Re: Squid Caching Software [7:53221]
John Neiberger wrote in message Are any of you using the Squid open source software on your own hardware? If so, are you happy with it? How does it perform in comparison to other caches you've used? I like Squid. ICP multicast seems like a very intelligent way to move content around. Content encapsulation with mod_gzip is nice on the sending side, but more people need to be caching content on the receiving side!!! Cache hierarchies are very nice, but as a content provider (no names here), I can tell you that all the CDN's and cache hierarchies in the world aren't going to solve any real world problems. Cache hierachies are for end-users, not content providers. Direct interconnection and/or smart routing (BGP performance and correct operation of multi-homed networks) has been and also currently is the champion for content providers, and where they should put the most investment in. Private/Public peering is also a better cost optimization (by leaps and bounds!) for heavy content providers than CDN's or cache hierarchies. Another big responsibility for content providers (and a MUST if they want to save huge amounts of time and money) is to provide the ability for their content to be cached by end-users and Tier-2's. The book Web Caching by Duane Wessels is excellent as are the RFC's. However, this is not a networking problem, it's an HTML and coding problem. While Cisco and Akamai don't understand this at all, they seem to be pushing their products to the wrong people, IMO, and this is why their product lines are suffering. DNS content routing mechanisms (e.g. RR DNS, Cisco Boomerang, Cisco DD, Radware Global Triangulation, et al) are proving to not work (because of DNS servers caching TTL's). To put this in easier terms, content routing can be done in a few forms: * DNS mode - done by client's DNS (not direct end-user), DNS server caching avoids adminitratively set TTL's, can be bypassed by using IP or different DNS name (http://yahoo.com instead of www.yahoo.com) * HTTP-Redirect mode - browser problems, bookmarks can bypass, DNS caching can still avoid administratively set TTL's (same problems really) * Edge-Intercept - now this does avoid DNS caching, but requires access to all the end-users and network. Still has problems with multi-homed users or users attempting to use different DNS servers. * BGP + Anycast - only real method that I know to solve global reachability of services across distributed data centers. Done with IP addresses (announcing single multi-homed blocks in more than one place, with /32's reachable for single IP's throughout the internal infrastructure (since you can't announce anything greater than a /24 in the Internet routing table). Anycast addresses are injected into the IGP (could be IBGP, doesn't have to be OSPF/EIGRP/ISIS) and marked with metrics showing distance. See the following paper for more details: http://www.cisco.com/public/cons/isp/essentials/ip-anycast-cmetz-03.pdf Cisco's IOS SLB and the Catalyst 6500 CSM have a feature called Route Health Injection, or RHI, which is capable of injecting a /32 host route and pulling it out of the routing table if the server (or VIP) is not available. The most current Cisco SRND's (e.g. Enterprise Data Center Design) and Cisco Sales Positioning are now recommending the use of RHI for content routing. I believe that investments should now be going into using this technology instead of letting people like Akamai invest in secrets like this for years. Going back to caching, I think the basics can be implemented by the content providers (such as using Reverse Proxy Caching (RPC) to replace use of servers and aid against flash crowd problems, etc). What I feel is more on the Enterprise or ISP/User side is use of cache hierarchies or CDN's. The strongest caching architectures should be found in those places, as they will have the most cost optimization and performance benefits from implementing such a solution. Here's a link to the Cache Now! campaign - http://vancouver-webpages.com/CacheNow/ I am very interested on the AOL cache architecture. They seem to be the only people who have figured it out. If anyone has any details of what they use / what they do, please post or email me privately. -dre Message Posted at: http://www.groupstudy.com/form/read.php?f=7i=53225t=53221 -- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
Re: Squid Caching Software [7:53221]
I guess I should have been specific about our circumstances. We're replacing an older Compaq TaskSmart cache server and we want to get two of something or other, and we're having a hard time making up our minds. Today I got the bright idea that we could simply buy two Sun Netra servers and put Squid on them. Sounds like a good idea to me but I wanted to hear the opinions of other Squid users. Thanks! John dre 9/12/02 3:04:07 PM John Neiberger wrote in message Are any of you using the Squid open source software on your own hardware? If so, are you happy with it? How does it perform in comparison to other caches you've used? I like Squid. ICP multicast seems like a very intelligent way to move content around. Content encapsulation with mod_gzip is nice on the sending side, but more people need to be caching content on the receiving side!!! Cache hierarchies are very nice, but as a content provider (no names here), I can tell you that all the CDN's and cache hierarchies in the world aren't going to solve any real world problems. Cache hierachies are for end-users, not content providers. Direct interconnection and/or smart routing (BGP performance and correct operation of multi-homed networks) has been and also currently is the champion for content providers, and where they should put the most investment in. Private/Public peering is also a better cost optimization (by leaps and bounds!) for heavy content providers than CDN's or cache hierarchies. Another big responsibility for content providers (and a MUST if they want to save huge amounts of time and money) is to provide the ability for their content to be cached by end-users and Tier-2's. The book Web Caching by Duane Wessels is excellent as are the RFC's. However, this is not a networking problem, it's an HTML and coding problem. While Cisco and Akamai don't understand this at all, they seem to be pushing their products to the wrong people, IMO, and this is why their product lines are suffering. DNS content routing mechanisms (e.g. RR DNS, Cisco Boomerang, Cisco DD, Radware Global Triangulation, et al) are proving to not work (because of DNS servers caching TTL's). To put this in easier terms, content routing can be done in a few forms: * DNS mode - done by client's DNS (not direct end-user), DNS server caching avoids adminitratively set TTL's, can be bypassed by using IP or different DNS name (http://yahoo.com instead of www.yahoo.com) * HTTP-Redirect mode - browser problems, bookmarks can bypass, DNS caching can still avoid administratively set TTL's (same problems really) * Edge-Intercept - now this does avoid DNS caching, but requires access to all the end-users and network. Still has problems with multi-homed users or users attempting to use different DNS servers. * BGP + Anycast - only real method that I know to solve global reachability of services across distributed data centers. Done with IP addresses (announcing single multi-homed blocks in more than one place, with /32's reachable for single IP's throughout the internal infrastructure (since you can't announce anything greater than a /24 in the Internet routing table). Anycast addresses are injected into the IGP (could be IBGP, doesn't have to be OSPF/EIGRP/ISIS) and marked with metrics showing distance. See the following paper for more details: http://www.cisco.com/public/cons/isp/essentials/ip-anycast-cmetz-03.pdf Cisco's IOS SLB and the Catalyst 6500 CSM have a feature called Route Health Injection, or RHI, which is capable of injecting a /32 host route and pulling it out of the routing table if the server (or VIP) is not available. The most current Cisco SRND's (e.g. Enterprise Data Center Design) and Cisco Sales Positioning are now recommending the use of RHI for content routing. I believe that investments should now be going into using this technology instead of letting people like Akamai invest in secrets like this for years. Going back to caching, I think the basics can be implemented by the content providers (such as using Reverse Proxy Caching (RPC) to replace use of servers and aid against flash crowd problems, etc). What I feel is more on the Enterprise or ISP/User side is use of cache hierarchies or CDN's. The strongest caching architectures should be found in those places, as they will have the most cost optimization and performance benefits from implementing such a solution. Here's a link to the Cache Now! campaign - http://vancouver-webpages.com/CacheNow/ I am very interested on the AOL cache architecture. They seem to be the only people who have figured it out. If anyone has any details of what they use / what they do, please post or email me privately. -dre Message Posted at: http://www.groupstudy.com/form/read.php?f=7i=53226t=53221 -- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and
Re: Squid Caching Software [7:53221]
As dre said squid works great. Has worked great for a long time and I don't see any reason for it to stop being good. The available tools for log analysis are broad and pretty good, better than those of the commercial vendors I've seen. The tools for content filtering on squid(though I don't advocate this so I don't try to track) are moderate and the commercial vendors seem to have done better at this. Squid has a much more visible process which makes it easier to support and you won't have to deal with vendors claiming black magic inside their box. The only headaches I've had from a deployed squid cache were from content developers who had no knowledge of caching and who's server was on the far side of a cache. Arguably they shouldn't have been writing content for that big of an audience, but a little education and guidance along with appropriate acl's make that type of problem disappear quickly. BTW, I'm also a fan of anycast when I put it in a few situations back in the mid 90s then saw other folks had been doing it in roughly similar ways, even went to work for one of them briefly. It works well too, only challenge there was finding midlevel support folks willing to wrap their minds around something a little different and in today's market that doesn't seem to be much of a challenge. Good Luck, Darrell always looking for the next big project... darrellhayaitacosnet John Neiberger wrote in message [EMAIL PROTECTED]">news:[EMAIL PROTECTED]... I guess I should have been specific about our circumstances. We're replacing an older Compaq TaskSmart cache server and we want to get two of something or other, and we're having a hard time making up our minds. Today I got the bright idea that we could simply buy two Sun Netra servers and put Squid on them. Sounds like a good idea to me but I wanted to hear the opinions of other Squid users. Thanks! John dre 9/12/02 3:04:07 PM John Neiberger wrote in message Are any of you using the Squid open source software on your own hardware? If so, are you happy with it? How does it perform in comparison to other caches you've used? I like Squid. ICP multicast seems like a very intelligent way to move content around. Content encapsulation with mod_gzip is nice on the sending side, but more people need to be caching content on the receiving side!!! Cache hierarchies are very nice, but as a content provider (no names here), I can tell you that all the CDN's and cache hierarchies in the world aren't going to solve any real world problems. Cache hierachies are for end-users, not content providers. Direct interconnection and/or smart routing (BGP performance and correct operation of multi-homed networks) has been and also currently is the champion for content providers, and where they should put the most investment in. Private/Public peering is also a better cost optimization (by leaps and bounds!) for heavy content providers than CDN's or cache hierarchies. Another big responsibility for content providers (and a MUST if they want to save huge amounts of time and money) is to provide the ability for their content to be cached by end-users and Tier-2's. The book Web Caching by Duane Wessels is excellent as are the RFC's. However, this is not a networking problem, it's an HTML and coding problem. While Cisco and Akamai don't understand this at all, they seem to be pushing their products to the wrong people, IMO, and this is why their product lines are suffering. DNS content routing mechanisms (e.g. RR DNS, Cisco Boomerang, Cisco DD, Radware Global Triangulation, et al) are proving to not work (because of DNS servers caching TTL's). To put this in easier terms, content routing can be done in a few forms: * DNS mode - done by client's DNS (not direct end-user), DNS server caching avoids adminitratively set TTL's, can be bypassed by using IP or different DNS name (http://yahoo.com instead of www.yahoo.com) * HTTP-Redirect mode - browser problems, bookmarks can bypass, DNS caching can still avoid administratively set TTL's (same problems really) * Edge-Intercept - now this does avoid DNS caching, but requires access to all the end-users and network. Still has problems with multi-homed users or users attempting to use different DNS servers. * BGP + Anycast - only real method that I know to solve global reachability of services across distributed data centers. Done with IP addresses (announcing single multi-homed blocks in more than one place, with /32's reachable for single IP's throughout the internal infrastructure (since you can't announce anything greater than a /24 in the Internet routing table). Anycast addresses are injected into the IGP (could be IBGP, doesn't have to be OSPF/EIGRP/ISIS) and marked with metrics showing distance. See the following paper for more details:
