Re: [SLUG] Need a lesson in routing [WAS: memory]
On Thu, Feb 26, 2009 at 06:56:57AM +1100, Kyle wrote: > Only just saw this msg. > > I did indeed load up wireshark and was using it. But I couldn't see any > packet "identifying" itself as an ipv6 packet. I realise that is > illogical. Every IP address registered by Wireshark was definitely an > ipv4 format. there might not have been an ipv6 request unless yuo had ipv6 routing setup. your machine would have made a dns request for a > > I s'pose I could switch it back on and go again, but at this point it > works. Once I read up on v6, I'll give it another shot. > > Thanks again folks. > > Kind Regards > > Kyle > > > Martin Visser wrote: >> Well done Michael! >> >> If you were using a wireshark to look at you application flow you would >> have seen the IPv6 query going out (but not getting an answer) and then >> a followup with a working IPv4 query and then a quick response. >> >> >> > -- > SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ > Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html > -- Falling rock. signature.asc Description: Digital signature -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
Only just saw this msg. I did indeed load up wireshark and was using it. But I couldn't see any packet "identifying" itself as an ipv6 packet. I realise that is illogical. Every IP address registered by Wireshark was definitely an ipv4 format. I s'pose I could switch it back on and go again, but at this point it works. Once I read up on v6, I'll give it another shot. Thanks again folks. Kind Regards Kyle Martin Visser wrote: Well done Michael! If you were using a wireshark to look at you application flow you would have seen the IPv6 query going out (but not getting an answer) and then a followup with a working IPv4 query and then a quick response. -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
Well done Michael! As I indicated earlier in my post, many issues with slow application performance these days are due to waiting for either unanswered queries or getting wrong answers. The quick check of raw throughput (your 10MB transfer in a *blip* proved that) can help you concentrate on the real problem. If you were using a wireshark to look at you application flow you would have seen the IPv6 query going out (but not getting an answer) and then a followup with a working IPv4 query and then a quick response. Regards, Martin martinvisse...@gmail.com On Sat, Feb 21, 2009 at 4:41 PM, Kyle wrote: > ... OK!!! > > That is indeed what it does Michael, when it doesn't timeout. I had > previously read up on F'Fox and turned on the various turbocharging options, > but hadn't thought of ipv6. > > So I changed "network.dns.disableIPv6" to true on the hosts behind the > switch and Wow! That's a bit more like what I might expect. > > ipv6 has always been a bit of a black box I've tried to avoid as long as > possible. Guess I need to start reading up on it. Or disabling it! > > Allow me here to thank each and every one of you that have put up with my > ignorance to assist in debugging this issue. > > I'm not convinced that's all there is to it just yet. For instance, the > Linux box is still an order of magnitude faster to load a page, > network.dns.disableIPv6 is true by default in T'Bird on the hosts which > still timeout on initial connection and all hosts are only using the linux > box itself as name server. > > But where we are now will go a long way to dispersing aggravation in the > local browsing community. > > Thanks again. > > ipv6 . mumble, groan, must read . pain in th. mumble, groan, > ipv6 > > > Kind Regards > > Kyle > > > > Michael Chesterton wrote: > >> >> >> Does it sit there for 11 seconds, then load all of a sudden, or does it >> start loading right from >> the start? >> >> I'm wondering if firefox is doing IPv6 lookups and failing. If you want to >> test, disable IPv6 in firefox (about:config) or use the same nameservers as >> the linux router >> >> -- > SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ > Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html > -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
> "Kyle" == Kyle writes: Kyle> If I switch off ipv6, will that help? If you don't have a complete ipv6 stack, turn it off. Most machines will try to talk with an ipv6 server if it's advertised --- and if your remote host doesn't have full connectivity, it'll attempt the connexion, time out, then do the real one. Does wireshark show this? -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
On Sat, Feb 21, 2009 at 2:22 PM, Martin Visser wrote: > Kyle, a few things. > > Firstly you talk about "15Kbps". In my mind this reads as 15 thousand bits > per second. This is slower than dialup speeds. (A little "b" is always bits > *not* bytes, which is "B" in communication speek). Even if you meant 15 000 > bytes per second (which equate to 150 000 is slow). So I am not sure what > you really mean here. > I discussed this with Kyle off-list, and we realised that he's missing some zeros - his dsl modem actually reports 15,000Kbps The rest of Martin's points stand :) -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
Perhaps you can use a socks proxy through a SSH tunnel and ask the Linux box to do DNS for you. This means the only thing that is affecting your speed behind the firewall is the actual connection between the host(running FF) and the firewall. e.g. on the Mac/Windows, ssh -D 1234 u...@linux_box and configure FF to use 'localhost' as the SOCKS proxy, port is 1234. Change 'network.proxy.socks_remote_dns' in FF to true. I wonder if this would result in any significant speed boost. Likewise, make a tunnel to your IMAP server, e.g. ssh -L 993:localhost: 993 u...@linux_box and configure TB to use 'localhost' as the IMAP server. If neither is speeding things up, it seems that might you have a slow connection between your Linux and the host you are running FF. Chris On 21/02/2009, at 4:41 PM, Kyle wrote: ... OK!!! That is indeed what it does Michael, when it doesn't timeout. I had previously read up on F'Fox and turned on the various turbocharging options, but hadn't thought of ipv6. So I changed "network.dns.disableIPv6" to true on the hosts behind the switch and Wow! That's a bit more like what I might expect. ipv6 has always been a bit of a black box I've tried to avoid as long as possible. Guess I need to start reading up on it. Or disabling it! Allow me here to thank each and every one of you that have put up with my ignorance to assist in debugging this issue. I'm not convinced that's all there is to it just yet. For instance, the Linux box is still an order of magnitude faster to load a page, network.dns.disableIPv6 is true by default in T'Bird on the hosts which still timeout on initial connection and all hosts are only using the linux box itself as name server. But where we are now will go a long way to dispersing aggravation in the local browsing community. Thanks again. ipv6 . mumble, groan, must read . pain in th. mumble, groan, ipv6 Kind Regards Kyle Michael Chesterton wrote: Does it sit there for 11 seconds, then load all of a sudden, or does it start loading right from the start? I'm wondering if firefox is doing IPv6 lookups and failing. If you want to test, disable IPv6 in firefox (about:config) or use the same nameservers as the linux router -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
... OK!!! That is indeed what it does Michael, when it doesn't timeout. I had previously read up on F'Fox and turned on the various turbocharging options, but hadn't thought of ipv6. So I changed "network.dns.disableIPv6" to true on the hosts behind the switch and Wow! That's a bit more like what I might expect. ipv6 has always been a bit of a black box I've tried to avoid as long as possible. Guess I need to start reading up on it. Or disabling it! Allow me here to thank each and every one of you that have put up with my ignorance to assist in debugging this issue. I'm not convinced that's all there is to it just yet. For instance, the Linux box is still an order of magnitude faster to load a page, network.dns.disableIPv6 is true by default in T'Bird on the hosts which still timeout on initial connection and all hosts are only using the linux box itself as name server. But where we are now will go a long way to dispersing aggravation in the local browsing community. Thanks again. ipv6 . mumble, groan, must read . pain in th. mumble, groan, ipv6 Kind Regards Kyle Michael Chesterton wrote: Does it sit there for 11 seconds, then load all of a sudden, or does it start loading right from the start? I'm wondering if firefox is doing IPv6 lookups and failing. If you want to test, disable IPv6 in firefox (about:config) or use the same nameservers as the linux router -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
On 21/02/2009, at 3:04 PM, Kyle wrote: Keeping it simple with HTTP (using Firefox), a site like smh.com.au (where I visit daily, so if there's any local caching going on, it's cached and I reckon internode would likely be caching smh.com.au) takes a minimum 11 secs to load and regularly 20+ secs. This is from behind the linux box. However, if I attempt to load smh.com.au from the linux box, it loads in 3secs flat. I don't have squid or any proxying server running myself - at least not that I have personally configured. Does it sit there for 11 seconds, then load all of a sudden, or does it start loading right from the start? I'm wondering if firefox is doing IPv6 lookups and failing. If you want to test, disable IPv6 in firefox (about:config) or use the same nameservers as the linux router -- http://chesterton.id.au/blog/ http://barrang.com.au/ -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
Lots of help coming in here, for which I am eternally grateful. Thank you all. Chris, Yeah, nope. I've scoured the maillog and there's no errors there. IMAP Backend is local file based. The conversation seems to have migrated to the mail server, but its not just that. As mentioned HTTP calls through the box take their time coming up too. Martin, thanks the detailed response. Yes, James Polley pulled me on that earlier too. Sorry. A case of mis-capitalisation (or dropped zeroes). I can never remember which is which there. The modem is reporting a 15559Kbps/1219Kbps Down/Up connection and I'm not more than 1Km from the exchange (So I suspect Netcomm have it wrong too, because I read that as 15 Megabytes per second). As mentioned Noise Margin: ~9dB, Attentuation: ~26dB. Modem connects over PPPoA and I have set MTU to 1492 all the way through the chain (LAN hosts, Linux eth1, eth0, Modem LAN, WAN). I am with internode and the 10Meg test you offered lands literally in a blip. Keeping it simple with HTTP (using Firefox), a site like smh.com.au (where I visit daily, so if there's any local caching going on, it's cached and I reckon internode would likely be caching smh.com.au) takes a minimum 11 secs to load and regularly 20+ secs. This is from behind the linux box. However, if I attempt to load smh.com.au from the linux box, it loads in 3secs flat. I don't have squid or any proxying server running myself - at least not that I have personally configured. Same token; Firefox on linux to load www.telegraaf.nl (a miscellaneous EU website) 13.7secs. Firefox on an OSX MAC (4GB RAM) behind the linux box (with only switch in between) 27+ secs before it got anywhere near 99%. It just seems the Linux box is the bottleneck. Especially when if I _first try to connect either with HTTP _or_ IMAP, I get timeouts. It's like the box takes time to wake up from something. Just to recap. The machine is all-in-one; Postfix/Dovecot/Spamassasin/Amavisd/Clamav mail server. DHCP, DNS server LAN Router Firewall By all accounts memory seems to be working as it should and is not overloaded. CPU Load rarely goes above 30-40% DHCP licences work and zones are updated with no errors DNS calls from cli return almost instantaneously. nslookup some domain in the EU which I happen to know exists and the server instance of 'named' comes back almost instantaneously with a response. I can ping servers I know are located in the EU and get avg. 340ms responses. Relevant iptables rules look like; # IMAP(S) -A chain-IN -p tcp -m tcp --dport 993 -m state --state NEW -j ACCEPT # HTTP(S) -A chain-IN -m tcp -p tcp --dport 80 -m state --state NEW -j ACCEPT -A chain-IN -m tcp -p tcp --dport 443 -m state --state NEW -j ACCEPT The more I read from you good folk, the more inclined I am to believe I have indeed done the right thing with the linux box and it "may" not be the linux routing processes itself. But I've only the one Linksys SD-208 switch between linux and the rest of the network and all reviews I've read about the linksys are good. It's run well for a number of years now. Kind Regards Kyle -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
Kyle, a few things. Firstly you talk about "15Kbps". In my mind this reads as 15 thousand bits per second. This is slower than dialup speeds. (A little "b" is always bits *not* bytes, which is "B" in communication speek). Even if you meant 15 000 bytes per second (which equate to 150 000 is slow). So I am not sure what you really mean here. Secondly as you seem to have different experience with different applications there is some value in splitting up your testing. The first thing I would do is make sure you are good getting good throughput (goodput) up and down. Your ISP probably has a webserver that will network-wise be close to you (not on the big-bad internet). You want to do a download from there. For instance Internode has a number of files on their mirror (which will be unmetered) specifically for this purpose - http://mirror.internode.on.net/pub/test/10meg.test. Your ISP may have something similar ( I know iiNet does) or even other largeish files like windows security updates that available there for easy update. To test upload speed, your ISP might have provided you with limited personal web space. You get one of those large files and then try uploading it. Firefox reports goodput, but you could also use something like wget. If something seems wrong, you can do a packet capture with wireshark you can get an idea of things like retransmissions, fragmenting and the like. Finally, even with good throughput you may have other application issues. For instance if you app needs to do a DNS look or go elsewhere to verify some credentials before the transfer you can have problems. For instance sshd in its default configuration often causes issues for users because it wants to do a reverse DNS lookup on the address of the connecting client. If your primary DNS can't give that answer (because it is a private unregistered address) then it can take some time to traverse multiple DNS servers before eventually giving up. Similar if your traffic is protected by SSL/TLS and the certificate presented has CRL (certificate revocation list) specified and for some reason it can't access the CRL server it could take 15 seconds or more to time out. To determine if such issues exist you can examine logs for the applications, (which often report that such timeouts, or use wireshark again to infer from the request/response sequence as to whether your app is getting the right answers in a timely manner or not. I'm not saying you have either of this issues, but it is important to try and separate out the layers - the lower ones (physical through transport) would be covered by the first tests, and then more detailed log/protocol examination would let you see any application layer issues. Regards, Martin martinvisse...@gmail.com On Sat, Feb 21, 2009 at 10:44 AM, Chris wrote: > Sorry I meant authentication and account information backend. If they are > stored in a remote ldap server and the traffic is slow to that server, in my > experience it can cause clients to get bad responses. Also can you take off > SSL and see if it is faster? > > Perhaps check syslog for errors on the IMAP server. And supply your private > key to wireshark to see the plain traffic. > > > On 21/02/2009, at 9:59 AM, Kyle wrote: > > Not sure I understand you there James. >> >> I telnet-ed in to test Peter's theories below. But for good measure, I >> just tried with openssl as a command too and that responds immediately. >> >> I just don't get it. One host behind the server/router is a MAC on OSX >> with 4GB, another WinXP with 2GB. The WinXP host is by far the worst. But >> irrespective the MAC is not exactly blindingly quick either. (Both wired >> connections) >> >> >> Kind Regards >> >> Kyle >> >> >> >> James Polley wrote: >> >>> you can use openssl s_client in place of telnet to connect - >>> >>> http://www.jaharmi.com/2007/09/26/using_openssl_securely_connect_your_imap_account >>> has a guide. >>> But for good measure Telnetted (and Wiresharked) over both my SSL IMAP port and 25. Both responses come back PDQ. And Wireshark shows traffic moving from one host to the other and return. I'm pretty confident of my iptables setup as I have refined that over a period of years. pe...@chubb.wattle.id.au wrote: > So, connexions to the (imap? smtp?) mail server time out. Can you run > wireshark on the server, and see what's happening? Does the server > have a correct route to the clients? > > If it's smtp, then try telnet from a client to the server (telnet > 192.168.1.1 25) on the inside of the firewall, while watching top on > the firewall. What does the load look like? Does the telnet session > time out? During which part of the connexion? > > -- >> SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ >> Subscription info and FAQs: http://slug.org.au/faq/mailinglists
Re: [SLUG] Need a lesson in routing [WAS: memory]
Sorry I meant authentication and account information backend. If they are stored in a remote ldap server and the traffic is slow to that server, in my experience it can cause clients to get bad responses. Also can you take off SSL and see if it is faster? Perhaps check syslog for errors on the IMAP server. And supply your private key to wireshark to see the plain traffic. On 21/02/2009, at 9:59 AM, Kyle wrote: Not sure I understand you there James. I telnet-ed in to test Peter's theories below. But for good measure, I just tried with openssl as a command too and that responds immediately. I just don't get it. One host behind the server/router is a MAC on OSX with 4GB, another WinXP with 2GB. The WinXP host is by far the worst. But irrespective the MAC is not exactly blindingly quick either. (Both wired connections) --- - Kind Regards Kyle James Polley wrote: you can use openssl s_client in place of telnet to connect - http://www.jaharmi.com/2007/09/26/using_openssl_securely_connect_your_imap_account has a guide. But for good measure Telnetted (and Wiresharked) over both my SSL IMAP port and 25. Both responses come back PDQ. And Wireshark shows traffic moving from one host to the other and return. I'm pretty confident of my iptables setup as I have refined that over a period of years. pe...@chubb.wattle.id.au wrote: So, connexions to the (imap? smtp?) mail server time out. Can you run wireshark on the server, and see what's happening? Does the server have a correct route to the clients? If it's smtp, then try telnet from a client to the server (telnet 192.168.1.1 25) on the inside of the firewall, while watching top on the firewall. What does the load look like? Does the telnet session time out? During which part of the connexion? -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
Just out of curiosity, what is your IMAP backend? e.g. LDAP, etc Chris On 21/02/2009, at 9:59 AM, Kyle wrote: Not sure I understand you there James. I telnet-ed in to test Peter's theories below. But for good measure, I just tried with openssl as a command too and that responds immediately. I just don't get it. One host behind the server/router is a MAC on OSX with 4GB, another WinXP with 2GB. The WinXP host is by far the worst. But irrespective the MAC is not exactly blindingly quick either. (Both wired connections) --- - Kind Regards Kyle James Polley wrote: you can use openssl s_client in place of telnet to connect - http://www.jaharmi.com/2007/09/26/using_openssl_securely_connect_your_imap_account has a guide. But for good measure Telnetted (and Wiresharked) over both my SSL IMAP port and 25. Both responses come back PDQ. And Wireshark shows traffic moving from one host to the other and return. I'm pretty confident of my iptables setup as I have refined that over a period of years. pe...@chubb.wattle.id.au wrote: So, connexions to the (imap? smtp?) mail server time out. Can you run wireshark on the server, and see what's happening? Does the server have a correct route to the clients? If it's smtp, then try telnet from a client to the server (telnet 192.168.1.1 25) on the inside of the firewall, while watching top on the firewall. What does the load look like? Does the telnet session time out? During which part of the connexion? -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
Not sure I understand you there James. I telnet-ed in to test Peter's theories below. But for good measure, I just tried with openssl as a command too and that responds immediately. I just don't get it. One host behind the server/router is a MAC on OSX with 4GB, another WinXP with 2GB. The WinXP host is by far the worst. But irrespective the MAC is not exactly blindingly quick either. (Both wired connections) Kind Regards Kyle James Polley wrote: you can use openssl s_client in place of telnet to connect - http://www.jaharmi.com/2007/09/26/using_openssl_securely_connect_your_imap_account has a guide. But for good measure Telnetted (and Wiresharked) over both my SSL IMAP port and 25. Both responses come back PDQ. And Wireshark shows traffic moving from one host to the other and return. I'm pretty confident of my iptables setup as I have refined that over a period of years. pe...@chubb.wattle.id.au wrote: So, connexions to the (imap? smtp?) mail server time out. Can you run wireshark on the server, and see what's happening? Does the server have a correct route to the clients? If it's smtp, then try telnet from a client to the server (telnet 192.168.1.1 25) on the inside of the firewall, while watching top on the firewall. What does the load look like? Does the telnet session time out? During which part of the connexion? -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
On Sat, Feb 21, 2009 at 8:39 AM, Kyle wrote: > It's all good. Keep the thoughts coming please. > > I actually use IMAP over SSL. you can use openssl s_client in place of telnet to connect - http://www.jaharmi.com/2007/09/26/using_openssl_securely_connect_your_imap_account has a guide. > But for good measure Telnetted (and > Wiresharked) over both my SSL IMAP port and 25. Both responses come back > PDQ. And Wireshark shows traffic moving from one host to the other and > return. I'm pretty confident of my iptables setup as I have refined that > over a period of years. > > Both NIC's in full-duplex (albeit negotiated down to 100Mbps for the switch > behind the router.) > > Is there anything in sysctl.conf I can mess with other than the single; > 'net.ipv4.ip_forward = 1' param which will help? > > If I switch off ipv6, will that help? > > > Kind Regards > > Kyle > > > > pe...@chubb.wattle.id.au wrote: >> >> So, connexions to the (imap? smtp?) mail server time out. Can you run >> wireshark on the server, and see what's happening? Does the server >> have a correct route to the clients? >> >> If it's smtp, then try telnet from a client to the server (telnet >> 192.168.1.1 25) on the inside of the firewall, while watching top on >> the firewall. What does the load look like? Does the telnet session >> time out? During which part of the connexion? >> > -- > SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ > Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html > -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
It's all good. Keep the thoughts coming please. I actually use IMAP over SSL. But for good measure Telnetted (and Wiresharked) over both my SSL IMAP port and 25. Both responses come back PDQ. And Wireshark shows traffic moving from one host to the other and return. I'm pretty confident of my iptables setup as I have refined that over a period of years. Both NIC's in full-duplex (albeit negotiated down to 100Mbps for the switch behind the router.) Is there anything in sysctl.conf I can mess with other than the single; 'net.ipv4.ip_forward = 1' param which will help? If I switch off ipv6, will that help? Kind Regards Kyle pe...@chubb.wattle.id.au wrote: So, connexions to the (imap? smtp?) mail server time out. Can you run wireshark on the server, and see what's happening? Does the server have a correct route to the clients? If it's smtp, then try telnet from a client to the server (telnet 192.168.1.1 25) on the inside of the firewall, while watching top on the firewall. What does the load look like? Does the telnet session time out? During which part of the connexion? -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
> "Kyle" == Kyle writes: Kyle> Must remember to hit "Reply to All" Yes, the mail server *is* Kyle> the box. It also serves DHCP and DNS. But I didn't think they Kyle> were all that heavy. So, connexions to the (imap? smtp?) mail server time out. Can you run wireshark on the server, and see what's happening? Does the server have a correct route to the clients? If it's smtp, then try telnet from a client to the server (telnet 192.168.1.1 25) on the inside of the firewall, while watching top on the firewall. What does the load look like? Does the telnet session time out? During which part of the connexion? It could be your firewall rules are broken, and replies are being dropped or something. Wireshark will tell you. -- Dr Peter Chubb http://www.gelato.unsw.edu.au peterc AT gelato.unsw.edu.au http://www.ertos.nicta.com.au ERTOS within National ICT Australia -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
Hi Alex, No, I doubt it. I have recently checked, tested and setup MTU settings all the way through the chain and that made it a little better, but still not what I would expect from 15Kbps download. Kind Regards Kyle Alex Samad wrote: jumping into the discussion, have you got an mtu problem ? -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
On Fri, Feb 20, 2009 at 12:35:09PM +1100, Peter Chubb wrote: > > "Kyle" == Kyle writes: > > Kyle> So I guess I need to look elsewhere as to why my experience is > Kyle> "slow". To clarify my thinking, my 'slow' experience relates to > Kyle> the Server/Router routing to/from the hosts behind it. > > Is this on individual connexions or after a connection is established? > If the former, I'd be looking at DNS services and timeouts. If the > latter, then measuring on the server box. > > Kyle> Hosts behind the box timeout frequently when contacting the mail > Kyle> server. Likewise HTTP calls through the box seem unusually slow > Kyle> despite an ADSL2+ running at ~ 15Kbps D'Load connection (noise > Kyle> margin and attentuation seem in reasonable levels). Yet an HTTP > Kyle> call from the Server itself loads fairly quickly. jumping into the discussion, have you got an mtu problem ? > > The mail server *is* the box? Is this box also serving DHCP? Who > serves DNS inside your firewall? > == > Dr Peter Chubb http://www.gelato.unsw.edu.au peterc AT gelato.unsw.edu.au > http://www.ertos.nicta.com.au ERTOS within National ICT Australia > A university is a non-profit organisation only in the sense that it > spends everything it gets ... Luca Turin. > -- > SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ > Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html > -- "You believe in the Almighty, and I believe in the Almighty. That's why we'll be great partners." - George W. Bush 12/10/2002 Washington, DC to Turkish Prime Minister Recap Tayyip Erdogan, signature.asc Description: Digital signature -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
Must remember to hit "Reply to All" Yes, the mail server *is* the box. It also serves DHCP and DNS. But I didn't think they were all that heavy. To address Kelvin's reply, DNS calls for internal and external machines are super quick, so I'm guessing the DNS server is also doing its job properly. Kind Regards Kyle Peter Chubb wrote: The mail server *is* the box? Is this box also serving DHCP? Who serves DNS inside your firewall? == -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
> "Kyle" == Kyle writes: Kyle> So I guess I need to look elsewhere as to why my experience is Kyle> "slow". To clarify my thinking, my 'slow' experience relates to Kyle> the Server/Router routing to/from the hosts behind it. Is this on individual connexions or after a connection is established? If the former, I'd be looking at DNS services and timeouts. If the latter, then measuring on the server box. Kyle> Hosts behind the box timeout frequently when contacting the mail Kyle> server. Likewise HTTP calls through the box seem unusually slow Kyle> despite an ADSL2+ running at ~ 15Kbps D'Load connection (noise Kyle> margin and attentuation seem in reasonable levels). Yet an HTTP Kyle> call from the Server itself loads fairly quickly. The mail server *is* the box? Is this box also serving DHCP? Who serves DNS inside your firewall? == Dr Peter Chubb http://www.gelato.unsw.edu.au peterc AT gelato.unsw.edu.au http://www.ertos.nicta.com.au ERTOS within National ICT Australia A university is a non-profit organisation only in the sense that it spends everything it gets ... Luca Turin. -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] Need a lesson in routing [WAS: memory]
maybe you should check your interfaces for half/full duplex and if there's errors or collisions... otherwise have a play with vmstat, iostat, mpstat etc - they could point you in a direction to look further, at least it will give you hints to see if the box is actively swapping (have swapped out data and swapping in/out data all the time are quite different, as James kinda mentioned) On Fri, Feb 20, 2009 at 10:57 AM, Kyle wrote: > Thanks the response and explanation James. > > I get the following, sooo... not _too_ bad I guess from that perspective. > > [k...@bottlenose ~]$ free > total used free sharedbuffers cached > Mem: 20729081987788 85120 0 1710841096132 > -/+ buffers/cache: 7205721352336 > Swap: 41929441124192832 > > So I guess I need to look elsewhere as to why my experience is "slow". To > clarify my thinking, my 'slow' experience relates to the Server/Router > routing to/from the hosts behind it. > > Hosts behind the box timeout frequently when contacting the mail server. > Likewise HTTP calls through the box seem unusually slow despite an ADSL2+ > running at ~ 15Kbps D'Load connection (noise margin and attentuation seem in > reasonable levels). Yet an HTTP call from the Server itself loads fairly > quickly. > > 'route' shows what it needs to show. I have only ever read of one param in > sysctl.conf that relates to routing. Where do I start to look? > > > Kind Regards > > Kyle > > > > James Polley wrote: > >> >> You haven't mentioned swap though - is your machine eating into swap? >> >> The best solution though is to get more RAM. It's cheap, and it makes >> everything faster. >> >> That is, assuming this is actually your problem >> >> On Fri, Feb 20, 2009 at 8:48 AM, Kyle wrote: >> >>> Hi Slugger's >>> >>> It appears I need a lesson in Linux and memory management. >>> >>> If you could treat this request as if coming from a complete numpty >>> please, and simply explain the differences between Cached, Buffered and >>> Application Memory as they pertain to Linux? >>> >>> According to KDE SysGuard, my CentOS 5.2 server appears to "cache" its >>> entire 2GB quotient of physical RAM. And my general experience of the box >>> (implemented as file server, mail server, firewall and router) is that it is >>> slow. >>> >>> Something tells me it shouldn't be behaving like this? >>> -- >>> >>> Kind Regards >>> >>> Kyle >>> >> -- > SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ > Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html > -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
[SLUG] Need a lesson in routing [WAS: memory]
Thanks the response and explanation James. I get the following, sooo... not _too_ bad I guess from that perspective. [k...@bottlenose ~]$ free total used free sharedbuffers cached Mem: 20729081987788 85120 0 1710841096132 -/+ buffers/cache: 7205721352336 Swap: 41929441124192832 So I guess I need to look elsewhere as to why my experience is "slow". To clarify my thinking, my 'slow' experience relates to the Server/Router routing to/from the hosts behind it. Hosts behind the box timeout frequently when contacting the mail server. Likewise HTTP calls through the box seem unusually slow despite an ADSL2+ running at ~ 15Kbps D'Load connection (noise margin and attentuation seem in reasonable levels). Yet an HTTP call from the Server itself loads fairly quickly. 'route' shows what it needs to show. I have only ever read of one param in sysctl.conf that relates to routing. Where do I start to look? Kind Regards Kyle James Polley wrote: You haven't mentioned swap though - is your machine eating into swap? The best solution though is to get more RAM. It's cheap, and it makes everything faster. That is, assuming this is actually your problem On Fri, Feb 20, 2009 at 8:48 AM, Kyle wrote: Hi Slugger's It appears I need a lesson in Linux and memory management. If you could treat this request as if coming from a complete numpty please, and simply explain the differences between Cached, Buffered and Application Memory as they pertain to Linux? According to KDE SysGuard, my CentOS 5.2 server appears to "cache" its entire 2GB quotient of physical RAM. And my general experience of the box (implemented as file server, mail server, firewall and router) is that it is slow. Something tells me it shouldn't be behaving like this? -- Kind Regards Kyle -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html