OS reliability report

[Steve]

June 06, 2006 8:21PM
UC annual IT reliability report list

The following OS's are used across all campuses in some manner or another. Other OS's are in use by staff and faculty, but not across all campuses. This list ranks each OS from most reliable (stability/lack of downtime, ability to complete assigned processing tasks, ability to be updated, repaired, and/or resecured on the fly). Overhead, customizability, etc. are not included. These rankings are based upon reliability reports and surveys from August 2005 through the beginning of June 2006.

1. Solaris 10
2.HP-UX
3.Solaris 9
4. AIX
5. BSD
6. IRIX
7. DOS 6.22
8. Debian
9. Windows XP
10. Ubuntu
11. Red Hat
12. Windows 2k

Due to incomplete record-keeping, BSD, OpenBSD, NetBSD, and FreeBSD have been lumped into one category.

Also related, see link to article from The Inquirer. http://theinquirer.net/?article=32224

 

[jlbenedict]

Pretty neat info

 

[Sunner]

Erm...I don't see anything remotely useful there.
Too much info lacking about just about everything.

 

[gsellis]

Thanks. But it is incomplete. They missed VM/CMS, MVS and TPF. Big iron could not out the top 3.

 

[pkme2]

Very interesting,

 

[nweaver]

yeah, shifty source, and no supporting data means it's worthless

 

[spyordie007]

If it's on thetnquirer.net I dont read it.

 

[n0cmonkey]

Uptime doesn't mean anything.

 

[RebateMonger]

They don't user Server 2003?

DOS 6.22???

 

[spyordie007]

Originally posted by: RebateMonger
They don't user Server 2003?

DOS 6.22???

Funny considering "ability to complete assigned processing tasks" is one of the criteria

 

[n0cmonkey]

Originally posted by: RebateMonger
They don't user Server 2003?

They should start since it's more reliable than Linux.

 

[spikespiegal]

I'd like to load the 70+ applications I have running on my Windows 2000 based Citrix servers (which often exceeed 5,000+ threads) on the XP, Dos, and the Solaris 10 box, and see how long they last.

Only way to fairly establish a list would be to run all the OS's inside of a VMware shell under the same conditions with the same load.

DOS 6.22? Where's Netware 4.x?

Toughest systems I've ever worked on were IBM's As400s.

 

[n0cmonkey]

Originally posted by: spikespiegal
I'd like to load the 70+ applications I have running on my Windows 2000 based Citrix servers (which often exceeed 5,000+ threads) on the XP, Dos, and the Solaris 10 box, and see how long they last.

Because Solaris is known for its instability and crappy threading. :roll:

Only way to fairly establish a list would be to run all the OS's inside of a VMware shell under the same conditions with the same load.

DOS 6.22? Where's Netware 4.x?

The way it read to me is that these were the results of their real world experience, and not meant to be a conclusive scientific study. Maybe it was just me.

 

[spikespiegal]

Solaris or Unix don't get nearly the garbage can treatment that Windows does, and you don't exactly load solaris with databases you buy at Best Buy, then run it on $300 E-machines.

I'm also curious how many ERP/MRP systems get run on Solaris, and how many drivers for $80 inkjet printers get run on it along with that same ERP/MRP system.

I've had Win2K servers in restaurants run for over a year until the UPS died, but I don't see vendors at my door offering to replace that box with HP-Unix or an IBM i-Series.

 

[Sunner]

Originally posted by: n0cmonkey

Originally posted by: spikespiegal
I'd like to load the 70+ applications I have running on my Windows 2000 based Citrix servers (which often exceeed 5,000+ threads) on the XP, Dos, and the Solaris 10 box, and see how long they last.

Because Solaris is known for its instability and crappy threading. :roll:

We had a program go haywire on an old Solaris 7 box once, a rather crappy one at that(E250 iirc).
Basically, it just spawned assloads of processes, I can't even remember how many it peaked at, but the load average was up to around 130 or so(yes, that 0 is supposed to be there ).
Funny thing is, since it wasn't a critical box, we didn't have anything more than some basic monitoring on it, network connectivity and such, so we didn't really notice since it did what it was supposed to do, just slower than usual, no one really checked until it became so slow that it got annoying.
Needless to say, we killed the offending processes as soon as we discovered this, but looking at the monitoring software, it had been increasing for about 80 days or so.
And yes, I know load average is useless by itself, but suffice to say, the box was quite loaded, took me around 10 minutes to logon, and running ps...let's not talk about that shall we...

 

[Nothinman]

I'd like to load the 70+ applications I have running on my Windows 2000 based Citrix servers (which often exceeed 5,000+ threads) on the XP, Dos, and the Solaris 10 box, and see how long they last.

We've got a handful of 2000 based Citrix servers and as long as they're setup and maintained properly they'll run indefinitely, or at least until you have to reboot for a patch. Our biggest problem was printing, but that's all in the past AFAIK.

Solaris or Unix don't get nearly the garbage can treatment that Windows does, and you don't exactly load solaris with databases you buy at Best Buy, then run it on $300 E-machines.

You buy your databases at BestBuy? No wonder you're having problems.

I'm also curious how many ERP/MRP systems get run on Solaris, and how many drivers for $80 inkjet printers get run on it along with that same ERP/MRP system.

The databases are probably on Solaris pretty frequently, the ERP application itself is (or at least should be if the app is half-respectable) a lot less resource intensive. The ERP system we use isn't exactly what I would describe as a well designed application but we haven't had any major issues running it on workstations or Citrix servers.

And as for your $80 printers, the drivers are the problem not the OS and I'm glad they only work on Windows. The fact that those crap HP SOHO printers require like 8 services and 3 kernel mode drivers for no real reason is no one's fault but HP's.

I've had Win2K servers in restaurants run for over a year until the UPS died, but I don't see vendors at my door offering to replace that box with HP-Unix or an IBM i-Series.

So? I don't think MS has ever come knocking at our door trying to replace our GS160s either, so what's your point?

 

[RebateMonger]

I think that ALL current OS's are pretty reliable. The problems come with drivers, applications, and hardware, which the OS makers often don't control. Proper OS maintenance and patches are also important in keeping small problems small, and keeping hackers away.

 

[n0cmonkey]

Originally posted by: spikespiegal
Solaris or Unix don't get nearly the garbage can treatment that Windows does, and you don't exactly load solaris with databases you buy at Best Buy, then run it on $300 E-machines.

I'm also curious how many ERP/MRP systems get run on Solaris, and how many drivers for $80 inkjet printers get run on it along with that same ERP/MRP system.

I've had Win2K servers in restaurants run for over a year until the UPS died, but I don't see vendors at my door offering to replace that box with HP-Unix or an IBM i-Series.

ERP? MRP?

Solaris is considered by many to be one of the best scaling OSes available. And please notice that I did not say anything bad about Windows, just filling you in in case you were under informed.

 

[Skitzer]

Hmmm ..... this may be a stupid question but where does Mac OS X fit into the equation? Maybe under BSD? Or Unix?

 

[Dravic]

Originally posted by: n0cmonkey

Originally posted by: RebateMonger
They don't user Server 2003?

They should start since it's more reliable than Linux.

If the servers i admin are down 3-5 time a year for a total of 10-20 hours a year.....

houston we have a problem...

 

[n0cmonkey]

Originally posted by: Skitzer
Hmmm ..... this may be a stupid question but where does Mac OS X fit into the equation? Maybe under BSD? Or Unix?

It probably doesn't.

 

[Skitzer]

Originally posted by: n0cmonkey

Originally posted by: Skitzer
Hmmm ..... this may be a stupid question but where does Mac OS X fit into the equation? Maybe under BSD? Or Unix?

It probably doesn't.

I have used both Windows XP and Mac OS X, (Jaguar through Tiger), for a few years now and I can tell you that my Mac is and has always been much more stable. Of course I am only one person.

 

[n0cmonkey]

Originally posted by: Skitzer

Originally posted by: n0cmonkey

Originally posted by: Skitzer
Hmmm ..... this may be a stupid question but where does Mac OS X fit into the equation? Maybe under BSD? Or Unix?

It probably doesn't.

I have used both Windows XP and Mac OS X, (Jaguar through Tiger), for a few years now and I can tell you that my Mac is and has always been much more stable. Of course I am only one person.

It probably doesn't fit in their report, whether it's stable or not (not that I'd know, having only used OS X since v10.0).

 

[drag]

Originally posted by: Dravic

Originally posted by: n0cmonkey

Originally posted by: RebateMonger
They don't user Server 2003?

They should start since it's more reliable than Linux.

If the servers i admin are down 3-5 time a year for a total of 10-20 hours a year.....

houston we have a problem...

If you have 15 hours downtime spread over a year you have a 99.83% aviability. If your dealing with PC-class hardware (including normal everyday Xeon/Opteron servers) that's about what I would expect for that sort of stuff, if you get better reliability it's just through dumb luck. Actually that would be pretty decent.

If you have automatic fail over between 2 machines that theoreticly should net you 99.999 percent uptime. Now of course due to power outages and any sort of emergancy patching you need to do increases the likelihood of them going down together. So it's more like 99.9 to 99.95 % uptime.

I don't know about Windows, but this level of reliability is generally easily acheivable. Just setup 2 machines and use Linux-HA (heartbeat) to have one machine startup a service if the other one fails. The hard part is making sure that the information on both systems is mirrored correctly, and for normal stuff that's easily done with drbd. Machines maintain a mirror'd partition with one machine mounting it at a time. One machine fails, the other machine detects it through the network, runs fsck on the partition, mounts it, and starts the service up. It'll range from about 30 seconds to 10 minutes depending on how clean the file system is.

So with that level of relaibilty you'd be safe to expect about a average of 10 minutes downtime per month. To bad it doesn't get spread out like that.

I think that people tend to heavily over estimate the effect of downtime. I estimate that 19 out of 20 businesses that figure they need 99.9% (about a average of 45 minutes downtime per month.) reliability don't and are just wasting their money. Unless your doing clustering stuff the hardware to get 99.9% or better starts to get very expensive.