previous 16
37
Ad-Servers

• Identified by identifying hosts that were named
  with the phrases ads and adserver.
• YES http//rmads.msn.com/images_47144_date_0429_5
  0.jpg.
• NO http//graphics4.nytimes.com/ads/scottrade_sov
  .gif.

38
Ad-Servers and DNS

• Number of pages with ad-servers.
• 9.5 of pages, 1.53 of total objects.
• Average of 2.4 ads per page.
• Objects not hosted on content server.
• DNS lookup may be large part of lookup cost.
• DNS caching doesnt give great improvement
• DNS caching improves parallel case 10.9.
• Compared with 12.2 over all pages.
• DNS caching improves parallel, persistent case
  8.
• Compared with 6.3 over all pages.
• DNS caching improves parallel, persistent w/ CDN
  4.7.
• Compared to 6.3.