If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below. |
|
|
Thread Tools | Rate Thread | Display Modes |
#1
|
|||
|
|||
Packet Length: Who Determines It?
I tried this in comp.os.ms-windows.networking.tcp-ip,
but no nibbles.... ----------------------------------------------------- Been playing around with something called LAN Speed Test - between my desktop PC, through the router, and to a NAS box. Back around June, I was getting what I thought were pretty good speeds: 300-500 Mbps write speeds and 350-650 Mbps read speeds. On all of those tests, LAN Speed Test listed "Packet Length" as 1,000,000. Recently I came back to LAN Speed Test, and have been getting much lower speeds (as in 11-20 Mbps write and 60-90 Mbps read. For all those much slower tests, LAN Speed Test shows Packet Length as 10,000. I do not see any way to set Packet Length in LAN Speed Test's UI, so I am guessing that somebody else determines it. True? If so, who?.... and is there any hope for getting back to 1,000,000? -- Pete Cresswell |
Ads |
#2
|
|||
|
|||
Packet Length: Who Determines It?
Per (PeteCresswell):
On all of those tests, LAN Speed Test listed "Packet Length" as 1,000,000. Reading here-and-there, I'm coming away suspicious of the 1,000,000 number. Yes, LanSpeedTest really did cite it.... but it's looking unrealistic to me. viz this quote from http://tinyurl.com/qz2j47e "Long packets, such as the 1,518-byte packets we used, are a good way to stress a network." Also, I am getting the impression that Packet Length is determined by applications and not globally for the entire network. Am I on the right track? -- Pete Cresswell |
#3
|
|||
|
|||
Packet Length: Who Determines It?
(PeteCresswell) wrote:
Also, I am getting the impression that Packet Length is determined by applications and not globally for the entire network. Am I on the right track? Network hardware (switches and NICs) determine maximum frame size, not application. The standard size is 1500 bytes (plus 18 bytes of headers) normally it's only gigabit or faster switches that support jumbo frames of 9000 bytes plus headers, and even then not usually the low-end gigabit switches you might find in a home. |
#4
|
|||
|
|||
Packet Length: Who Determines It?
"(PeteCresswell)" wrote in message ... Per (PeteCresswell): On all of those tests, LAN Speed Test listed "Packet Length" as 1,000,000. Reading here-and-there, I'm coming away suspicious of the 1,000,000 number. Yes, LanSpeedTest really did cite it.... but it's looking unrealistic to me. viz this quote from http://tinyurl.com/qz2j47e "Long packets, such as the 1,518-byte packets we used, are a good way to stress a network." Also, I am getting the impression that Packet Length is determined by applications and not globally for the entire network. Am I on the right track? Pete, Not sure if this is applicable but, I recently setup a small LAN with 1 Windows 2008 Server and 25 Windows 7 64-bit workstations and a NAS device. Data transfer seemed slow (every device including Router and Ethernet switch is gigabit) I found this setting on the NIC's that made a huge difference on LAN transfers of data. Large Send Offload v2 (IPv4) This is found by right-clicking on your NIC (I assume you know how to get to and change properties on your NIC, if not let me know) and selecting Properties. Then click on "Configure" then click on the "Advanced" tab. There are several parameters (depending on the manufacturer of your NIC) that you can configure. Large Send Offload v2 (IPv4) is enabled by default (I am assuming you are using IPv4, if not select Large Send Offload v2 (IPv6) By disabling this parameter my LAN speeds increased dramatically specifically the backup to the NAS device. I have begun to revisit some of my other recent installs and made this change resulting in backup times going from several hours to less than 1 hour to the LAN NAS device. Hope this helps! JT -- Pete Cresswell |
#5
|
|||
|
|||
Packet Length: Who Determines It?
(PeteCresswell) wrote:
Per (PeteCresswell): On all of those tests, LAN Speed Test listed "Packet Length" as 1,000,000. Reading here-and-there, I'm coming away suspicious of the 1,000,000 number. Yes, LanSpeedTest really did cite it.... but it's looking unrealistic to me. viz this quote from http://tinyurl.com/qz2j47e "Long packets, such as the 1,518-byte packets we used, are a good way to stress a network." Also, I am getting the impression that Packet Length is determined by applications and not globally for the entire network. Am I on the right track? The article on Jumbo Frames says 9000 is the accepted limit. http://en.wikipedia.org/wiki/Jumbo_frame I've only run Jumbo just the one time, directly, between two computers. In my mixed network with decrepit gear here and there, I can't fix everything to run Jumbo, and it would likely break at the router anyway. If you want a speed test, try this one. It's for computer to computer. You start a receiver copy on one machine. Then transmit to that machine, from another machine. And it'll measure the speed. http://www.pcausa.com/Utilities/ttcpdown1.htm The reason I was using that, is I have a file sharing setup with a 20MB/sec visible limit. And I can't figure out why it is running slow. The weird part is, if I start a copy of "dumpcap" tied to the port in question, file sharing runs faster. But I don't know what that means and why that should make a difference. Paul |
#6
|
|||
|
|||
Packet Length: Who Determines It?
On Wed, 17 Sep 2014 17:23:08 -0400, "(PeteCresswell)" wrote:
I tried this in comp.os.ms-windows.networking.tcp-ip, but no nibbles.... ----------------------------------------------------- Been playing around with something called LAN Speed Test - between my desktop PC, through the router, and to a NAS box. Back around June, I was getting what I thought were pretty good speeds: 300-500 Mbps write speeds and 350-650 Mbps read speeds. On all of those tests, LAN Speed Test listed "Packet Length" as 1,000,000. Recently I came back to LAN Speed Test, and have been getting much lower speeds (as in 11-20 Mbps write and 60-90 Mbps read. For all those much slower tests, LAN Speed Test shows Packet Length as 10,000. I do not see any way to set Packet Length in LAN Speed Test's UI, so I am guessing that somebody else determines it. True? If so, who?.... and is there any hope for getting back to 1,000,000? Hi Pete, I know you're asking about packet size, but others have answered that so I wanted to suggest something else. The faster speeds make me think you were testing on a Gigabit network, while the slower speeds make me think you were testing on a Fast Ethernet (100 megabit) network. Since the throughput speed of any connection is only as fast as the slowest link, is it possible that the second set of tests were run at 100 meg? The reason I ask is that I've had that happen to me. -- Char Jackson |
#7
|
|||
|
|||
Packet Length: Who Determines It?
JT wrote:
"(PeteCresswell)" wrote in message ... Per (PeteCresswell): On all of those tests, LAN Speed Test listed "Packet Length" as 1,000,000. Reading here-and-there, I'm coming away suspicious of the 1,000,000 number. Yes, LanSpeedTest really did cite it.... but it's looking unrealistic to me. viz this quote from http://tinyurl.com/qz2j47e "Long packets, such as the 1,518-byte packets we used, are a good way to stress a network." Also, I am getting the impression that Packet Length is determined by applications and not globally for the entire network. Am I on the right track? Pete, Not sure if this is applicable but, I recently setup a small LAN with 1 Windows 2008 Server and 25 Windows 7 64-bit workstations and a NAS device. Data transfer seemed slow (every device including Router and Ethernet switch is gigabit) I found this setting on the NIC's that made a huge difference on LAN transfers of data. Large Send Offload v2 (IPv4) This is found by right-clicking on your NIC (I assume you know how to get to and change properties on your NIC, if not let me know) and selecting Properties. Then click on "Configure" then click on the "Advanced" tab. There are several parameters (depending on the manufacturer of your NIC) that you can configure. Large Send Offload v2 (IPv4) is enabled by default (I am assuming you are using IPv4, if not select Large Send Offload v2 (IPv6) By disabling this parameter my LAN speeds increased dramatically specifically the backup to the NAS device. I have begun to revisit some of my other recent installs and made this change resulting in backup times going from several hours to less than 1 hour to the LAN NAS device. Hope this helps! JT Some NICs are just badly designed. They cause an excessive number of interrupts per handled packet. And no amount of fiddling with the NDIS features, will rescue them. And the after-market for NICs is practically gutted now. You just can't find anything worth buying. Of the couple of Intel chips, one seems to have an uncorrected driver problem. I want a chip like the old Marvell one in my P4 machine, as that one worked well. I don't know if it is an inability to support the latest NDIS that causes chip designs to "fall out of the market" or what. But I wish the process could be reversed. Paul |
#8
|
|||
|
|||
Packet Length: Who Determines It?
Per Char Jackson:
I know you're asking about packet size, but others have answered that so I wanted to suggest something else. The faster speeds make me think you were testing on a Gigabit network, while the slower speeds make me think you were testing on a Fast Ethernet (100 megabit) network. Since the throughput speed of any connection is only as fast as the slowest link, is it possible that the second set of tests were run at 100 meg? The reason I ask is that I've had that happen to me. You are correct in that I *think* I'm running gigabit.... Was thinking that something in the network might be dragging me down to 100, so I ran the test with the PC and NAS plugged directly in to the router and everything else disconnected.... but got the same results. -- Pete Cresswell |
#9
|
|||
|
|||
Packet Length: Who Determines It?
On Thu, 18 Sep 2014 12:38:39 -0400, "(PeteCresswell)" wrote:
Per Char Jackson: I know you're asking about packet size, but others have answered that so I wanted to suggest something else. The faster speeds make me think you were testing on a Gigabit network, while the slower speeds make me think you were testing on a Fast Ethernet (100 megabit) network. Since the throughput speed of any connection is only as fast as the slowest link, is it possible that the second set of tests were run at 100 meg? The reason I ask is that I've had that happen to me. You are correct in that I *think* I'm running gigabit.... Was thinking that something in the network might be dragging me down to 100, so I ran the test with the PC and NAS plugged directly in to the router and everything else disconnected.... but got the same results. Which begs the question, is the router's embedded switch running at Gig speeds? And did it negotiate Gigabit connections to the NAS and the PC? :-) If either of those connections didn't negotiate properly, then the overall connection won't be GigE. I thought you might say that you connected the NAS directly to the PC as a test. If both are GigE-capable, they'll properly figure out the MDI/MDIX question. Windows will tell you, in Task Manager, what speed it's connected at. Never mind, I may be chasing rabbits. It's just that GigE requires all of the cable pairs in the Ethernet cable to actually be connected properly, while FE (100M) doesn't, so a bad cable can drop you back to FE. But as a second test, try using iPerf or jPerf. Both are free and both work well. I especially like jPerf, but I don't remember why. Neither tool writes anything to the filesystem during the test, so it's just a 'network speed test' type of tool. A couple of years ago I was getting 300-450 megabits/sec on one segment of my LAN, so I assumed it was the cheap $6 NICs I was using. Using jPerf, though, the test reported 960 mbps+, so it wasn't the NICs. In my case, it was a slow/dying drive at one end. -- Char Jackson |
Thread Tools | |
Display Modes | Rate This Thread |
|
|