[Fwd: Re: Coping precedence from cache->server conenction to cache->client.]

attached mail follows:

For better QoS we want that TOS of connection between server and proxy
will propagate to the client to proxy connection. Here is example:

Let's assume that due to QoS issue packets that come from certain
links are marked with certain precedence on ISP's main router. Later
secondary route to which customer is connected does traff shaping based on
the TOS/precedence of the packet. So without proxy everything works fine,
clients goes to directly to www server and all packets coming from the
server are marked by main router. Here is small diagram:

          __________ ___________ __________
client ->|sub router|---->|main router|---INET-->|www_server|
         | | | | | |
client <-|shaping |<----| |<-------- | |
          ---------- ----------- ----------
                   TOS=0x01 TOS=0x00
As you can see main router sets some TOS lets assume 0x01, and
sub-router(might be same as main router) makes some QoS based on the TOS.
If I put a squid(transparent mode) between sub router and the main router,
precedence will not propagate to sub-router and wrong QoS decision will be
made. I've already patched our squid to set defined TOS value on
squid-client side if request was hit, for example we can shape MISSes and
not to shape HITs(btw do you think this patch worse submitting to the
squid source tree ?). Next step we want to do is to read TOS of packets
that come to squid from main_router, and set same TOS to the client->squid
connection. So if squid received object from the net it will be sent to the
client with same precedence as it was received by squid.

I was thinking to expand store_entry structure to hold TOS value that came
with the object (for now I don't see any needs for it to be saved in the
swap files).

Let me know what do you think about it, and if you have any ideas how to
implement it the best way.

P.S.: We have some plans to fix the MTU discovery problem with transparent
proxy, so if you have to say anything about it other than in your last
Usenet article, we would really like to hear about it.

On Mon, 10 Jan 2000, Henrik Nordstrom wrote:
|Sorry. I do not follow what you are trying to do. Can you please
|reprhase the question? Preferably giving an example.
|
|--
|Henrik Nordstrom
|Squid hacker
|
|
|Leonid Igolnik - LiM wrote:
|>
|> I am considering to implement precedence propagating from cache->www_server
|> connection to the one made with the client. I think to expand store_entry
|> to store the precedence value at the first stage, and maybe storing it with
|> object's meta-data (I know that I will lose all my current cache). Any
|> ideas, comments, suggestions and pointers are welcomed.
|> Leonid Igolnik aka LiM
|

Leonid Igolnik aka LiM
Received on Mon Jan 10 2000 - 18:37:51 MST