66803 - questions about cache

Status: RESOLVED FIXED

(Core :: Networking: Cache, defect)

Description

[HJ]

I filed this bug report because I would like to know something more indeed
about:cache It's not a real bug report but more an collection of questions. If
you do know bugs related to one of the questions here, I would like to know the
bugnumber(s). 

1 - is it normal that Mozilla has entries in about:cache with Content Length: 0?
I see a lot of them in my about:cache and all of them are image related.
2 - when should I see entries in mem:cache? 
I never ever seen a single entry in it.
3 - when a site uses image rollovers, I see a lot of images been cached with
Flags set to partial. Should Mozilla cache broken images?
4 - the information in about:cache, how important is that? I mean does the
source use those values in a way for other purposes?
5 - when I visit a website with more then 512 images, what exactly does Mozilla
then? I have 2GB memory so I can set cache to a real big value (most images are
small).
6 - how big can the disk cache and mem cache be?
7 - when I visit a site, I often see that the about:cache information is out of
date. I have to restart Mozilla first to update the information in about:cache.
8 - when I set disk and mem cache to 0 the pages are loading fast
9 - I use a satellite connection (vpn on Astra sat) for downstream and a cable
connection for the uplink station in Luxemburg. It seems that Mozilla only has
all sort of timer related problems. Netscape 4.76, Opera 4/5 and all MSIE
version are working well. But if I switch disk and mem cache off, then it seems
to work pretty well. Does the new cache implementation correct this problem in
the near future?

Thank you very much for your time, HJ

Comment 1

[gordon]

Here are some answers, more will follow.

#1: cc'ing darin for expected HTTP behavior.

#2: The memory cache is currently only used by IMAP.  If you aren't using IMAP 
for mail, then you shouldn't see any mem cache entries.  I believe we have some 
open bugs on this however; the memory cache may be using some memory when it 
shouldn't be.

#3: cc'ing Chris Saari and Pavlov for more info.

#4: The information in about:cache is derived from data structures kept by the 
cache itself.  It may be out of date by the time you actually view the 
about:cache page however.

#5: The image lib cache is currently a separate cache (not the disk cache, not 
the memory cache).  It stores all of its entries in memory (but not the same 
memory allocated to the memory cache).  We have plans to integrate the image lib 
cache and the memory cache.  The limit of 512 entries applies to the disk cache.  
An image could be in the image lib cache, after it has been evicted from the disk 
cache.  Chris Saari or Pavlov would have more precise information on the current 
image lib cache.

6# The disk and mem cache can be as large as you have room to make them, however 
the disk cache is currently limited to 512 entries, and the mem cache is only 
used by IMAP.  We plan to expand or remove these limitations, and extend the use 
of the memory cache to HTTP, ImgLib, and other clients.

#7: How are you determining that the about:cache information is out of date?  I 
need a bit more information before I can address this issue.

#8 Pages will initially load faster with no cache because it takes time to fill 
the cache.  You would only see a time savings on revisiting a site that has an 
unexpired entry already in the cache.  If the entry has expired, we would still 
have to refetch it from the net, even though it was in the cache.  Due to 
inefficiencies in the current cache design, it's possible the cache provides not 
benefit whatsoever, depending on your OS, drive, and network connection.

#9: I would need more information on the nature of the timer related problems.  
It would be great if you could open a separate bug for this, with as much 
information as you can share.

Comment 2

[HJ]

Thanks Gordon, 

# 5 - Ok, but I still see images in the disk cache, is that normal?
# 6 - would be nice, when??
# 7 - I use an protocol analyzer and compare the captured data with the
information from about:cache. The information is not the same.

# 8 - does that count for win32 and the Linux OS too?

# 9 - if you like, I will open new bug, but what kind of information do you need?

note: I don't get any javascript and style sheet cached. I mean they pop up in
the disk cache but after an reload/refresh Mozilla requests for a new version.
However, this does not occur when I put that same javascript and/or style sheet
in an HTML file!

Comment 3

[gordon]

#5  - Yes, it's normal.

#6 - I'm currently working on it.  This isn't just something we're planning for 
the future.

#7 - What fields have discrepancies?

#8 - Yes, it could.  SCSI vs. IDE drives seem to make a big difference.  The 
different OS's have very different performance characteristics for 
opening, and creating files.  Obviously, a faster network connection reduces the 
benefits of caching.

#9 - I'd like to understand more precisely what you mean by "timer related 
problems".  It would also be useful to have a detailed description of your 
environment (machine specs, OS, network connection, etc.) so we can better 
understand how the code behaves for you.  Thanks.

Comment 4

[HJ]

My system configuration (custom build):
Intel Pentium 4 at 1.50+ GHz (sorry, non-disclosure)
Intel D850GB motherboard
2 GB PC800 RDRAM
2 * 64 MB 4xAGP videoadapters
1 * Adaptec 29160 Ultra160 SCSI adapter with ASC-29160 Drivers/Microsoft Windows
NT4.0 SP6b
this one is about 240% faster than UDMA and 60% faster than Fibre Channel.
2 * Quantum Atlas? 10K III disk drives of 73.4 GB (Window NT 4/5)
2 * 40GB Ultra ATA-100 hard drives connected to the mainboards controller (for
RedHat 7 only)
2 * Nanao FlexScan L771 1600 x 120 LCD's (seems relevant because it takes a lot
more time to fill up the entire screen)

Sat configuration:
Uplink via Chello cable to Luxemburg (active 7/24)
8 Gb downstream

I like to use SCSI because the CPU usage is very low, compared to fast ATA. I'm
using an Cisco router with Cisco IOS Firewall feature set software for protection. 

Oh, a Cisco engineer is just arrived (7/24 support) he will take the GigaSwitch
adapter out of the system, because there might be trouble with that one he just
told me. Well I have to submit, sure hope this engineer is faster then the last one!

Comment 5

[HJ]

1 - I see that Cookie information is attached to all external javascript files
in cache, why is that? Can't JavaScript use something like document.cookie if
that's not done? If not, then it's only unnecessary data transfer.
2 - Other browsers insert an Referer in the header for images/javascript and
style sheets. While Mozilla only sets an Referer for images. Might be related to
# 3.
3 - I can clearly see JavaScripts and linked Style Sheets in disk cache. But it
seems that Mozilla does not validate this cache content with an
If-Modified-Since request, but request a forced reload for those files! Is
support for this kind of content not yet implemented? Caching Style Sheets and
JavaScript may pretty well speed up things.
4 - If I take a look at the flags in about:cache the I see some flags set to
PARTIAL. The flag state Partial for me means an HTTP 206 header. But I'm unable
to locate a single header in my HTTP capture log! Is PARTIAL in about:cache
something different, or should that be the same as in the RFC 2116?
5 - does Mozilla support Meta tags for Cache issues?
6 - oops almost forgot one important thing, both Pragma and Cache-Content for
Javascript and Style Sheets are set explicitly to no-cache, that might cause #
3, right? 

For the first #7, I will let you know, after that Cisco has confirmed that it's
not related to any other stuff I'm using Ok? Otherwise we might run into other
peoples problems/bugs and that should not be!

Comment 6

[gordon]

I'm sorry.  I meant your system information should go in the new bug your were 
going to create for your question #9.

We have open bugs for cache issues related to JS and CSS (for example, see bug 
29370)

Since most of this bug seems to be purely informational, it may be better handled 
through the news groups.  Please create separate bugs for any problems you can 
reproduce, and we'll try to handle your general informational questions in the 
mozilla.netlib newsgroup.  Thanks.

Comment 7

[Darin Fisher]

#1 - HTTP will cache headers without any data to indicate special conditions
such as URL's that should not be cached and static redirects.

Comment 8

[gordon]

Marking this bug as FIXED, because the discussion has petered out.  Further 
discussion can take place in the news groups.