Matt
Take me down to the paradise city
Premium
join:2003-07-20
Jamestown, NC
 ·North State Commun..
| reply to djrobx
Re: [TWC] Pricing model changing soon
Informal Test |
said by djrobx  : quote:
Okay, you're up to 800Mb/s. Let's see you do 10Gb/s with a Windows machine.
I know this is really OT on such an active thread, but I do think Windows does seem to generally have trouble utilizing a gigabit network. I spent many hours trying to tweak and tune the performance of my server. Sometimes I can get them to transfer data at around 40gb/s, and even around 70-80gb/s "bursts" from cache, but the performance is not consistent and nowhere near the performance of the local disks. I think I've tried every single performance tuning article and adjusting every network setting there is. And these are with Intel server adapters! Local: DISKSPEED (C) Alexander Grigoriev, alegr@aha.ru Test File: "R:\$$test$$.tst" Test File Size: 512 MB Testing Uncached New File Write Speed.... Data Transfer: 40.21 MB/s, CPU Load: 2.4% Testing Uncached Write Speed.... Data Transfer: 56.12 MB/s, CPU Load: 2.9% Testing Uncached Read Speed.... Data Transfer: 29.34 MB/s, CPU Load: -0.2% Testing Cached Write Speed.... Data Transfer: 30.18 MB/s, CPU Load: 4.1% Testing Cached Read Speed.... Data Transfer: 50.97 MB/s, CPU Load: 6.2% Networked: DISKSPEED (C) Alexander Grigoriev, alegr@aha.ru Test File: "R:\$$test$$.tst" Test File Size: 512 MB Testing Uncached New File Write Speed.... Data Transfer: 13.16 MB/s, CPU Load: 1.2% Testing Uncached Write Speed.... Data Transfer: 6.53 MB/s, CPU Load: 0.9% Testing Uncached Read Speed.... Data Transfer: 6.49 MB/s, CPU Load: 1.1% Testing Cached Write Speed.... Data Transfer: 27.61 MB/s, CPU Load: 1.2% Testing Cached Read Speed.... Data Transfer: 10.16 MB/s, CPU Load: 1.7% With GigE I should have plenty of bandwidth to get close to my local disk speed. But the results are awful. It's just fast enough to tell me that I'm not stuck at 10/100. If I reboot the server the numbers may jump up a bit for a while. Those disk throughput numbers are terrible, I'm not surprised you have throughput trouble. My RAID arrays sustain over 160MB/sec and burst to 550MB/sec or so.
In addition, you're using a program written in 2002. You may want to try updating your benchmark tool. iOMeter or even SiSoft Sandra can test network shares if you want disk-to-disk network performance. To truly test OS level IP stack efficiency, you need something like iXChariot to test the connection by transferring from the memory of one server to the memory of another.
Anyway, sorry for hijacking your thread OP!
Perhaps a mod should split this discussion out and drop it in the network forum? |
|
B
Premium,MVM
join:2000-10-28
| reply to djrobx
You're using a mapped drive for your test and are thus relying on NetBIOS/SMB.
I assume you've tested with lower level transfers as well (ftp et al.?).
See also the Intel link I posted above, if you haven't already seen it.
-- B
--
In a realm outside causality and function |
|
djrobx
join:2000-05-31
Valencia, CA
 ·PHONE POWER
 ·AT&T U-Verse
·AT&T CallVantage
·Time Warner VOIP
 ·RoadRunner Cable
| reply to B
quote:
Okay, you're up to 800Mb/s. Let's see you do 10Gb/s with a Windows machine.
I know this is really OT on such an active thread, but I do think
Windows does seem to generally have trouble utilizing a gigabit network. I spent many hours trying to tweak and tune the performance of my server. Sometimes I can get them to transfer data at around 40gb/s, and even around 70-80gb/s "bursts" from cache, but the performance is not consistent and nowhere near the performance of the local disks. I think I've tried every single performance tuning article and adjusting every network setting there is. And these are with Intel server adapters!
Local:
DISKSPEED (C) Alexander Grigoriev, alegr@aha.ru
Test File: "R:\$$test$$.tst"
Test File Size: 512 MB
Testing Uncached New File Write Speed....
Data Transfer: 40.21 MB/s, CPU Load: 2.4%
Testing Uncached Write Speed....
Data Transfer: 56.12 MB/s, CPU Load: 2.9%
Testing Uncached Read Speed....
Data Transfer: 29.34 MB/s, CPU Load: -0.2%
Testing Cached Write Speed....
Data Transfer: 30.18 MB/s, CPU Load: 4.1%
Testing Cached Read Speed....
Data Transfer: 50.97 MB/s, CPU Load: 6.2%
Networked:
DISKSPEED (C) Alexander Grigoriev, alegr@aha.ru
Test File: "R:\$$test$$.tst"
Test File Size: 512 MB
Testing Uncached New File Write Speed....
Data Transfer: 13.16 MB/s, CPU Load: 1.2%
Testing Uncached Write Speed....
Data Transfer: 6.53 MB/s, CPU Load: 0.9%
Testing Uncached Read Speed....
Data Transfer: 6.49 MB/s, CPU Load: 1.1%
Testing Cached Write Speed....
Data Transfer: 27.61 MB/s, CPU Load: 1.2%
Testing Cached Read Speed....
Data Transfer: 10.16 MB/s, CPU Load: 1.7%
With GigE I should have plenty of bandwidth to get close to my local disk speed. But the results are awful. It's just fast enough to tell me that I'm not stuck at 10/100. If I reboot the server the numbers may jump up a bit for a while. |
|
B
Premium,MVM
join:2000-10-28
| reply to bigunk
I can't find much in the way of transfer rate comparisons for Linux v. Windows at 1 Gbps or 10 Gbps Ethernet.
»www.yale.edu/fastcamac/gigabit/g···nnt.html is quite old and indicates an initial advantage to Linux.
»www.enterprisenetworkingplanet.c···/3485486 comes very close but does NOT mention the Windows performance, only that Linux got 895 Mbps. They imply, but DON'T say, that Windows was in the same ballpark.
Finally, the devil is in the details -- »softwarecommunity.intel.com/arti···1259.htm shows that interactions of the XP OS and the ext3 file system, of all things, can lead to drastic differences in measured throughput.
Oh, and as to the OP, I'm sure Verizon will be very happy to hear about this.
-- B
--
In a realm outside causality and function |
|
Matt
Take me down to the paradise city
Premium
join:2003-07-20
Jamestown, NC
·North State Commun..
| reply to bigunk
said by bigunk :said by Matt :said by bigunk :Not to get off topic, but tell that to my gigabit Windows 2003 Servers. I routinely transfer 100+MB (megabytes) a second between my FTP servers. of course, they have extremely beefy internal RAID-10 arrays, but to say Windows can't do that is simply not true. My Windows machines and Linux machine both exhibit the same transfer rates, so we're likely switch/nic limited, not OS or software limited. Okay, you're up to 800Mb/s. Let's see you do 10Gb/s with a Windows machine. You give me the proof from a speed measurement tool that is recognized and accepted, and I will gladly change my position. I simply believe the *NIX world is faster and more efficient than Windows, and that is one of the gating factors. In other words it is the OS architecture that is to blame here. Can Windows get faster? Sure. Will it get faster? More than likely. I refer to today, not in the future (where *NIX will probably still show an edge). I'm not arguing, just pointing out that I have Windows machines doing 1Gbps, which you said wasn't possible.
I don't have 10Gbps capability and I really don't care to debate the merits of Windows vs. Linux. I run both Windows and Linux servers, they each have their strengths and weaknesses. I will point out however that Myricom makes 10Gbps cluster gear that runs on Windows using the standard Windows IP stack and 10Gbps gear for the IBM BladeCenter line has been available for ages. |
|
bigunk
Gort, Klattu Birada Nikto
join:2001-02-10
Santa Clarita, CA
·AT&T Yahoo
| reply to Matt
said by Matt :said by bigunk :Not to get off topic, but tell that to my gigabit Windows 2003 Servers. I routinely transfer 100+MB (megabytes) a second between my FTP servers. of course, they have extremely beefy internal RAID-10 arrays, but to say Windows can't do that is simply not true. My Windows machines and Linux machine both exhibit the same transfer rates, so we're likely switch/nic limited, not OS or software limited. Okay, you're up to 800Mb/s. Let's see you do 10Gb/s with a Windows machine. You give me the proof from a speed measurement tool that is recognized and accepted, and I will gladly change my position. I simply believe the *NIX world is faster and more efficient than Windows, and that is one of the gating factors. In other words it is the OS architecture that is to blame here. Can Windows get faster? Sure. Will it get faster? More than likely. I refer to today, not in the future (where *NIX will probably still show an edge).
--
There is not a man in the country that can't make a living for himself and family. But he can't make a living for them AND his government, the way his government is living. What the government has got to do is live as cheap as the people.
- Will Rogers |
|
Matt
Take me down to the paradise city
Premium
join:2003-07-20
Jamestown, NC
·North State Commun..
| reply to bigunk
said by bigunk :said by dentman42 :Only a gigabit?  That's just our tap off a much larger pipe. We plan to go faster in the near future. Problem is our ISP (who is connected to something like 12 different carriers) would need to put in a faster switch facing us, and so would we facing back to him. That's pricey stuff. Not to mention we would need to upgrade every NIC in every one of our machines if we want that bandwidth to propogate. Keep in mind Windows won't run that kind of bandwidth anyways. We can get our UNIX and LINUX machines to do it though. Not to get off topic, but tell that to my gigabit Windows 2003 Servers. I routinely transfer 100+MB (megabytes) a second between my FTP servers. of course, they have extremely beefy internal RAID-10 arrays, but to say Windows can't do that is simply not true. My Windows machines and Linux machine both exhibit the same transfer rates, so we're likely switch/nic limited, not OS or software limited. |
|
bigunk
Gort, Klattu Birada Nikto
join:2001-02-10
Santa Clarita, CA
·AT&T Yahoo
| reply to dentman42
said by dentman42 :Only a gigabit? That's just our tap off a much larger pipe. We plan to go faster in the near future. Problem is our ISP (who is connected to something like 12 different carriers) would need to put in a faster switch facing us, and so would we facing back to him. That's pricey stuff. Not to mention we would need to upgrade every NIC in every one of our machines if we want that bandwidth to propogate. Keep in mind Windows won't run that kind of bandwidth anyways. We can get our UNIX and LINUX machines to do it though.
--
There is not a man in the country that can't make a living for himself and family. But he can't make a living for them AND his government, the way his government is living. What the government has got to do is live as cheap as the people.
- Will Rogers |
|
69742511
Premium
join:2003-08-09
Costa Mesa, CA | reply to bigunk
Funny thing is, my sites are also in LA (well, Pomona to be exact); on solar powered hosting - and guess who their ISP is?
Yep, it's TW.
Business class though (of course). |
|
dentman42
join:2001-10-02
Columbus, OH | reply to bigunk
Only a gigabit? |
|
bigunk
Gort, Klattu Birada Nikto
join:2001-02-10
Santa Clarita, CA | reply to 69742511
You're always welcome to park your sites on my platform. I sit in a colo in LA, right on a tier 1 tap, with 1Gb/s of bandwidth. |
|
djrobx
join:2000-05-31
Valencia, CA
·PHONE POWER
·AT&T U-Verse
·AT&T CallVantage
·Time Warner VOIP
·RoadRunner Cable
1 edit | reply to saber11
quote:
Also new speed tiers by 3rd quarter.
This alone makes this whole thing suspect. Time Warner has no consistency across its markets in terms of speed tiers, so "new speed tiers by 3rd qarter" makes no sense, unless it is in reference to a single market. |
|
Hall
Premium,MVM
join:2000-04-28
Dayton, OH
·EarthLink
·AT&T Midwest
·Earthlink Cable Mo..
| reply to saber11
said by saber11 :I use to be a Tier III tech in western ohio, I got an email from someone in the RNOC... Which is in Columbus. I suspect that kshymkiw works there so.... |
|
69742511
Premium
join:2003-08-09
Costa Mesa, CA | reply to saber11
Man you better be joking. If they do that I'm either moving to where I can get FiOS; or going to (much) slower DSHell. I use a lot of bandwidth (forced to - I run 4 websites); so cable is my best option. |
|
saber11
Check Six
Premium
join:2000-06-09
Clayton, OH
1 edit | reply to kshymkiw
. |
|
