Email delivery speed.
How long it takes
a message to arrive.

Five tiles are the delivery curve in five numbers: how long it took half, 80, 95 and 99% of delivered messages to arrive, and how long the slowest one took. It's a quick answer to how fast your messages arrive, without the misleading average. In the example half of 118,403 messages arrived within 5 seconds, 95% within 0:08:59, and the slowest message to orange.fr took 18:59:43. The tiles are all one color: there's no “fast or slow” threshold, you judge by the numbers themselves. This is an exact measurement: the slowest message across all servers is the maximum from their reports. The last percent, 1,184 messages, holds the slowest deliveries; why they waited is shown on the “Where the time goes” screen.

Enlarge

• Half — of 118,403 messages — within 5 seconds.
• 80% — within 43 seconds.
• 95% — within 0:08:59.
• 99% — within 2:48:27.
• Slowest — 18:59:43 — orange.fr, server tx-02, an exact measurement.

The tiles are one color: there's no “fast or slow” threshold, judge by the numbers.

The delivery curve plots time against the share of delivered messages — on a logarithmic scale, from a second to a day. It shows where delivery slows down sharply: in the example the curve shoots up after 99%, and the time grows sevenfold. Under the curve is a comparison with the previous collection, so you notice if things got slower. Usually the slowest are messages that had to wait: a rate limit, backoff mode or retries after a temporary bounce. In the example 95% is 0:08:59 versus 0:08:39 last time, 99% is 2:48:27 versus 2:44:02 — no noticeable change. Which of these causes you have is shown on the “Where the time goes” screen below.

Enlarge

• Line — for all servers — an estimate from intervals.
• 100% point — measured.
• Versus the previous collection — 95% — 0:08:59 versus 0:08:39, 99% — 2:48:27 versus 2:44:02.
• Bend — after 99% — seven times longer.

Usually the slowest are messages that had to wait.

The table is the same delivery curve in numbers, for when you need an exact cutoff: how many messages reached each share and in what time. It's handy to quote in a report or in a conversation with your hosting provider. In the example 50% is 59,202 messages within 0:00:05, 95% is 112,483 messages within 0:08:59, and 100% is all 118,403 messages. For all servers together the shares are estimated and the 100% row is a measurement; for a single server the table is exact, taken from its MTA's report. For a single server the table has two columns: the average time of the messages in that share and its cutoff. Here 90% of messages arrived within 0:02:09, 98% within 1:03:29. If the 95% cutoff grows from one collection to the next, messages have started getting delayed.

Enlarge

• 50% — 59,202 messages within 0:00:05.
• 90% — 0:02:09.
• 95% — 112,483 messages within 0:08:59.
• 98% and 99% — 1:03:29 and 2:48:27.
• 100% — all 118,403 — the slowest 18:59:43.

For a single server the table is exact, taken from its MTA's report.

This screen breaks delivery time down into the intervals the MTA itself counts: how many messages arrived within five seconds, within half a minute, within a minute and longer. It's handier than the curve when you need to say what share of messages was delayed. In the example 61,373 messages (51.8%) arrived within five seconds, 84.7% within a minute, 2.0% took over an hour, and 32 messages took over 12 hours. Small intervals are shown as “<0.1%”, not “0.0%”: these are real messages. Another 29,506 messages arrived within half a minute, another 9,442 within a minute, and 18,082 took longer than a minute. If the “over an hour” share grows, check the retries on the “Where the time goes” screen.

Enlarge

• Within 5 seconds — 61,373 messages — 51.8%.
• Within a minute — 100,321 — 84.7%.
• Over an hour — 2,417 — 2.0%.
• Over 12 hours — 32.

Small intervals — “<0.1%”, not “0.0%”.

This screen breaks a message's journey into parts: waiting in the queue before the first attempt, the trip to the mailbox provider's reply and retries after a temporary bounce. It tells you where a delay comes from — your server or the recipient's server. In the example the queue takes 4 seconds on average, the reply 21 seconds, and retries 1:06:17 for 9,473 messages; 92.0% of messages go out on the first attempt. A switch shows the average, 95% or worst time. The most attempts — 9, for that same message to orange.fr that took 18:59:43. The Hub calculates waiting and retries itself with a quick pass over the accounting files: the MTA's report utility doesn't provide them.

Enlarge

• Queue — 4 seconds before the first attempt.
• To reply — 21 seconds for 108,930 messages on the first attempt.
• Retries — 1:06:17 for 9,473 messages.
• Attempts — first — 92.0%, second — 4.5%, later — 3.5%; the most — 9.

The switch — average, 95% or worst.

The best window is the highest sending speed of the period, by number of messages and by volume. The Hub finds it across every record of the accounting file, not by whole clock hours, so a short burst doesn't get averaged away. The screen shows how many messages a server pushes through at its peak. In the example the best by messages is tx-02 on the evening of Oct 21: 23 in a minute, 566 in an hour; by volume — tx-01 with 15.0 MB in an hour. Server peaks don't add up, so the server is always named. By volume tx-01 also has 10.2 KB/s for a minute. The same best-window search is on the statistics overview — there it's shown as the week's peak load.

Enlarge

• By messages — tx-02 on the evening of Oct 21 — 23 in a minute at 23:29, 75 in five minutes, 566 in an hour.
• By volume — tx-01 — 10.2 KB/s for a minute, 15.0 MB in an hour.

Server peaks don't add up — for all servers together the screen shows the best stretch of a single server and names it.

The table shows recipient domains by average delivery time — who your messages take longest to reach. It helps you tell a general problem from a problem with one mailbox provider. In the example messages to the corporate veltrano-group.com take the longest — 0:11:51 versus 0:05:37 on average across all servers, followed by aol.com, yahoo.com and libero.it. A switch shows the same data by vMTA and by node — with the half, 95% and the share of messages over a minute. A domain with just a couple of messages is marked “few messages” and doesn't rank above domains with real volume. Per domain the MTA only gives the average and worst time; the full domain profile is on the “Domains” tab.

Enlarge

• veltrano-group.com — 0:11:51 across 6,425 messages.
• aol.com — 0:09:22.
• libero.it — 0:08:14.
• yahoo.com — 0:08:37 across 11,701 messages.
• All servers — 0:05:37 on average.

The “by vMTA” and “by node” views — with the switch.