RE: Re: [WIRELESS-LAN] many clients, one room
John: Thanks for responding. Two points: - Its not reasonable to ignore retransmits. One of Merus key technology strengths is its claim to pseudo-schedule client access. This reduces retransmits due to collisions. Meru argues (and the last Novarum study appeared to demonstrate) that in dense client situations Merus approach provides a higher aggregate throughput per AP. If you recall one of the first graphics on their web site many years ago was of a chart with the number of clients along the x axis and aggregate throughput along the y-axis. I dont want to ignore the fact that the other vendors involved in Novarums test didnt have an opportunity to optimize their product or want to participate, but not unlike ATM and Token Ring, it appears that Merus approach, in situations of high client density, should outperform the traditional approach. In other words, in the PowerPoint scenario you described, Meru would do better than their competitors. Their competitors would argue that the network should be designed differently .. - More (non-overlapping) channels is almost always better. The enterprise WLAN vendors could stack multiple APs on top of each other, each operating at one or more non-overlapping 5 GHz frequencies, but omni-directional antennas will make channel planning difficult. Xirrus does a nice job of packaging that up, and its directionality increases coverage and limits co-channel interference with neighboring arrays. My summary viewpoint: most enterprise WLAN vendors have been able to avoid the channel-stacking and co-channel interference challenges because actual usage levels have been low, they havent had to worry about it. Theyve been granted a reprieve with 802.11n. While one might be tempted to say that this will catch up on them, I believe that raw speed will continually increase, either through more efficient modulation schemes or smart antenna technologies. Its a little like enterprise-deployed Ethernet we generally dont deploy QoS in our network, it was cheaper to go from hubs to switches, 10 to 100, 100 to 1000 Mps, and later, it will be 10 Gbps. Its a lazy approach, but it deals with usage and service level issue problem 99.99% of the time. Frank From: The EDUCAUSE Wireless Issues Constituent Group Listserv [mailto:[EMAIL PROTECTED] On Behalf Of Jon Freeman Sent: Friday, April 11, 2008 12:52 PM To: WIRELESS-LAN@LISTSERV.EDUCAUSE.EDU Subject: Re: [WIRELESS-LAN] many clients, one room Some math offers insight on this question Assuming the following: · were looking at a single area (i.e. lecture hall), · No retransmits are allowed (not real world, but is a best case example) · were talking about an average sized PPT of 10MB (looking through my PPT folder this was just my average) · Student and teacher expectations of speed is drawn from their homes (i.e. cable and DSL), less than this will be noticed and likely complained about · the rooms average data rate is 54Mbs (10 people by 10 people = 50ftx50ft) · 100 people, all downloading at the same time · max radio density for Meru is 3 (i.e. 3 channels of coverage, this is the most non-overlapping channels you can light in this area without interference problems using their latest gear) · Actual throughput for TCP data is 20Mbs per channel (54Mbs less Wi-Fi management overheads this is a number referred to in the 802.11 spec and one Ive observed many times) · Max radio density available from other shipping solutions today is 15 channels Meru Solution: · 20Mbs x 3 = 60Mbs converting to Bytes /8 = 7.5MB/sec /100 people = .075MB/sec (using 1024KB to the MB, this is 76KB/sec/user of TCP!) · Time to download 10MB/.075MB = 133 sec/user to download a 10MB file (about 2 minutes), so a 40MB file would take ~8min/user . · Link throughput then is 76KBs TCP for each user .you decide if thats acceptable 14 channel solution: · 20Mbs x 15 = 300Mbs · 5 times the bandwidth = 5 times the throughput · 76KBs/user x 5 = 380KBs TCP for each user of link throughput (and this is a little bit better than most uplink speeds on home broadband, http://www.speedtest.net www.speedtest.net is what Ive used on many LANs) · Instead of 2 minutes waiting, the 10MB file downloads with this solution in 26 seconds, and about 1 ½ min for a 40MB file, versus 8 minutes. So, we can assume that Franks interviews from 2 years ago dont account for the latest technologies. Sorry Frank, I dont mean to poke holes in your study, but it is 2 years old and we are talking about technology. Didnt we stop trying to manage limited bandwidth when ATM failed? When did we go back to thinking thats ok? I like more power, more speed, better, faster . Jon 303-808-2666 Xirrus Array...the Air is the Network...visit us at www.xirrus.com
RE: [WIRELESS-LAN] many clients, one room
Added a couple of notes to Frank's message below... Jon 303-808-2666 Xirrus(tm) Array...the Air is the Network(tm)...visit us at www.xirrus.com From: The EDUCAUSE Wireless Issues Constituent Group Listserv [mailto:[EMAIL PROTECTED] On Behalf Of Frank Bulk - iNAME Sent: Saturday, April 12, 2008 2:03 PM To: WIRELESS-LAN@LISTSERV.EDUCAUSE.EDU Subject: Re: [WIRELESS-LAN] many clients, one room John: Thanks for responding. Two points: - It's not reasonable to ignore retransmits. One of Meru's key technology strengths is its claim to pseudo-schedule client access. This reduces retransmits due to collisions (JON - true but what they don't point out is that this is similar to the 11g collision avoidance technique already part of the spec - I've not seen them argue they do any better than 22Mbs which is only a 10% difference resulting in a few seconds difference from the calculated result, not enough to compare to the 4 times faster demonstrated). Meru argues (and the last Novarum study appeared to demonstrate) that in dense client situations Meru's approach provides a higher aggregate throughput per AP (JON- as noted in my last comment, this may be, but the small percent difference can't come close to lighting up more total channels). If you recall one of the first graphics on their web site many years ago was of a chart with the number of clients along the x axis and aggregate throughput along the y-axis. I don't want to ignore the fact that the other vendors involved in Novarum's test didn't have an opportunity to optimize their product or want to participate, but not unlike ATM and Token Ring, it appears that Meru's approach, in situations of high client density, should outperform the traditional approach (JON - actually the opposite is true as the stand alone AP environment offers a new pool of capacity per AP where the Meru blanket approach only offers a single pool of capacity across multiple APs that everyone share, in effect creating a single hub for the entire area of coverage that is only 3 channels in size, so depending on the size of the coverage area the Meru approach could provide a significantly less amount of total bandwidth). In other words, in the PowerPoint scenario you described, Meru would do better than their competitors (JON - yes, this is true for everyone except the example used for the Xirrus Array which provides 4 times the speed, and since we're talking about classroom teaching time this difference is significant in terms of impact on the learning effect of students). Their competitors would argue that the network should be designed differently.(JON - actually most competitors might say that you can't support this number of people in a closed space since they will deal with near field interference issues) - More (non-overlapping) channels is almost always better (JON - we agree on this point completely). The enterprise WLAN vendors could stack multiple APs on top of each other, each operating at one or more non-overlapping 5 GHz frequencies, but omni-directional antennas will make channel planning difficult (JON - actually the planning would be more likely impossible as any APs placed in close proximity would cause each other near field interference, like what you hear when your cell phone is near your telephone, both operate on difference frequencies but their close proximity causes interference...the Array has several passive and active technologies that eliminate this problem, a benefit of integration that can't be solved by stacking APs, anyone who's tried stacking can offer their experience). Xirrus does a nice job of packaging that up, and it's directionality increases coverage and limits co-channel interference with neighboring arrays. (JON - agreed, and thank you!) My summary viewpoint: most enterprise WLAN vendors have been able to avoid the channel-stacking and co-channel interference challenges because actual usage levels have been low, they haven't had to worry about it (JON - true but we're seeing this problem coming to a head in about 30% of the Wi-Fi implementations today with a very rapid growth). They've been granted a reprieve with 802.11n (JON - .11n is now set for ratification in 2009, it does provide a good indication of the need for speed if you review the level of interest, FYI - the array with .11n will provide fast Ethernet switch replacement speeds - 12/24/48 port speeds, allowing you to get the switch benefit without the costs of the wires). While one might be tempted to say that this will catch up on them, I believe that raw speed will continually increase, either through more efficient modulation schemes or smart antenna technologies. It's a little like enterprise-deployed Ethernet - we generally don't deploy QoS in our network, it was cheaper to go from hubs to switches, 10 to 100, 100 to 1000 Mps, and later, it will be 10 Gbps. It's a lazy
Re: [WIRELESS-LAN] many clients, one room
Frank Bulk wrote: I’ll let others with production networks pipe in with their experiences. I'm missing this aspect of the current discussion. Does anyone out there have real-world experience to confirm or counter Jon's claims for Xirrus? -- Regards, -- Cal Frye, Network Administrator, Oberlin College www.calfrye.com, www.pitalabs.com The United States is in no sense founded upon the Christian doctrine. -- George Washington, founding father and first President ** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/.
