As explained to me, the issue is that retrieving successive 4K blocks
of an object in memory takes non-linear amounts of CPU time - it takes
X cycles to retrieve the first 4K, but 2*X to retrieve the second 4K,
3*X to retrieve the third 4K, etc. etc.
-C
On May 24, 2009, at 12:19 PM, Dror Galron wrote:
Thank you for your answers,
I have not understood your statement of "Squid-2 has 4KB buffers to
store objects, so the larger ones have some issues doing read seeks"
Could you please emphasize on this?
Thank you,
Dror
On Thu, May 21, 2009 at 6:26 AM, Amos Jeffries
<squid3@xxxxxxxxxxxxx> wrote:
Hi,
I am considering implementing Squid as my web cache for Video
streams
(YouTube etc).
I am going to configure Squid over SAN centralized storage.
I am aware of the additional plug-in required to "normalize" YouTube
URL's.
I have few questions:
1) Are there any example installations of Squid as Video oriented
cache
server?
I'm not aware of anything published.
2) If I implement Squid peering (either digest or ICAP), how does
I think you mean: CARP.
ICAP is a filtering or adaptation method.
Squid solves "popular object" problem, when one cache within the
cluster serves the most popular movie. As far as I understand, in
this
case all requests for that movie would be served from one particular
server; this will cause overloading of that server.
The versions of Squid-2 which have the storeurl features for
normalizing
you-tube requests also contain collapsed_forwarding which damps
this type
of overload down a lot. Squid efficiency rises enormously under
this type
of hot-object scenario up to close around 100% on the single
object. Note
this occurs at BOTH levels of the squid mesh, receiving and source
Squids
doing effective multicast for HTTP.
This is one reason CDN people love Squid so much.
3) Are there any limitations / recommendations for maximal storage
size that has many separate physical disks?
No more than one cache_dir per disk. Squid can easily handle up to 63
cache_dir entries and thus disks. Beyond that certain types of RAID
do
actually start to be useful.
4) Are there any limitations regarding maximal cached object size?
Squid-2 has 4KB buffers to store objects, so the larger ones have
some
issues doing read seeks. I forget what the limits were.
Amos
--
Dror Galron
