Sonic.net

<I> people who are using slow (non-Sonic) networks for their transit</I>

That's me. We moved from Sonic's area years ago, but I've kept up the service just for a dependable archive of 20+ years of email, and Sonic's great support -- and to support Sonic's attitude about big national internet issues.

Now I'm in Washington State boondocks at the end of miles and miles of DSL, which sometimes works, always slowly.

dane wrote:

lr wrote:I hope this is the "official" thread for the new webmail. If it isn't, please let me know.

I just did some network usage measurements on the old versus new webmail.
...

RoundCube has more interface elements, which result in more data transmitted initially. But, after first use, these items are cached, so the penalty is not as bad as your testing the first time would indicate.

Wrong. I ran the test three times (each time doing both Squirrelmail and the new Beta Webmail), and reported the third measurement, without shutting the browser or computer down. All this was done in the span of about an hour (during a boring meeting).

If you want to disagree with me on measurements, please feel free to repeat the measurements yourself (or have staff do it) and report your numbers. I would love to enter a technical debate on what the actual numbers are. But that's not the important question. The important question is: did anyone at Sonic think of measuring the performance of the new webmail *before* rolling it out and making the switch mandatory (a decision which has, fortunately, been delayed for now)? If yes, what reasoning was used to go ahead anyhow? And if no, do you (or does your staff) feel that it is responsible project management to roll something like this out without performance regression testing?

If you (or staff) don't feel comfortable discussing this in public, you know my e-mail address.

Ralph,

We tested with what we considered a worst case scenario, dialup using a "56k" modem with an older laptop with IE. here are the results we measured:

With 56k dialup, average of 3 samples each.

Beta webmail:
From login page to Inbox (default of 50 messages)- 14.9 seconds
From login page to Inbox (set to 15 messages)- 13.0 seconds
From login page to Inbox (200 messages)- 43 seconds
Open 3kb message- 5.5 seconds
Open 1.1MB message (picture in body of message which is the default
setting)- 344seconds (5 min 44 seconds)
Return to inbox (default of 50 messages) from 3kb message- 11.1 seconds
Compose message (load Compose message page from Inbox)- 7.5 seconds

Squirrelmail:
From login page to Inbox (default of 15 messages) displayed in- 12.0
seconds
From login page to Inbox (set to 50 messages) displayed in- 17.2
seconds
From login page to Inbox (200 messages) displayed in- 73 seconds
Open 3kb message- 5.6 seconds
Open 1.1MB message (picture in body of message, requires display
attached images with message option)- 156seconds (2 min 36 seconds)
Return to inbox (default of 15 messages) from 3kb message- 8.6 seconds
Compose message (load Compose message page from Inbox)- 5.5 seconds

Hmm, if I add this up (and assume that the time is strictly proportional to bandwidth, which is saying that the 56k modem is the only bottleneck), I get for my scenario: Beta webmail 14.9 + 5.5 + 7.5 = 27.9 seconds, Squirrelmail 12.0 + 5.6 + 5.5 = 23.9 seconds. And that's even wrong, because I'm comparing 50 messages in beta mail with 15 message in squirrel mail (and in my scenario, the message was on the 3rd or 4th screen in squirrelmail). On the other hand, my scenario is not exactly comparable, because I attached a small .pdf file when sending the composed message, but that should only have added 2.5 seconds, and that adder should be constant. Your numbers clearly show that the new mail is only about 15% or so "slower" (measured by transit speed alone). Odd that we disagree so much.

If you assume that the modem actually transports 50 kBits/s (realistic), with my byte counts squirrel mail should have taken 42 seconds (actually, a little bit more because you have to pay a few roundtrips, and ping latencies on modems tend to add 200ms or so). The 24s calculated above is awfully fast. A little bit of that could be explained by packet headers being much shorter on ppp versus ethernet, and compression on the modem might explain the rest. It's at least in the ballpark. But your numbers really don't agree with my packet and byte counts showing a factor of 2-3 between Squirrel and beta, versus your 15%. It's probably not IE versus Safari, as my wife sees a large slowdown using IE too (but hasn't measured that).

I will try to find some time to further figure out what network usage pattern causes such a radical difference in my measurements. I can set up a 56K modem and reproduce your numbers too. But this will not happen soon, as it's pretty busy in the next few days (and long boring meetings in the office are not all that common, fortunately).

Well, at this point I need to apologize. So you had measurements, and they showed that the slowdown from the beta web mail is "minor" (using a sensible definition of "minor"). I do not understand why my measurement shows a much larger slowdown, and some other posters have also seen that qualitatively. Given the information you have, you did the right thing. I'm sorry to wrongly accuse you of being careless or worse. But please consider whether completely turning off all old web mail is a good idea, since there is a set of people and situations where the new one does considerably worse (either in speed, or in "user acceptance").

lr wrote:[...]I do not understand why my measurement shows a much larger slowdown, and some other posters have also seen that qualitatively.

The discrepancy could be anywhere between the network and the machine. Check out modem stats, router rules/settings, use wired networking instead of wireless, CPU/disk load, and application settings, including trying another browser.

Guest wrote:

lr wrote:[...]I do not understand why my measurement shows a much larger slowdown, and some other posters have also seen that qualitatively.

The discrepancy could be anywhere between the network and the machine. Check out modem stats, router rules/settings, use wired networking instead of wireless, CPU/disk load, and application settings, including trying another browser.

My byte/packet count measurements were taken at a place with an "infinitely" fast network (my office very likely has outside connectivity comparable to all of Sonic in Santa Rosa). Concepts such as "modem" and "router" don't really apply there.

But as soon as I find some time for it, I'll try to repeat the measurements with (a) a deliberately slow network (i'll probably set up a modem at home for that), and (b) different browsers. Not going to happen today or Thursday though, real work takes priority.

lr wrote:My byte/packet count measurements were taken at a place with an "infinitely" fast network (my office very likely has outside connectivity comparable to all of Sonic in Santa Rosa). Concepts such as "modem" and "router" don't really apply there.

Unless you work in the networking group and have full access to a company's load balancers, proxy servers, and traffic shaping appliances, it would be difficult to ascertain how they interact with your connectivity.

Thanks for the updated webmail interface. As a Firefox user, I never was able to get anything except the most primitive "works with any browser" interface to function, so this is a major improvement in convenience, especially for composing messages, adding attachments, etc.

I still use Thunderbird when at home, but for remote access this is highly satisfactory.

kgc wrote:Checkboxes on Message Lists - This was a big surprise to me. I didn't realize how much people like using checkboxes and/or are unaware of (or don't realize that they should try) CTRL/SHIFT-Click as standard GUI usage. In response, there will be checkboxes on the message list. I'm not sure if this will come out in the next release or not but will be in place before it is moved to production.

I'm glad to hear the checkboxes will be back. I was getting ready to add that request myself, for the following reason: I use a Wacom Cintiq screen with a stylus, and it's really nice to be able to tap a bunch of checkboxes on the screen to select a bunch of messages to flag, move, or delete. As a tech I'm fully aware of the standard CTRL/SHIFT-click GUI way of doing things, but it's not progress to force users to use two hands, when on-screen checkboxes allow us to do the same work with one hand. A column of checkboxes take up very little screen real-estate, and make multiple selections quicker, simpler, and more intuitive. That said, I do like SHIFT-click for selecting a large range of messages, by selecting the two ends, so having both systems is ideal. Checkboxes for random selections (or CTRL-clicks for people who want to use two hands), and SHIFT-clicks to select a larger group of messages by picking the top and bottom of the group.

Search Results - Back to List

I have been using the search function to look for certain messages. After I find my message and forward it, the window goes back to my original list, rather than to my search results, so I have to do the search again. I believe this is not the case with Squirrelmail. So either the function has changed in the new webmail, or it's not obvious in the new app.