Computer: Relatively modern laptop (about 3-4 years), Windows XP, browser is IE (version unknown). No specific web or ppp accelerators. Laptop is overloaded with software, running low on memory, starting any program takes "a long time". Installing interesting software on her laptop is a no-no, since it's controlled by her office's IT department. But once the browser is running, loading web pages is limited by internet speed only. Browsing the web works perfectly fine, and when the going is good, she can listen to internet radio, watch sports events video over the net, copy big files to/from her office over a VPN connection, and generally get work done. A web-based mail reader is just perfect for this situation.
The problem is our last mile connection. Because we live in an area where the nearest DSL or cable-modem is several miles away, we get our connection through a wireless internet co-op, which is overloaded at several points, in particular the final ingress/egress. To mitigate that, individual customers are capped at 2Mbits/s. Right now, I'm getting ping times to shell.sonic.net of 31ms, and throughput of 230 kByte/s (just measured that with wget of a large, sequential file that will not compress). This is acceptable, and at this point, most things would work reasonably well. But that's in the early afternoon, when few people are at home. In the evening (when everyone is trying to use this network to watch videos), typical throughputs are so variable they are hard to measure; they tend to average 50-100 kBit/s (getting 10-20 kBytes per second is typical). At those times, watching video over the net is hopeless, and internet radio is very iffy. The problem are the ping times: it's not uncommon to see them average 900ms to 1200ms, with measurable packet loss (say 20% or 30%). For a tcp connection of a long sequential file, this is bad, but not a big issue, as modern tcp implementations can handle that and stream files anyhow (so those ping times and packet losses are reflected in the measured bandwidth above). That's why even at those times, you can look at static web pages: http opens one tcp connection, sucks in one page, and the underlying reliable transmission layer handles all the mess you throw at it. The price you pay is in the connection setup/breakdown time.
I would hypothesize that the problem with the new mail is that it doesn't use put/get to prepare http pages at the server, and instead uses Ajax to run tight-loop protocols over the network, transmitting perhaps individual mouse events and key strokes, and returning screen updates at the level of individual characters (just an educated guess). Over a network with pingtimes O(1s), that's not going to be joyful.
My wife has been deliberately trying to use the new web mail (to some extent in order to defuse the situation, and make me less upset). Yesterday evening it took her 45 minutes to send one e-mail.
I'm thinking of an alternate solution: Just use IMAP. Set up a hidden Windows machine for her (for example in a VM on my server at home), and exports its user disk via sharing. Install a sensible mail reader (Thunderbird?) there, and have her use that machine over VNC from her laptop. The only problem is that attachments can't be directly saved from the mail reader to her machine, but would have to be double-bounced through a shared (network-mounted) directory.
Or convince her to completely separate home and work, and use a home machine with a sensible mail mechanism at home. We have a half dozen spare laptops around the house. But for her it is a question of convenience that everything is in one place.
I understand that our situation is quite unusual (not many people use a local internet connection with such performance characteristics), but it does point out that there is a difference in the underlying architecture between new and old web mail, which is reflected in different response to environmental stimuli.
P.S. While typing this (and fixing lunch for my kid), the network speed dropped to 7 KByte/s with ping times to sonic of 380ms and 35% packet loss. But I'm still browsing the web: looking at Google news, reading and writing in a technical discussion forum, getting e-mails.
The problem is our last mile connection. Because we live in an area where the nearest DSL or cable-modem is several miles away, we get our connection through a wireless internet co-op, which is overloaded at several points, in particular the final ingress/egress. To mitigate that, individual customers are capped at 2Mbits/s. Right now, I'm getting ping times to shell.sonic.net of 31ms, and throughput of 230 kByte/s (just measured that with wget of a large, sequential file that will not compress). This is acceptable, and at this point, most things would work reasonably well. But that's in the early afternoon, when few people are at home. In the evening (when everyone is trying to use this network to watch videos), typical throughputs are so variable they are hard to measure; they tend to average 50-100 kBit/s (getting 10-20 kBytes per second is typical). At those times, watching video over the net is hopeless, and internet radio is very iffy. The problem are the ping times: it's not uncommon to see them average 900ms to 1200ms, with measurable packet loss (say 20% or 30%). For a tcp connection of a long sequential file, this is bad, but not a big issue, as modern tcp implementations can handle that and stream files anyhow (so those ping times and packet losses are reflected in the measured bandwidth above). That's why even at those times, you can look at static web pages: http opens one tcp connection, sucks in one page, and the underlying reliable transmission layer handles all the mess you throw at it. The price you pay is in the connection setup/breakdown time.
I would hypothesize that the problem with the new mail is that it doesn't use put/get to prepare http pages at the server, and instead uses Ajax to run tight-loop protocols over the network, transmitting perhaps individual mouse events and key strokes, and returning screen updates at the level of individual characters (just an educated guess). Over a network with pingtimes O(1s), that's not going to be joyful.
My wife has been deliberately trying to use the new web mail (to some extent in order to defuse the situation, and make me less upset). Yesterday evening it took her 45 minutes to send one e-mail.
I'm thinking of an alternate solution: Just use IMAP. Set up a hidden Windows machine for her (for example in a VM on my server at home), and exports its user disk via sharing. Install a sensible mail reader (Thunderbird?) there, and have her use that machine over VNC from her laptop. The only problem is that attachments can't be directly saved from the mail reader to her machine, but would have to be double-bounced through a shared (network-mounted) directory.
Or convince her to completely separate home and work, and use a home machine with a sensible mail mechanism at home. We have a half dozen spare laptops around the house. But for her it is a question of convenience that everything is in one place.
I understand that our situation is quite unusual (not many people use a local internet connection with such performance characteristics), but it does point out that there is a difference in the underlying architecture between new and old web mail, which is reflected in different response to environmental stimuli.
P.S. While typing this (and fixing lunch for my kid), the network speed dropped to 7 KByte/s with ping times to sonic of 380ms and 35% packet loss. But I'm still browsing the web: looking at Google news, reading and writing in a technical discussion forum, getting e-mails.
