Why did I choose Comcast HSI and SolutionPro as my Internet Service Providers?

Background: After retiring from Microsoft the natural tendency was to continue using MSN to which I had been a charter subscriber. I had one of those limited-use, free-for-a-year accounts from AT&T WorldNet (since they are my long distance carrier).  During some later years that MSN didn't support Windows NT, I switched to MCI and then to GTE IS but when GTE restricted to 100 hours/month I switched back to unlimited MSN (from Costco at $36/3 months).

In November 2000 I got cable modem (leased RCA DCM215; later I switched to purchased Toshiba PCX1100) from AT&T@Home and except their disallowing connection of any server, it is great!

Before running the tests below, it is smart to try something like ShieldsUP! and/or Shield Probe to verify the security of your always-on internet connection.  Also consider visiting a site like DSLReports.com and/or speedguide.net to see if your OS settings are "tweaked" (my Windows 2000 seemed okay but I bumped TcpWindowSize from default (17520) to 32120 for faster cable without proxy).  I experimented with whether there is much disadvantage to setting proxy:8080; @Home seems to use Traffic Server.

Since I was one of the first in my neighborhood things might go downhill later, but I reached around T-2 (DS2) downstream speed (upstream was capped at 128Kb/sec).  Performance slows between about 4:45pm and 12:45am weekdays but otherwise Pinging from a FreeBSD 4.x machine to two machines in nearby cities takes only about 16ms (through at least five hops).  Speed tests aren't perfect but I look at the following:  FTP of large files from the ISP shows 415KB/sec (457*), and from fast servers show 471KB/sec (527*) and 399KB/sec (439*) - trying multiple servers at the same time seemed to show at least 568KB/sec (575*).   The highest speed test I saw was Rutgers' which said 9415Kbps down, 127Kbps up (after increasing default pump down).  It seems to report about 15% packet loss, but I can't trust those fastest results.
My nearest Speakeasy.net says 5554kbps down, 122kbps up.  At DSLReports.com, choosing MegaPath  says 5280K (4152*) down, 123Kbps up.  McAfee says 5.05Mbps.  telus.net said 4.996Mbps (1.943*).  Later testing (from my county's library) reached around 4500K down from SFO or LinkLine and around 1800K up (McAfee only said around 3.3M down).
The following don't seem quite able to measure full speed download:  PC Pitstop said 5000Kb/sec (2060*) download and 131Kb/sec upload (later 740K up).
CNET said 5361Kbps (3817*) - I'm still studying it due to the variability; it seems too short to be robust.  MSN used to say up to 8160Kbps 1000KBytes/sec (20400 2500*) but its testing method probably wasn't robust enough (ran <1 second?) - now seems to be like CNET and says 1938Kbps.
Another that was interesting was Sitka/Triumf (rarely ~7317-3997Kbps) but it had weird negative timings.

Note: the second numbers in parenthesis with an asterisk (*) are a result from a time when modem was direct to computer, while the other numbers are through a NetGear router.  Default TcpWindowSize direct hurt DSLReports by ~25%, Speakeasy and McAfee by ~50%; PC Pitstop by >50% (try 64240 or proxy to reach near 5000 down), CNET and MSN maybe by 20% or more (seem too variable; try proxy to reach near 4750).

On December 1'01, @Home went away and attbi.com started.  Cap lowered to 1500/128, confirmed by new speed test results.  In late May'02, the 128 upload cap was enhanced to 256.  Comcast HSI took over early 2003 and raised download cap to 1800; in early 2004 raised download to ~3300; late March 2005 raised to ~4300/384 (I download ~440KB/sec); late summer 2005 raised to 6600+ if you subscribe to cable TV (I download ~790KB/sec).

Since I'd like to experiment with server(s), I had Verizon hook up (R?)ADSL in November 2000, specifying my ISP to be Lightrealm (later Hostpro), now SolutionPro, with features like no transfer limit, static IP (and server allowed).  At first they wrongly connected me to Verizon.net - the speed tested below 256/64, with ping times maybe 35 or 73ms through a variable number of hops.  Since that was less than half the desired 768/128Kbps Bronze+, they must have configured it to Bronze.  They eventually corrected my ISP, but it took a couple more calls to remove the Bronze speed limit - there was variability so perhaps there was still 3600 feet of excess bridge taps on my line but speed tested around 700/118 (FTP downloads about 85KB/sec).  Ping times were usually around 25ms through at least five hops.  Then in mid-2001 I "upgraded" to Silver 384/384 and tested around 365/365 with similar ping times.  For free DNS I used zoneedit.  Unfortunately I discovered that Lightrealm wasn't a nice Verizon ISP; I was seeing non-broadcast traffic for other IP addresses in my same subnet (SPro said they fixed with newer Cisco router).

So In addition to some security ideas mentioned above, for bridged DSL modems it might make sense to verify that you don't see unexpected traffic (MAC addresses other than the gateway and your modem).  A trace to any non-gateway address similar to yours should go through the gateway (not direct).  Lightrealm apparently had that design flaw, because I saw such issues!

Update: I moved cross-country, into the territory of AT&T (formerly BellSouth) and Bright House Networks.  I don’t watch cable TV and I like a traditional landline for various reasons (such as monitored security alarm), so I chose FastAccess DSL Xtreme 6.0 which seems fine (although I wish it had faster upload than 430kbps).  Ping times start around 8ms and my signal seems fine since I’m around 500m from the DSLAM.  UVerse is available but ping times are supposedly higher and I don’t want their TV service.

Why did I choose pair Networks as my Web Presence Provider (for Web Hosting)?

Since MCI didn't seem to offer any reasonable personal home page hosting, I looked elsewhere. There seem to be a huge number of companies offering free, inexpensive, or expensive hosting! I tried looking into some of the free offerings, but it seemed that they had often started charging, or disappeared, or had small disk space offered, etc (update: I'm looking at GeoCities Free Home Page Information). Some of the higher priced providers offer service that I don't need.

Some providers (especially lower priced) want you to match their template of a home page - they don't allow truly custom pages (via FTP). Some providers (typically lower priced) restrict your site to certain transfer limits per day or month. I didn't want such restrictions - although I didn't anticipate my site being visited much, I didn't want to be penalized if it was (update: my <7Mb site apparently transfers around 300Mb/month with peaks around 16Mb/day). Even after ruling out those types, there were still plenty of choices.

To find my semi-finalists, I ruled out providers who cost over $10/month or seemed to have relatively slow access (in this case fast access supposedly meant T3 also known as DS-3, while slow typically meant T1). I also didn't consider providers outside North America. I could have spent more time researching, but I had at least a half-dozen semi-finalists which I felt was reasonable.

Since I'm accessing the Internet by dial-up, you'd think it might be hard to tell "fast" from "very fast" A trick is to go to someplace like CMU's "Network Development - Traceroute Gateway" (or the one at Novagate) and/or the similar offering from Interconnected Associates (not limited to MCI's backbone but the results when using MCI aren't quite as fast as CMU or Novagate). Enter the address of your potential provider, and look at the bottom line results. For example, the numbers for a fast provider might be around 100 milliseconds but very fast numbers might be closer to 50 milliseconds. Watch out for some of the other Traceroute/Ping services because the bottleneck in checking performance might be on that service's side! Perhaps cooler (but more intrusive) is Visual Ping Tool (helps you discover the bandwidth despite the latency). You can also try to trace connectivity to specific backbones such as Sprint/GSL or UUNet/pubnix. Note that such analysis is just interesting data and I'm not promising that it is perfect (FYI, typically the best performance of the day seems to be around midnight to 5am Pacific time especially near 3am, and best days are Sunday or Saturday, and worst is around 8:30am-5pm weekdays, as show by graphs such as MAE and NAP).

To find my finalist(s), I compared characteristics such as the monthly fee and included disk space, monthly cost of additional disk space, sign-up fee, and any tradeoffs versus those characteristics (such as speed, etc). For example if there were two providers offering $5/month for 1Mb but one offered additional disk space at $2/Mb rather than $1/Mb, they'd move down in my list. If two providers had the same offering for $5/month but one had no sign-up fee, they'd move up in my list. When a provider that offered 10Mb of space for the same price as only 1Mb from others but performance seemed too slow, they moved down on my list. Bottom line was that DigitalDaze offered 1Mb of space on a very fast server whose ISP has a redundant connection from UUNet (and probably also from Sprint) for $5/month (billed quarterly to a major credit card) plus $1/month for each additional Mb, and no sign-up fee (with no transfer limit and allowing full customization via FTP). Of course that was just my finding in mid-October '96 and I'm sure it is subject to change. You can even look at regular testing of their connectivity using Digital Daze(tm) Internet Health Report.

Later I switched to an ISP which included personal web space and dropped DigitalDaze. Since this meant a URL address change, I try to publish the redirectable URL http://www.niceties.com (NetNames USA reserves a domain name and optionally redirects URL/email).

Update: In November '98 I got tired of URL "redirection" and might change ISPs someday, so I researched again with the added requirement of "virtual" domain name, POP email, and web logs.  I am very happy with the "FTP account" I selected at pair Networks; the only drawback seems to be the 3GB/month transfer limit, which I can accept since that is around 10x more than my site gets.


Back to Doug Hogarth's Home Page