Re: Cache Replacement Algorihtms

From: Eran Balter <baltere@dont-contact.us>
Date: Tue, 8 Aug 2000 12:19:31 +0200

----- Original Message -----
From: Duane Wessels <wessels@packet-pushers.com>
To: Eran Balter <baltere@inter.net.il>
Cc: <squid-users@ircache.net>
Sent: Monday, August 07, 2000 7:14 AM
Subject: Re: Cache Replacement Algorihtms

Hi Duane,

Lot of thanks for the reply. Yet, I have some questions I could not resolve.

> > 2. Is there a field that measure the transfer time, by
size(document)/time(fetch of document)?
>
> The transfer time is currently logged, but not stored.
> Look for this line in client_side.c:
>
> http->al.cache.msec = tvSubMsec(http->start, current_time);

Does this tme is precise? meaning: is it the time measured exactly from the
beggining of the transfer till it ends, without connection start times, time
till the first byte arrives, ...

> > 3. Is there a data structure that saves those measures for each server?
>
> There is the "network measurement database" (netdb.c). It measures RTT
for
> servers in groups of /24 subnets.

1. Do you mean that the network range is based on IP addresses with shared
24 bit (if I guess correctly, there would be a hash table that stores linked
lists of netdbEntries, that each of these entries is a network of something
about 2^8 IP addresses)?

2. What is the policy of refreshing the RTT times in netdb (netdbSendPing, I
guess)? Does it happen when a request for a URL document to a server is
submitted or periodically? If periodically, how do we remember to make this
refreshment (meaning: I guess it's done without passing through the whole
data structure searching for expired information) ?

3. Is there a function that computes the transformation from a URL address
to an IP address or URL address to server's address (I guess that would be
in urn.c, but don't know exactly)?

Regards,
Eran B.
Received on Tue Aug 08 2000 - 03:26:14 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:54:46 MST