- From: Eric J. Bowman <eric@bisonsystems.net>
- Date: Mon, 19 Jan 2015 19:25:07 -0700
- To: Chris Palmer <palmer@google.com>
- Cc: Noah Mendelsohn <nrm@arcanedomain.com>, "Michael[tm] Smith" <mike@w3.org>, Mark Nottingham <mnot@mnot.net>, Henri Sivonen <hsivonen@hsivonen.fi>, Tim Berners-Lee <timbl@w3.org>, Public TAG List <www-tag@w3.org>
Chris Palmer wrote: > > Noah Mendelsohn wrote: > > > [1] http://www.w3.org/DesignIssues/NoSnooping.html > > """This takes a lot of server CPU cycles, making server farms more > expensive. It would slow the user's computer. It would effectively > slow down the whole net.""" > > That was not true in 2009, and it's certainly not true now. > Show me any review of Celeron or Sempron server CPUs on tomshardware or anandtech which support your contention. I'm a dinosaur who'd rather purchase old systems vs. new CPUs in the $50-$60 range to handle way more unencrypted Web traffic, but I see TBL's point (then and now) vis- a-vis server farms. Especially when I look at "cloud" hosting rates nowadays -- at these prices I can't cost-justify retaining independence by running my own hardware, assuming ubiquitous HTTPS. User CPUs are now soldered on with integrated GPUs, but I think we can agree that's irrelevant to user-perceived performance nowadays, even back in 2009. Network slowdowns are ulikely, but more expensive server farms is spot-on from my POV. Please don't leave it to me, or TBL, to undertake the research showing how much of the Web is hosted on Celeron and Sempron processors, or shows how badly their performance degrades when handling HTTPS-centric loads. IMNSHO, claiming that even 5+ years ago this was a fallacy, puts the onus on you to back it up with verifiable numbers which discount what I've been reading on tomshardware, anandtech, etc. regarding CPU performance on Web workloads over that timeframe. Your arguments assume various processor enhancements which have yet to filter down, with no guarantee they will anytime soon; after this many years I'm not willing to bank on promises they will at the $50-$60 CPU cost driving the commodity webhosting/cloud industries. I'm also not willing to assume that budget hosting plays on Celeron and Sempron CPUs falls under the 80/20 Mendoza line. What I don't have, is the wherewithal to undertake such research myself. Had it occured to me, I'd certainly have collected an arsenal of bookmarks supporting my contention for the sake of future mailing-list discussions. My first multi-core CPU was what, 2002-ish? But just made it to Celeron last year? This tells me that optimizations for ubiquitous HTTPS are a ways off for budget server CPU purchasers, unless proven otherwise, based on experience. -Eric
Received on Tuesday, 20 January 2015 02:26:02 UTC