Email Size Analysis

[ The following analysis was done by one of my colleagues at work. We are building an email search and storage subsystem and the sizes and uniqueness of attachments was interesting. The document is looking at the size of the body and attachments (not the headers) of over 18 million corporate emails. Here is his analysis. Thanks Peter. ]

The following is an analysis of a large collection of email messages. Some notes:

• "Bad Messages" are messages the message parser did not understand. (Note that many of the .tar.gz files are corrupt, and therefore there are probably many more messages that were never parsed at all.
• "Clear Size" is sum of the uncompressed sizes of all asset parts. This analysis does not include overhead sizes for storing the tree of asset parts that constitute an asset; it is merely the sum of the parts.
• "Deflated Size" is the sum of the size of each part after compression by the Deflater algorithm.
• "Unique Parts" are, well, unique. That is, among 40 million parts (bodies and attachments), only 21 million are unique and the remaining 19 million are redundant.
• "Unique Clear Size" and "Unique Deflated Size" are the clear text and compressed size totals for just those parts that are unique. Thus "Unique Deflated Size" is the number of bytes we would actually store with SIS and compression turned on. (Of course, you need to multiply by N for replication and add some overhead to each part to get a complete picture.

Item	Value	Percentage
Total Included Assets	18,568,722
Total Bad Assets	36,619	0.2%
Total Asset Parts	40,885,616	100.0%
Total Clear Size	1,424,967,245,112	100.0%
Total Deflated Size	793,987,580,620	100.0%
Unique Parts	21,213,630	51.9%
Unique Clear Size	700,819,336,446	49.2%
Unique Deflated Size	361,929,754,290	45.6%

So this table is a bit strange. Don't think of it as rows. The columns are independent (unfortunately). So in the <16b column, 14,356 clear files were <16 bytes, 108,351 files were deflated to that size, 7,409 unique clear text files were that size, and 4,249 deflated unique files.

Byte Size	Total Clear Bytes	Total Deflated Bytes	Unique Clear Bytes	Unique Deflated Bytes
<1b	10,645,434	10,645,434	2	2
<2b	51,275	0	12	0
<4b	39,115	0	160	0
<8b	10,065	0	1,422	0
<16b	14,356	108,351	7,409	4,249
<32b	41,912	40,519	23,212	21,505
<64b	160,667	186,782	60,399	65,489
<128b	971,882	1,126,347	242,473	350,998
<256b	1,496,287	2,580,006	1,127,850	1,713,909
<512b	2,878,892	5,742,009	1,658,630	3,828,936
<1k	4,519,315	6,743,444	3,134,157	5,328,728
<2k	4,795,537	5,979,510	3,775,815	5,005,490
<4k	4,719,521	3,059,155	3,851,320	2,510,388
<8k	3,506,609	1,360,555	2,931,410	941,587
<16k	2,173,706	767,800	1,752,428	394,225
<32k	1,554,018	623,130	1,040,344	249,027
<64k	1,219,832	578,461	666,137	233,756
<128k	757,401	505,886	348,406	207,676
<256k	535,854	281,989	242,087	120,564
<512k	313,384	250,260	136,979	106,641
<1m	229,601	162,524	100,396	68,163
<2m	143,667	85,135	62,202	35,827
<4m	67,973	39,598	31,307	17,065
<8m	26,946	11,237	12,337	5,483
<16m	6,390	6,516	3,518	3,418
<32m	5,260	786	2,799	388
<64m	575	136	321	88
<128m	110	39	77	22
<256m	24	5	14	4
<512m	6	2	5	2

The following table is estimating the amount of space wasted based on the blocksize of the filesystem.

Block Size Bytes	Total Unused Slack Space	Percent Overhead
512b	5,326,831,438	1.5%
1k	11,470,299,470	3.1%
2k	26,209,376,590	6.8%
4k	62,533,699,918	14.7%
8k	142,984,419,662	28.3%

Free Spam Protection   Android ORM   Simple Java Magic   JMX using HTTP   Great Eggnog Recipe   Massachusetts Covid Vaccine Sites