edoceo: Latin "to inform fully, instruct thoroughly"

../webpage/int-main/20030924-windows-update-bandwidth.phtml


A column by Brian Livingston in eWeek magazine (22 Sep 2003) asked a question about the bandwidth usage of Microsoft Windows updates. We did some really rough research into the topic. These are the unwarranted conclusions.

Disclaimer

The calculations were based on the reported sizes of downloads listed on the Windows Update web site or by what size the download manager reported. These are by no means as accurate as a report one might get from MRTG or extensive iptables logging. The measurements are rounded to the nearest kilobyte. All measurements were made on our lone Windows 2000 Server, no measurements or calculations were made with regard to Windows NT4 Workstation or Server or for Windows 2003 Professional. Reasonable guesses have been made where necessary (see 30% reference below), this limits accuracy but gives context. Use my contact form for disputes or flames. Please draw your own conclusions from your own educated research.

The Downloads

When we checked for this test (24 Sep 2003) our Windows 2000 Server needed six critical updates.

ItemSizeNotes
Service Pack 6 589 KB/15,628 KB First download (589 KB) is only the installer. Our binaries were 15628 KB
IE6 Patch 822925 2.1 MB (~2,150 KB)
Security Update KB824105 232 KB
Security Update KB819696 790 KB This one was not needed after SP4 installation. It is not referenced in any calculation.
Security Update KB824146 917 KB
MDAC Update 823718 1.6 MB (1,638 KB)

The installation procedures first required SP4 installation that removed the requirement for update KB819696. One reboot was required after SP4 and one after the next set of four updates.

These five critical updates consumed (589+15628+2150+232+917+1638) = 21,154 KB of bandwidth. The amount used to request and receive the web pages involved is of course trivial by comparison.

21,154 KB perspective

How much bandwidth is that really? This table should help out. Keep in mind that line speed is measured in bits, the downloads are measured in bytes. Also the line speed is a "best condition" speed; this requires planetary alignment so actual speeds are slower.

Line SpeedTime
56 Kb56 Kb/s = 7 KB/s; 21154 KB / 7 KB = 3022 seconds = ~50 minutes
T1 (1581 Kb)1581 Kb/s = ~197 KB/s = 21154 KB / 197 KB = ~107 seconds = ~1.8 minutes

That doesn't seem so bad does it? We used a Windows 2000 Server as our test, and according to this cNet/news.com report Microsoft Servers are 49% of the market share. Netcraft has a survey that states that there are 4,839,264 servers running Windows IIS, not all Windows 2000 Server. Take 30% of those to be Windows 2000 Server hosts (sound reasonable?) that is ~1,451,779 servers. Those would use (21,154 KB * 1,451,779) 30,710,932,966 KB. Over a T1 that would take (30,710,932,966 KB / 1581 KB/s) = (19,425,005s / 60) = 323,750 minutes, that is ~224 days! Updates were released over 22 days (20 Aug 2003 to 10 Sep 2003).

What about Linux

During that time frame the Linux 2.4.22 kernel was released, the full download was 28,774 KB. With 1,451,779 servers and T1 speeds that is (28,774 KB * 1,451,779) / 1581 KB/s = (26,422,194s / 60) = 440,369 minutes.

Notes

To my knowledge the Linux kernel or other Linux based applications have not had any major virus/trojan/worm attacks like: Slammer, LovSan or SoBig.F that also cause their own bandwidth havoc.

If your server is connected via 10 Mb network to an OC-12 with your hosting provider the download times for anything are trivial (Linux 2.4.22 took lest than 1 minute 10 seconds).

Document Change Log

XHTML 1.1. and CSS2.1/3 © 1999-2008 Edoceo, Inc.
Edit this Page