sietec

I might have already replied to this topic a while ago, so if I did, forgive me...I just didn't check whether I had or not. I'd just like to give a quick summary of my view of TMN vs. "the others." I am a networking guy by profession as well as by degree and certifications, so it is natural for me to be the "curious cat" about everything networking and to try to "fool the system" (e.g. find bugs that cause erroneous results) as well as attempt to prove or disprove the validity of someone's claim (in this case, the accuracy of TMN). Many (probably most) people do not realize that there have been TCP (and other transports) benchmarking for just about as long as the transport itself has been around. Some of the most powerful are command line tools found (typically) in Linux systems that offer extreme flexibility in testing (e.g. packet sizes, compression algorithms, hardware offload for things like checksums and VLAN or QOS tagging, certain kernel path bypass mechanisms, window scaling heuristics, and literally dozens of other adjustable parameters to test the maximum Tx/Rx speed between two or more computers). That said, I've used pretty much all of them at one point or another and have done very meticulous comparisons to the results on TMN. The margin of error is astonishingly low (generally less than 5%). Compared to Ookla's Flash based test, this is a factor of 10 difference in accuracy because the average from that site that I have found is around 50% (with a huge standard deviation). One day, I will perform the tests again and post the results in a forum here..I didn't save everything last time and want to make my post "legitimate" by including all methods and screenshots utilized. I'll try to get around to it sometime soon. That was the first thing to get out of the way. Secondly, as explained in other posts, Flash is a VERY expensive technology (expensive meaning processor and memory intensive) and adds quite a bit of latency as well due to the complex paths the data flow must go through. Is it appealing to the eye? Absolutely. Would I choose the most graphically appealing test if my goal is to get the most accurate results? Absolutely not. See paragraph above...command line is about as ugly as you can get - but also as accurate as you can get (using the correct tools). I'll preface point three by saying that I don't know enough about the internals of either TMN or Ookla to make a totally accurate claim about their inner workings but I may through my obversations, I can assure you that Ookla has major problems with regard to the results. One of the best ways to test this on your own is to use a program called Wireshark and start a capture of your network packets (make sure to choose the correct network card!!) during a TMN test and an Ookla test. The first thing to notice is the amount of data that is transferred during a test. I cannot figure out, for the life of me, the algorithm with which Ookla determines how much actual data to transfer. By "data," I mean, for example, how many megabytes are transferred to your box during the test. Second thing I noticed was a lot of "noise" in the packets that seemed to be upstream communication to the Ookla host server from my computer during a download test (NOT ACKS, so please don't call me out saying it was ACKS). There is some type of communication to the server going on - which OBVIOUSLY interferes with an accurate download score if a Download test is also simultaneously transmitting information other than standard TCP Acknowledgments, replies, etc. That doesn't occur on TMN. There is simply an ACK and SYN as expected during a raw transfer. The TMN server determines whether or not more data is needed to determine an ACCURATE result based on how quickly you down- or uploaded the information (7 seconds down and 5 seconds up for a specific amount of data transfer). Assuming a download, at first you will receive the smallest continuous piece of data and if it took less than 7 seconds to transfer, TMN will push the next size to you. This process repeats until: The seven seconds expire and you have not received the entire download chunk - or - You reach the maximum size (200MB) and complete the download in less than seven seconds. This way, during the download, the only cost incurred on your PC is that of the Operating System's networking routines and the CPU usage for offloaded tasks (for instance, checksum offload). BTW, this occurs during ANY network communication and there is no way around it. Another thing to consider if you are receiving results that are inconsistent amongst testing sites is the location of the hosting server. If you go to speedtest and live in Atlanta, Speedtest will choose the location closest to you with the least latency (and, in this case, would be in Atlanta). The further a byte has to travel, the more latency introduced and (generally) more hops must be taken to reach the destination. All of which introduces decreases in speed with increase in hops and latency. So, if you're testing on Speedtest in Atlanta on an Atlanta based server and then hop over to TMN and use a Dallas server, it is only natural to expect that the transfer speed will (again, typically) be slower and vice-versa. So, a more accurate way to compare the sites would be to choose a Dallas location on speedtest, take the test and then test via Dallas on TMN. Or, you can just trust me..TMN is better The last point I'll make in this post is that with TMN, the data transfer occurs via standard HTTP, which is how the vast majority of your real world downloads and browsing occurs. One exception is on a secure site that uses SSL and is preceded with "https://" - that normally occurs on port 443 instead of 80 as in HTTP and incurs a heavy performance penalty for the encryption and decryption of the data after is is received. There are tons of other protocols such as FTP, SCP, SSH, CIFS, SMB, NFS, etc. but, like I said, 99% of the typical user's internet browsing occurs on HTTP. I don't know exactly how the data is transmitted and received on Ookla based sites but I do not believe it is HTTP - I think it is an embedded part of the Flash wrapper. So, to close this post that I meant to be short and to the point and went way overboard, my opinion and experience is that the most accurate measurement of your bandwidth is going to be found on TMN. I apologize for the rambling. I hope at least someone finds this helpful! Take Care....more to come (in the future, sometime!) --SIETEC--

From a SUSE Open Linux Virtual Machine with a VXNET3 NIC for the public IF. If you are interested in the setup of my environment, read on below the link...warning: you might get bored because I tend to ramble sometimes It's been a while since I've had the pleasure of visiting TMN, been out of the country for some time now. Nice to be back...anyway, I've been tweaking a SUSE box (SUSE is my new obsession recently) that I have installed as a network lab on one of my ESXi vCenter servers....anyway, I'll get to the point. Just completed a transition from the "default" Enterprise Plus packaged Distributed Virtual Switches to using the Cisco Nexus 1000V DVS. Not only is it AWESOME to have the IOS Cisco shell to work with in the virtual environment, but that switch kicks the sh** out of the standard DVS. If you're an advanced and enthusiastic VMware (or Xen, for that matter....and, I guess, also *gag* Hyper-V...), install the 1000V and get back control of your networking. It is a really powerful box and Cisco offers a 100% fully free perpetual version now with very few limiting features. No, I promise I don't work for Cisco. Now, here is the tech specs of the server for those interested (the cluster on which this test was run): Three Dedicated, 2U each, Servers with the following identical configuration on each one: Supermicro X9Dri-LN4+ PROC Intel Xeon (Sandy Bridge Chipset) E5-4620 Quad-Socket, Hex-Core (48 Logical Processors) @ 2.00GHz each (96GHz per server) MEM 128GB RAM (server0) 128GB (server1) 192GB (server2) ETH Utilizing 6 - 1Gb, 2-10Gb and 2-100Mbit NICs per server; configured as follows: Public (Internet) - 6 Bonded gigabit NICs with individual port profiles at the upstream switch and Active LACP (link aggregation) enabled on the servers with load balancing and distribution via IP Hash algorithm. Private (Internal network) - 2 individual 10Gig with zero link aggregation but using multiple failover modes and logical seperation with both VLANs and PVLANs and intra-server, non-routed communication via layer 2 VXLANs (virtual eXtensible lans...which make non-physically connected networks behave as if they were all connected to the same switch...then the same subnet without routing is able to be used...plus adds a huge layer of security). The private network is mainly for SAN and iSCSI and management/intraserver traffic 802.1q (QinQ) functionality is provided using two 100Mbit uplinks which enable STP (spanning tree protocol) and Cisco portfast configs to be immediately learned by the upstream switches (e.g. a VM powers on and has a different MAC address than the previous occupant of the associated IP address, typically the upstream switch tables would take from 20-60 seconds to update, whereby the port would not be "up" until the switch updated...portfast and the QinQ allows immediate failover and MAC learning .... thus, only need 100Mbit cards for this since it is simply "telling" the upstream switches (2 of them, hence two cards) the virtual environment network parameters). STORAGE 20TB shared Network attached storage (NFS, iSCSI, SMB, etc) and 5TB local (RAID striped and mirrored) per server Currently the three host servers in this cluster are configured for fault tolerance and transparent failover of any VM that operates on it and, on average, there are approximately 100 virtual machines running at any given point in time with average host utiliztion of about 10% CPU and 30-50% RAM and around 110Mbps 24h average network transfer on the public network and 700Mbps average on the private network with peak of around 1200Mbit public and 7500Mbit private (during backups, vmotion, etc). This puts the system at a very cool 10% overall load average (plenty of room for more VMs) There are 7 Windows server 2008R2 boxes, 5 Win server 2012R2, about 25 Linux web and file servers and then the standard infrastructure of SQL, Oracle DB servers and etc. Also running is our new Virtual desktop platform with linked clones which provide "thin client" access of Windows 7 Enterprise machines through HTML5 or Horizon View PCoIP via IPSEC VPN access. Basically, I explained all of that in order to say that computers today are so unbelievably powerful, that with some basic simple tweaking, you can get unreal performance out of them. I mean, there were 30 other "computers" running on the same host that this SUSE box was running on, simultaneously, and the performance is still mind blowing. SIETEC over and out...later

Hello again, My apologies for the long time away...it's been crazy! I was reading comments on the MTU and maybe I should have given a little more information before just spitting it out there! there are few reasons "normal" systems need an MTU above 1500, but like FiberGuy said..in general it is an auto adjusting thing. If the OS is old, or jumbo frames not supported, it can cause problems. However, I seem to notice more problems when I have the MTU set at 1500 than up at 9000 .. mostly because one of my second obessions is VPN with IPSEC and/or L2TP over IPSEC and anyone who plays around with these things know that you actually have to use fragmentation in order to keep the packet overhead from putting the overall packet size over 1500. Or set the packet size to 1436 in the host server and use an autoconfig for the client (e.g. during the phase 1 negotiation). So, in order to maximize the bandwidth and minimize latency, I always make sure (even my home boxes) are set at jumbo frames and I have noticed an increase of about 40% overall performance when operating under encrypted VPN. Again, in most cases, adjusting the MTU will do no harm, but I would stress that anyone changing these settings, who isn't comfortable in doing so, check with someone who can help first. You don't want to completely lose your internet connectivity! Since I've been back in town, I've been scouring resources on what to put in my next post here and I plan on updating it over the next day or so. The first one was just quick, off the top of my head stuff, but I have tried to come up with some more generic, applicable to everyone suggestions for the next go. That is, suggestions that aren't redundant, and with what a great job of thoroughness the TMN team has done, it's hard to come up with unique ideas that will actually help. Lastly, I didn't want to come off as just an advertisement for my web services -- my main source of income has nothing to do with networking/etc. I just have always wanted to put together an extremely reliable and FAST, SECURE, platform for advanced users who need a platform that is completely customizable to do with what they want. I will be offering completely customizable systems on any operating system you like, with full admin/root access with up to 48Ghz processor resources and 24GB RAM per system with a 2GB/s public port. You will have full control of what you do on it (within the law, of course!) Essentially, I have completed my initial testing and security analysis and should be ready to go very shortly. PM me if you have interest, but I definitely want to keep the topic here on improving speeds, not on the services I have to offer!! Hope all is well with everyone here...I'll talk to you later! Zac

I will preface this post by saying that I am a Windows dummy, so I don't have a whole lot to offer to the Microsoft bunch out there. However, my quick little tips don't apply so much to a particular OS as to general system settings good for almost any OS, so here goes: 1. Minimize the amount of running services on your system. I cannot overstress the importance of this one -- the more services & applications you have running simultaneously on your box, the more overhead utilized. So many people have all kinds of virus scanners, supposed "speed up" apps, unnecessary running services/daemons/etc. which do not need to be running 24/7. For instance, say you have a virus scanning platform. **Please note, I am not advocating that you turn off your virus protection -- and only you can determine the level of protection you need, so in the end, use your own judgement and don't hold me responsible (please) ** Anyway, virus scanners are generally not needed constantly. You're not going to get a virus if you don't have one already and you are not downloading files, visiting malicious sites or running untrusted software. There are times the preceding statement if false, but rarely enough to cause concern. So, I suggest that you only have your virus software set to scan files on demand - e.g. when you download something or when you are reading emails. Use your browser security settings to help minimize the chance you wonder into an insecure site and always make sure the sites you are using for sensitive information (such as bill payment, purchases, sensitive data, etc) are encrypted with trusted encryption (check the pad lock icon and make sure it is a "https" connection). Using your browser to help with security is a performance sparing method to minimize the processor intensive constant scanning by a virus detection package. If you have a ton of system services running in the system tray of Windows, take a look at which ones you really need and eliminate the others. Same for system startup -- clear out everything that does NOT need to be started upon startup; this will save memory (RAM) and processor usage...which leads to quicker page response and transfer speeds. Advanced users should check actual "services" in Windows and turn off those you don't need. If you're running a linux flavor, I assume you're already pretty versed in modifying your configuration. If not, PM me. e.g. . chkconfig --list to see a list of services or service --status-all to see what's running currently (RH/CentOS/etc.) 2. Change your MTU (max transmission unit) to 9000 if you're on a high speed cable or ethernet link. This allows your network interface to send much larger frames than the standard 1500 value. Basically, this parameter sets how large a single transmission frame may be, minus overhead. So, if your system includes overhead in the setting, set it to less than 9000 so you don't get an error for exceeding 9000 (e.g. set it to 8900 to be safe). Just make sure you do this if you know what you're doing only and if you're sure it is appropriate for your set up. It shouldn't cause any problems, but make sure you know how to get back in there and change it to the previous value in case it breaks your connectivity (which I doubt!) 3. Try different browsers - not only are some browsers actually much better at speed natively, they save on system resource usage as well. My suggestion: firefox or opera. I was never a fan of opera until recently when they made a lot of improvements. Firefox is always a good choice. IE would be my last choice, but many people love it. Chrome is pretty good too but I have noticed some issues with it lately. 4. Stop background file transfer daemons such as online data backup utilities, cloud sync systems (iCloud, etc). If you have things like that running in the background, they obviously will take a lot of your bandwidth away (unless you specifically configure them not to, they will use as much bandwidth as they can get their hands on). So, shut down iCloud from syncing while the system is active and if you use something like Carbonite backup, let it work when you are not using your box (e.g. overnight). 5. If you are using wifi, make sure you have a good signal and make sure you are using the faster encryption settings. You should use 802.11g at a minimum and preferably 802.11n for best connectivity. Most network connections (note I said most) will not realize much (if any) performance gain by going direct ethernet unless the previous 3 suggestions (signal, encryption, protocol) are not followed. If you cannot adjust the first 3 settings, then you probably should connect directly to your router's ethernet connection or to your cable modem/DSL/etc. You can also change the "power" or "energy" saving settings on your wifi box by setting the transmission power to 100% all the time. Search your manufacturers KB or google for "increasing wifi transmission power" or, more generally, "tutorial to speed up wifi" Note that wifi is going to give you a theoretical maximum of 450Mbits per second in the best and hardly ever realized conditions. Generally 300Mbits is as fast as you'll see and if you're not using 802.11n, 54Mbits optimum. These optimal speeds are what the manufacturers tout, but it is nearly impossible to see them. Believe me, wifi can interfere (pardon the pun ) with even a moderate cable speed of 30-40 Mbits/second. Summary: wired is always better, but there are ways to maximize your wifi connection. I will put the last 5 of my first 10 tips online over the next few days ... but, please feel free to reply to this post or PM me if you need help doing these things. I hope some of this helped! SIETEC President & Lead Network Engineer SIETECserverNet :: sietecFASTcom :: sietecMATRIXcom Coming soon, I will be launching a hosting service unlike anything you have ever seen -- take a look at my test results here: My clients will be able to achieve similar, depending on the service chosen, and it is my goal to help them do so. Interested in pre-sales information? Drop me a line at [email protected] and let me know what you are looking for ... deep discounts to the first 25 people! My website is being re done right now, but I basically offer anything most hosting companines offer, just in a virtual box. (which, believe it or not, that is how I achieved most of these results, through one of my virtual machines). You still get to pick how many processors, RAM, type of HD and space on disk, NIC speeds, etc. You can even choose a free virtual appliance that suits what you are doing and launch it on my platform. Or, I can assist you in installing your OS and let you take it from there .. full root/administrator access! Again, if interested, email or PM me!

Hi all, I just wanted to put my two cents in here for what it's worth. First of all, I have no interest in any online speed test. I host mirrors for the two most popular speed test sites, one of which you are on right now. My only interest is in helping people get accurate and informed information when it comes to measuring their internet speed. I will probably upset some people with this post, but what I am saying comes from my 12+ years of experience in tweaking web servers and meticulously measuring the details of different transports. For http (e.g. "internet" or web-browser compatible) speed testing, I can say with 100% certainty that you will not find a testing methodology more accurate than testmy.net. The attention to detail and every little tidbit of information involved in http/tcp transport is second to none. I can say that since I host a mirror on here and can measure the results from both ends. For example, if you make a request on my box, apache keeps a log of the transfer information and precisely how long the socket was open and how much information (exactly) was transferred. The timing is accurate to less than 1/10000 of a second. Therefore, I can measure the precise time it took to transfer, say 100MB from my box to your computer and determine from that the speed in which the transfer took place. TMN goes a bit further than most, in that you can see the actual speeds during the xfer as it fluctuates. The other speed test engines take an average of your speed over the duration of xfer and throw away a significant amount of the data as "erroneous." My question is, if it is erroneous yet contributes to the overall time it takes to transfer a file, it is NOT erroneous and should NOT be discarded as such. This is where you will see significant differences on the "other" sites. Sometimes it is disappointing to see the real numbers your machine is capable of. Believe me, I know. I get upset when I think a measurement is wrong. But, it gives me valuable insight into how to increase the throughput and achieve better results. Take a look at my max and averages These come from my VNC (e.g. linux remote desktop) connections to one of my boxes using Firefox or other browsers to test the box. The average has been skewed by testing all kinds of different settings (e.g. jumbo frames, different MTUs, browsers, servers, etc) and so could be much higher if I dedicated an account singly to testing with my best settings...but that would kind of defeat the purpose, in my opinion! If you consider the overhead on the processor and network interface just to support my 1920x1080 VNC connection, then factor in that I am remoting into a server to test it's connection; furthermore that it is just a TCP link with limited threads that can be opened (due to the nature of the protocol), my box is capable of much more. But when it comes to true INTERNET ability, I have never measured anything faster than I can through testmy.net. Consider that the "other" sites use Flash and other visual enhancements, a limited amount of data and basically, nondiscriminatly discards data assumed to be erroneous, which site do you think will be more accurate? I will soon be starting a blog regarding how to improve you internet connection speed and will post the link when I publish it. I would like people to try before and after modifying system settings as well as the other speed test sites. I say everyone should try the other sites as well and see where the discrepancy lies. But if you are truly looking to improve your performance, in my opinion, you have found the spot (that is, unless you want to write your own code and try to start from scratch what has taken a decade to get where it is...good luck by the way). Anyone with questions, please feel free to ask. Understand, please, that I am often distracted and cannot immediately reply -- but I will reply. Probably your best source of information is in the threads on this site, or directly from Damon. He is very active on his site and generally answers anyones questions or posts fairly rapidly. Thanks for reading and forgive my spelling and/or rambling! Good luck! SIETEC

Does it "stall" during the upload (e.g. while showing the progress bar) or after the progress bar reaches 100% and when you should be getting the results? SIETEC

Thank you, I have been very impressed with this box - I wouldn't complain about anything because, frankly, I was shocked to see a browser via http over one pipe perform like this. Hopefully you are right and browsers will implement additional protocols for general use. I think you are correct, everything is very symmetrical as far as the upstream/downstream xfer goes. I would imagine you need at least 7-10 seconds in order to measure the speed accurately when you account for the propogation delay, handshakes, etc. I imagine 33MB is generally plenty...most people don't check in with an unmetered gigibit device You're welcome and I look forward to seeing what you are working on! I definitely will be one of your frequent visitors; this is by far the most realistic and accurate site I have found thus far for bandwidth tests. I can't stand speedtest.net, etc. with those silly .swf apps that look like a speedometer or whatever. Take care, talk with you soon . Zac

Test ran 235.3Mbps (28.7 MB/s) Down & 122.4Mbps (15.3 MB/s) Upstream This is, of course, not your typical home computer. This is my dedicated Centos box with two 1Gbps Connections, unmetered, no throttling. I pay $490 per month for this machine and the 12 IP addresses on it. Not bad for such a blazing fast machine...it is awesome! I have actually seen through the network monitors on the box, with simultaneous connections from multiple VM's that I was transferring files from...I have seen >60MB/s on the downstream node and the upstream was doing about 20 MB/s....it was close to 700 Mb/s together so it was pushing one ofo those gigibit connections pretty close! Awesome site here - and great job with the coding! Zac