Adrian Chadd disse na ultima mensagem: ... > > So as long as you're able to store small objects seperately from large > objects and make sure one doesn't starve IO from the other then you'll > be able to both enjoy your cake and eat it too. :P > that is really *the* issue I guess coss certainly is the first step in this direction when you seperate by object size you can even tune the file system and OS exactly for this kind of file size which can give you extreme performance boost since you can manage cache_dir very well by minimum_object_size and maximum_object_size (unfortunatly there is no minimum_object_size_in_memory ...) this is an easy approach often it seems difficult to do it on one machine and on a small network may not be a budget for having 2 or 3 cache server, or the benefit does no justify it Now, interesting issue that I can tune the partition for larger files since COSS is a large file and so i could use diskd together with it I use today one squid instance with null fs cache_dir, storing only small objects up to 300k in memory and two more instances storing mid size objects on disk from 300k up, I have another server with very large disks only to store objects >50Mb as parent proxy-only. most people alert me about memory stuff and so but I do not care, hardware is easy and cheap even if expensive because at the end it pays off because I get a constant 30-40% tcp:80 benefit, in peaks very very much more, 200-500% to be exact. I measure the incoming tcp:80 on the outside of my transparent proxy router and measure the outgoing tcp:80 on the inside. Means, supposing 2MB tcp:80 incoming, I get 2.6 - 2.8Mb - that is money. May be for some countries this is not important but here we pay about US1200-1500 per 2Mb so each byte I get out is important, no matter how Michel ... **************************************************** Datacenter Matik http://datacenter.matik.com.br E-Mail e Data Hosting Service para Profissionais. ****************************************************
