Actual Speeds

[yoda291]

Originally posted by: Sunner

Originally posted by: yoda291

Originally posted by: spidey07

Originally posted by: randal

Originally posted by: Sunner
The fastest real world transfer I've seen was between two servers directly connected to each other.
Both are HPaq Proliants with Broadcom bcm5700 NIC's, getting about 40-50 MB/sec, this is sending data from a Win2K box to a Linux box running Samba, no tweaking.
Of course, this measurement is HIGHLY unscientific, seeing as I just looked at the files transferring, somewhat easy though since the files were all 40-50 MB in size and took about 1 sec/file

And yes, PCI is 133 MB/Sec.
PCI-X goes from 66 - 133 MHz, ~ 532 MB/Sec - 1064 MB/Sec.
Then there's PCI-X 2.0, which increases this to a max of 533 MHz, ~4 GB/Sec.

Considering that a standard 1000mbps NIC is PCI@33Mhz/32bits, it becomes nearly impossible for the machine to saturate a gigE network; there simply isn't enough bandwidth on the machine's bus to accomodate moving tons of data from a HD/memory to the NIC and back without everything having to compromise/slow down.

Hence why big iron servers have ridiculous amounts of I/O & bus availability - going so far as to have completely separate busses for different devices.

hence why "real" servers (not the puny wintel crap) have no trouble filling multiple 1000
Base-t ports.

I fail to see how a server running an intel proc or windows as an operating system affects bus speeds. Last time I checked, most servers sold nowadays use an intel proc.

Well, an IBM pSeries certainly has alot more I/O bandwidth than any Wintel server around.
But again, even lowly 2-way $2K servers use PCI-X buses, oftentime several of them, so they shouldn't have much of a problem, so long as you're not talking loads of GigE ports.

actually a unisys es7000 would be around those IO numbers as well. I'd have to bench them and I don't have several hundred thousand dollars to throw around. Funny sidenote,I conference called a unisys sales rep and an IBM sales rep last year when spec'ing out a database machine (they were all giving me conflicting numbers and facts) -- hilarity ensued, and it eventually degraded into playground name-calling tactics. Fun was had by all.

 

[Sunner]

Originally posted by: yoda291

Originally posted by: Sunner
Well, an IBM pSeries certainly has alot more I/O bandwidth than any Wintel server around.
But again, even lowly 2-way $2K servers use PCI-X buses, oftentime several of them, so they shouldn't have much of a problem, so long as you're not talking loads of GigE ports.

actually a unisys es7000 would be around those IO numbers as well. I'd have to bench them and I don't have several hundred thousand dollars to throw around. Funny sidenote,I conference called a unisys sales rep and an IBM sales rep last year when spec'ing out a database machine (they were all giving me conflicting numbers and facts) -- hilarity ensued, and it eventually degraded into playground name-calling tactics. Fun was had by all.

Well, I intentionally disregarded some of the more "oddball" Intel bigiron, I was referring more to the average Intel server.
I don't have any numbers for SGI's Altix boxes, but I'd guess those have a fair amount of I/O bandwidth as well.

And that conference call sounds like fun, gotta do that with IBM and HP some day

 

[spidey07]

Originally posted by: yoda291

Originally posted by: spidey07

Originally posted by: randal

Originally posted by: Sunner
The fastest real world transfer I've seen was between two servers directly connected to each other.
Both are HPaq Proliants with Broadcom bcm5700 NIC's, getting about 40-50 MB/sec, this is sending data from a Win2K box to a Linux box running Samba, no tweaking.
Of course, this measurement is HIGHLY unscientific, seeing as I just looked at the files transferring, somewhat easy though since the files were all 40-50 MB in size and took about 1 sec/file

And yes, PCI is 133 MB/Sec.
PCI-X goes from 66 - 133 MHz, ~ 532 MB/Sec - 1064 MB/Sec.
Then there's PCI-X 2.0, which increases this to a max of 533 MHz, ~4 GB/Sec.

Considering that a standard 1000mbps NIC is PCI@33Mhz/32bits, it becomes nearly impossible for the machine to saturate a gigE network; there simply isn't enough bandwidth on the machine's bus to accomodate moving tons of data from a HD/memory to the NIC and back without everything having to compromise/slow down.

Hence why big iron servers have ridiculous amounts of I/O & bus availability - going so far as to have completely separate busses for different devices.

hence why "real" servers (not the puny wintel crap) have no trouble filling multiple 1000
Base-t ports.

I fail to see how a server running an intel proc or windows as an operating system affects bus speeds. Last time I checked, most servers sold nowadays use an intel proc.

I guess I was talking about servers that don't use intel processors or use PCI buses and aren't affected by their limitations.

 

[ktwebb]

"actually a unisys es7000 would be around those IO numbers as well. I'd have to bench them and I don't have several hundred thousand dollars to throw around"

If I owned ours and not the company, I'd gladly give you a big discount just to get the piece of sh*t out of the server room and under my management watch. I was the available server tech when it was delivered and setup by Unisys so I got to be the ES7000 "expert" Unisys gave me a 20 minute demo on the sentinel server, the partition and the service processor. ES7000 /400 series. 410 I believe. I don't that one is several hundred thousand. Seems like I heard 130K. Might be the components in the package. 8 Itanium proc's, 32 GB or RAM. with moitoring 2k3 server (sentinel) on a Dell 1750 and XP management workstation (service Processor), some off brand, in the solution with the Partition (Itanium box). I have very little good to say about it, including Unisys's support.

 

[spidey07]

Originally posted by: ktwebb
"actually a unisys es7000 would be around those IO numbers as well. I'd have to bench them and I don't have several hundred thousand dollars to throw around"

If I owned ours and not the company, I'd gladly give you a big discount just to get the piece of sh*t out of the server room and under my management watch. I was the available server tech when it was delivered and setup by Unisys so I got to be the ES7000 "expert" Unisys gave me a 20 minute demo on the sentinel server, the partition and the service processor. ES7000 /400 series. 410 I believe. I don't that one is several hundred thousand. Seems like I heard 130K. Might be the components in the package. 8 Itanium proc's, 32 GB or RAM. with moitoring 2k3 server (sentinel) on a Dell 1750 and XP management workstation (service Processor), some off brand, in the solution with the Partition (Itanium box). I have very little good to say about it, including Unisys's support.

that was kinda my point - real servers don't aren't wintel

more to the point - wintel "servers" are nothing more than glorifed PCs.

 

[ktwebb]

My problems with the ES7000 has nothing to do with windows or intel. Nor do I have a problem with our 3 hundred plus Wintel servers. Hey to each his own but "real servers"? Yeah, ok spidey.

 

[spidey07]

just trying to stir the pot KT

 

[yoda291]

Originally posted by: ktwebb
"actually a unisys es7000 would be around those IO numbers as well. I'd have to bench them and I don't have several hundred thousand dollars to throw around"

If I owned ours and not the company, I'd gladly give you a big discount just to get the piece of sh*t out of the server room and under my management watch. I was the available server tech when it was delivered and setup by Unisys so I got to be the ES7000 "expert" Unisys gave me a 20 minute demo on the sentinel server, the partition and the service processor. ES7000 /400 series. 410 I believe. I don't that one is several hundred thousand. Seems like I heard 130K. Might be the components in the package. 8 Itanium proc's, 32 GB or RAM. with moitoring 2k3 server (sentinel) on a Dell 1750 and XP management workstation (service Processor), some off brand, in the solution with the Partition (Itanium box). I have very little good to say about it, including Unisys's support.

I have 2 of those sitting in my NOC already. I also went to the demo. Overall, I'd say that once I got over the hassle of setting up the damn thing, the es7000s haven't ever really given me a problem. As far as setting them up even, a hard reset makes the machine post for 20minutes, when you're first putting in patches, THAT makes me agitated. well,that and if you ask unisys about it they'll tell you "it's supposed to do that".

that reminds me tho, I don't see you in OT all that much anymore spidey.

 

[ktwebb]

Now that I have spent some time with it I don't have as big a problem as I did. One of the major problems I had was, I had never seen this solution and the guy came in, sat with me about 20 minutes and off he went. Never even addressed the monitoring on the sentinel server. Never got an agreeable response as to why they can't put the service processor software and sentinel monitoring software on one box. Can't put SP2 on the XP box. Can't put SP1 on the itanium box when it's released until they do their thing, whenever that will be. The XP box only came with 512 MB of ram. If you let it run it will start paging within a week. No answer there from them. Can't get to the partition except through the RAC. I have a laundry list but had I been better satisfied with the intial support I might not have such a big chip on my shoulder. It's a very powerful box. No doubt about that. How could it not be without that much horsepower but I don't like the implementation and I have no respect for how they handled a very expensive transaction as it related to their intial support and training.

The funny thing is, the business unit that bought this could have done what they wanted on a poweredge 1850. It's completely absurd they are only going to utilize about 10% of the resources on a 100+k, 8 way 32GB ofs memory box.

 

[n0cmonkey]

Originally posted by: yoda291

Originally posted by: ktwebb
"actually a unisys es7000 would be around those IO numbers as well. I'd have to bench them and I don't have several hundred thousand dollars to throw around"

If I owned ours and not the company, I'd gladly give you a big discount just to get the piece of sh*t out of the server room and under my management watch. I was the available server tech when it was delivered and setup by Unisys so I got to be the ES7000 "expert" Unisys gave me a 20 minute demo on the sentinel server, the partition and the service processor. ES7000 /400 series. 410 I believe. I don't that one is several hundred thousand. Seems like I heard 130K. Might be the components in the package. 8 Itanium proc's, 32 GB or RAM. with moitoring 2k3 server (sentinel) on a Dell 1750 and XP management workstation (service Processor), some off brand, in the solution with the Partition (Itanium box). I have very little good to say about it, including Unisys's support.

I have 2 of those sitting in my NOC already. I also went to the demo. Overall, I'd say that once I got over the hassle of setting up the damn thing, the es7000s haven't ever really given me a problem. As far as setting them up even, a hard reset makes the machine post for 20minutes, when you're first putting in patches, THAT makes me agitated. well,that and if you ask unisys about it they'll tell you "it's supposed to do that".

that reminds me tho, I don't see you in OT all that much anymore spidey.

I had a SunFire 4800 that essentially posted for 30-45 minutes. Definitely annoying.

 

[spidey07]

something was seriously wrong with that box then.

but yeah, they are like 7 years old. Like most "real" servers they would run diags on all the blades but that only took like 1 minute per blade and IIRC the 4800 was 4 blades.