Date: Fri, 24 May 1996 09:11:21 -0400 From: "Louis A. Mamakos" <louie@TransSys.COM> To: dennis@etinc.com (Dennis) Cc: "Karl Denninger, MCSNet" <karl@mcs.com>, hackers@freebsd.org Subject: The view from here (was Re: ISDN Compression Load on CPU) Message-ID: <199605241311.JAA04673@whizzo.transsys.com> In-Reply-To: Your message of "Thu, 23 May 1996 15:34:55 EDT." <199605231934.PAA29049@etinc.com> References: <199605231934.PAA29049@etinc.com>
next in thread | previous in thread | raw e-mail | index | archive | help
I'll briefly put my UUNET hat on, the one that I've worn when I was directly involved the design, specification and recommendation of hardware selections of well in excess of $10M last year. Consider that you have many, many unmanned POP locations, all over the planet. These are generally co-located in telephone central offices which are not manned 24x7. We colocate in interexchange carrier central offices for a variety of reasons, including cost (you don't need to purchase $5000/mo local loops on each of your DS3 trunks). There is also good environmental conditions and power available in the form of -48V DC power plants that run all of the telco transmission stuff. We do have contracted "remote hands" (NOT "remote brains!") which can be dispatched when required to go touch things, swap boards, etc. This costs about approximately $100 per hour, with something like a 3 hour minimum. And in some cases, they could be a few hours away depending on the time. And in some cases, break other stuff that used to be working 'cause they bumped into it or unplugged the wrong thing. So the way that you want to think about these installations is that the hardware is in a room filled with poison gas, and the last thing that you want to do is send a human in to diddle with stuff. [Did you know that one of the leading causes of telecom related outages is classified as "human intrusion failure"?] Now, PC platforms in this environment: we've searched far and wide for a PC platform which can be deployed in these spaces, not for actual network traffic, but for support purposes such as RADIUS servers, DNS servers, NNTP news senders, etc. Try to find one which does not require a keyboard and display to do system maintenance activities. Try to find one which also operates from -48VDC. Try to find one which is *completely* controllable from a serial port. You can't find one that works anywhere near as well as a Sun Sparc platform, where you can do *anything* you need from the serial port, like change the boot device, format the disk, boot the system, etc. In fact, GNP makes a NEBS complient Sparc platform suitable for just this environment; it's really *very* cool. Cool enough that we are going to incur the considerable extra cost of maintaining a Solarius platform as well as the BSDI platforms we already have. For computing applications which require bulk storage, I have no problem with installing extra disks and boot media to be able to QUICKLY recover the system with NO PHYSICAL INTERVENTION. It's just not reasonable to have to have someone visit a site literally on the other side of the planet in Singapore to punch the reset button. And for access server sorts of products, there is just no good reason why I have to wait for my box to 'fsck' should they crash and reboot. There is no reason why there has to be spinning magnetic media, which will eventually crap out and have to be replaced, in a piece of hardware which is intimately involved in service delivery when there are fine alternatives available which use more reliable technology. Imagine flash memory instead; heck, the Cisco 7500 routers we're deploying have 8MB of flash on the RP, and a pair of 8MB and 20MB PCMCIA flash cards plugged into them as well. > I really hate this garbage argument. Novell servers with uptimes > over a year are commonplace...PC bus, spinning media and all. > you only have problems with drives that are too fast, too hot and > too overworked, which simply isnt the case with a router scenario. Consider this: you as a service provider have some set of costs incurred in providing a service to your customers. - The cost of transmission and telecom services is a big one; it's also somewhat difficult to reduce in a big way since the pricing is somewhat inflexible for your T3 or ISDN PRI. - The cost of the hardware is the next thing that people think about. In either two ways: 1) it's a fixed, one-time expense, compared to the recurring expenses; so there is a CASH issue. 2) the actual COST is relatively low if you consider that you're depreciating the hardware over 60 months (or whatever you convice the IRS of). Compared to the recurring cost of the stuff you plug into the box, the hardware in many cases is almost "free." Assuming you don't have a CASH problem. - The operations cost. That is, how much do you keep spending to keep the box in a state whereby it continues to generate revenue. This includes stuff like maintenance contracts, power, and LABOR. This is the one area where Smart People can directly affect the cost incurred. If you expend more one-time dollars at the outset, you can reduce the recurring expense, and you win in the long run. And don't forget the OPPORTUNITY COST associated with operating this equipment; if you're spending your (limited) time whacking away keeping one box running, you're not doing some other new revenue producing thing, even if you wanted to and otherwise the resources. So when a vendor comes to me with a box which has absurd operating costs because it has to be touched, or takes up too much rack space or the operations component can't be automated, then he's not getting very far in the door. This is the area that an ISP can directly apply his experience and brains to optimize and hopefully cause the REVENUE to exceed the COST at the end of the day, so I can go buy beer. Err.. make the stockholders happy; yeah, that's it :-) So, while optimizing the capital cost of acquiring hardware may be attractive, it's only part of the larger picture you need to consider. [It helps to be able to drive a spreadsheet to be able to tell if the right thing is happening.. That's why the other partition of the disk has Windows95 on it.] The thing that the engineering folks at UUNET are just obsessed with is scalable solutions; if it can't be scaled up, they you'll become a victim of your own success. While I will not claim that my requirements are the same as very many others, you cannot dismiss them at just the ravings of a lunatic. I've put my money (more than $10M of it) where my mouth is. Sorry for the long message; I'll take my UUNET hat off now, and go away quietly. Louis Mamakos
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199605241311.JAA04673>