What kind of hardware do ISPs run? What's in those 3U rack cases?

[spidey07]

Bozz, Nothinman.

Very true. But again I don't really care about storage, that's what the san is for.

Do IDE drives fail more often than a strong SCSI array? How much per year in failed IDE drives does it cost in man hours?

Sorry guys, my perspective may be unique. I come from a world where 4 hours of down time is murder and my guys get fired when there is. The year is 2002, there is no excuse for downtime anymore. The minimal capital outlay for hardware is no longer an excuse. Cost isn't an issue.

SO, why not take a regular PC and load linux, solaris, win2K on it? Its still a server right? We've saved some capital bucks right?
But then again capital doesn't hit the G&A quite as hard as people hours does it? I'd rather fork over the capital and save on expense in terms of manpower.

I only offer my opinion based on "been there, done that, don't want to do it again."

 

[Sunner]

Popular OS's are all over the board. I would have to go with flavers of *nix being the most popular ( unix and other like OS's )

irwincur
I don't work for an ISP, so Im not 100% sure, but Im willing to bet that *NIX's of various kinds are far more common among ISP's than any Win32 platform.

And IMO for a good reason, but since this thread is about hardware, let's skip the OS evangelism

 

[Bozz]

Spidey - Your right about downtime, it is, in fact, most often cheaper to buy more expensive hardware rather than put up with downtime, it all boils down to the total cost of ownership. All the site's I administer use hardware RAID 5 (Adaptec 2100S) for the SCSI arrays and software RAID 1 for the IDE backup drives.

You cannot be too careful I say! Make backups of the backups to be safe.

 

[Sunner]

Originally posted by: spidey07
Bozz, Nothinman.Very true. But again I don't really care about storage, that's what the san is for.Do IDE drives fail more often than a strong SCSI array? How much per year in failed IDE drives does it cost in man hours?Sorry guys, my perspective may be unique. I come from a world where 4 hours of down time is murder and my guys get fired when there is. The year is 2002, there is no excuse for downtime anymore. The minimal capital outlay for hardware is no longer an excuse. Cost isn't an issue.SO, why not take a regular PC and load linux, solaris, win2K on it? Its still a server right? We've saved some capital bucks right?But then again capital doesn't hit the G&A quite as hard as people hours does it? I'd rather fork over the capital and save on expense in terms of manpower.I only offer my opinion based on "been there, done that, don't want to do it again."

Spidey, I agree for a mission critical environment where downtime is unacceptable, say a bank for instance.
But like Nothinman said, in some cases IDE drives, or even a home built server might make sense.

And sometimes the powers that be(CFO, CEO, etc) will limit your budget to the extent that you simply can't afford what you want or should get.
That's happened to me lots of times, and I simply argue a bit to see if I can change their minds, if I can't Ill simply say something to the effect of "Ok, Ill do what I can with this budget, but it won't be optimal, and don't expect miracles".

Every case differs, and not every company has loads of money to spend on SAN's, Sun enterprise boxes, Cisco's top end equipment, etc.

 

[spidey07]

If you're just serving up web services ala web, ftp, dns, cache then we could come up with a low cost solution.

Home built PCs, dozens of them all hanging off of two redundant L4-L7 switches. Keep a couple spares built and ready with the exact hardware config as the others. "server" fails? simply yank the disk and install the spare.

In this respect we let the network and load balancers handle the redundancy aspect and reduce a server to "just proc, memory, disk"

 

[irwincur]

I don't work for an ISP, so Im not 100% sure, but Im willing to bet that *NIX's of various kinds are far more common among ISP's than any Win32 platform.

UNIX and LINUX account for less than 10% of the market.

 

[Armitage]

Originally posted by: irwincur

I don't work for an ISP, so Im not 100% sure, but Im willing to bet that *NIX's of various kinds are far more common among ISP's than any Win32 platform.

UNIX and LINUX account for less than 10% of the market.

Reference?
Or is this just a PIOOMA thing?

 

[spidey07]

Just so more misinformation is not spread...

The ISPs and service provider industries are DOMINATED by *nix flavors. Enterprise and large applications/DBs are DOMINATED by unix. Any serious computing is performed on unix, windows handles all the little stuff. There's just a lot more little stuff than large scale.

 

[Sunner]

UNIX and LINUX account for less than 10% of the market.

Now this is something I would definately be interested in seeing a source for.

I work at a small company that's in the financia business, and we're almost 100% UNIX, as are most companies in this sector.
And I know every ISP I've used personally is all UNIX as well.

 

[RemyCanad]

Any large ISP I have ever visited uses UNIX. But many of the little "Home Town" IPSs use Windows NT or 2K. Why might you ask? From what I can tell from talking to the guys who started the 3 that I know of they decided that they wanted ot start an ISP becuase there town didn't have one or they had to dial long distance.... So they hire someone to come in a build them one. Sadly someone comes in builds a custom server (Really bad in this case) and puts windows on it becuase the person that is getting the ISP build knows nothing of UNIX or even what it is. And on top of this they usually decide they can handle the technical stuff them selves (based off of what the person build it said "If the server locks up just restart it") . So in one instance the owner was using the server every morning to do his stock trading! And he usually crashed it every morning.

 

[Noodles71]

There are 2 common denominators when it comes to what real ISP's will run.

1. Sun Hardware & Solaris Unix.
2. Cisco Networks.

When hardware crashes on these systems, replacements are dispatched from a nearby parts store by an internationally known courier and it arrives in about 1 to 1 and a half hours. If an engineer is required they are onsite at the 2 hour mark, normally about 20 minutes after any parts they need arrive. You pay for Sun and Cisco because they are predicatable and when it does fail you have the hardware and OS expertise on hand right away with no other vendors for engineers to point any fingers at.

If you don't have a need to be online 100% of the time then you can play with the pretend vendors that mix and match hardware and have a support system in place where repair turnarounds are measured in days and not hours.

If you want to be cheap you might be able to get away with it on web servers but if you have any sort of database and you are serious about being online 100% then you will have to spend some serious dollars. Most online trading firms have a 'A' and a 'B' side to their systems. Not only have they paid for a rack of load balanced Cisco and Sun gear but they then duplicate it in the cabinet next door. If for some reason the load balancing within the half dozen or so servers in that cab fails then they switch over to the second rack. Redundant PSU's are present on anything that isn't a small Cisco device or web appliance and are run on different phases fed from different PDUs (Power Delivery Units - very big UPS's). NIC cards are doubled up and run to separate blades on the switches connected to similar but different VLANs (Virtual LANs), sometimes even separate switches as well. Some run 3 different brands of hardware firewalls before you then get to the software ones. Some even then have an identical test network of all of the above. And then there are some companies that have all that duplicated across many sites globally. It all comes down to how much money (and respect) your business will lose if your system goes down.

Without wading too much into the OS debate I have one thing to say "predictability=uptime". If Windows is completely happy with the hardware you have it installed on then you are not going to get any nasty suprises. You are just going to have to pay for the level of hardware development you need to make such outages predictable.


take care,
Greg

 

[Buddha Bart]

For an ISP;

Invest in a really good piece of SAN hardware,

then use something like heartbeat or Windows Clustering and buy 2 of every machine. Get cheap, simple 1U (or blade if you're feeling bleeding edge) machines. They don't need redundant powersupplies, there's a redundant machine. They don't need redundant storage, there's a SAN. Something like an IBM x300 or a Dell PE350.

If you're starting out small you can honestly run everything (webserver, access server, shell server, mail, dns) on one pair of machines. Then grow out pieces as you need them. This way you can afford the longer (probably even up to a full 24 hours) responce time from your hardware vendor, and your sleection of OS/Software doesn't have to be nutzoid in favor of stability.

bart

 

[capybara]

as for what OS most webservers use, seenetcraft
and it appears to me 90% of webservers are unix compliant (FreeBSD, OpenBSD,
Solaris, Linux, HP-UX etc).
>>>>>irwincur, you have alot of knowledge to share, pls try to do so without
insulting everyone. grrrrr!

 

[capybara]

Originally posted by: irwincur
UNIX and LINUX account for less than 10% of the market.

that does not appear to be factual:
using solaris: visa, mastercard, bank of america, equifax,att, aol, bbc, netsol
using macOSX:heatware
using linux: google, amazon, slashdot, whitehouse.gov
using freeBSD: yahoo, fico
using microsoft: Anandtech, ebay
total of above using unix-compliant OS: 15 of 17 sites = 88%
source for this info:

netcraft

 

[drag]

according to this article (although it is a bit dated) says that unix share is a 14 percent.

This is not bad though since Linux is used in 20% of servers... posix compliant servers make up 34% they are a getting close to microsoft's 41 percent.
combined that with novell then the majority of servers are non-microsoft.

I can't seem to find any stats after 2001. It seems that Microsoft has paid off a stats company (garnter) to publish contrary server figures to inflate the percent of Microsoft server share growth.

If someone is using freebsd and bought 2 copies and installed it on 10 servers and another company bought microsoft and installed it on 8 servers, the resualting sale stats would show that Microsoft owns 80 percent of the server market because the second company would of had to by 8 licsenses for each of it's 8 servers... This combined with the fact of contrary figures makes it nearly impossible to say for certain which OS owns what percent of the server market.

OK with that being said...


The only logical answer to the question of what hardware is running at various locals is:

How much money do they have to spend: Small budgets to large...

off the shelf home PCS.... custom PC machines with a hodge-podge of hardware to suit specific purposes.... X86 servers from brandname companies and farming out database managment to more capable companies... a mixture of of various servers and hardware with serveral levels of redundancy with racks and racks of stuff.... Large high end beowolf-type clusters of 50 or so X86 servers and/or large mainframes and exotic stuff like that.

like the sign in the speed shop (car stuff) says: "Speed costs money, how fast do you want to go?"

I think if you want a REAL server the only thing to go with is a new Cray SV1 with a couple of thousand processors (or a nice scalable cray hpc linux cluster), also hire a bunch of programmers to create your own OS to run it just the way you need it...(i'll even set it up for you for free, right after I get my new rocket assisted Jeep)

 

[bizmark]

well I guess the OS debate is kind of thrown off by the lack of good data. drag the link you posted just shows data for "servers". So this everything from webservers and database servers to the file server in a 10-person office. Many businesses invested in Novell stuff long before Linux became a viable solution, so it's not that surprising that Netware continues to hold a large part of the market share in the 'server' market. However, when we talk about enterprise-level servers, I'd bet that the numbers are somewhat different. Just a guess, and a fuzzy one at that.

 

[Sunner]

UNIX owns the enterprise at this pont, and that's definately a fact.

Pretty much every sector that requires big iron hardware, the magical 5 9's, or huge clusters use *NIX.

 

[ynotravid]

UNIX isn't just cheap. Among other things it shines when it comes to uptime. You can't blow your nose on a Windows machine without having to reboot.

And as for:

OK it is abvious that none of you really know what you are talking about...
...
But after reading through damn near the entire thread I noticed that most of the responses were off base or simply wrong. On most forums the highly technical section is for remotely educated answers on difficult questions...

...you may actually learn something. I only responded because I saw quite a few "myths" getting in the way...

UNIX and LINUX account for less than 10% of the market

There is a whole world outside of your office irwincur.

 

[Soybomb]

Of the 6-7 isp's in the area I'm familiar with (ranging from fresh starts with 30 customers to bigger ones with 30k dialup customers) only one runs *nix. I'm a *nix guy, but we don't run it here, we're windows and I'll admit it does the job acceptably well. You'd need to turn to AOL or other massive isp's to find the massive redundancy some of you guys are talking about I think. Maybe its just our low income part of the country, but ISP's just don't make alot of money with "normal" rates to be able to afford really expensive equipment bills.

Its a little late but to answer the posters original question...sorta.....I think you'd be better to maybe carry cases and parts or something for rack mounts. If the ISP doesn't purchase it from Compaq or someone, they'll have someone on staff who can build it easily.