[hcs-d] Ethernet speed

Joshua Kroll jkroll at fas.harvard.edu
Thu Apr 23 01:11:57 EDT 2009


Apparently, I am not smart enough to hit reply-all.

Josh

On Wed, Apr 22, 2009 at 11:04 AM, Joshua Kroll <jkroll at fas.harvard.edu> wrote:
> This is in fact not the case.
>
> Also: since this e-mail is long, a one sentence summary. They upgraded
> your switch (probably).
>
> As background, FAS IT is in the last stages of a massive, multi-year
> project to rebuild the entire (wired) network. A few years back, they
> replaced the ailing network core with four Cisco 7600's and laid down
> redundant 10Gbit connections between these and also to their border.
> They then replaced the distribution layer over the course of almost a
> year. Finally, they started on the several hundred "edge" switches,
> which you might see in your dorm basement or hear in mechanical
> closets in buildings around campus. The goal was end-to-end gigabit
> connectivity for everyone. I'm told that all or almost all of the edge
> switches have OC3 or better connectivity to the MDF/distribution layer
> and connectivity is often better between the MDF and the core. Also,
> the core is now spread between multiple datacenters, so things like
> the fire of a few years back won't knock out the (whole) network
> anymore.
>
> As part of the "edge project" as they call it, FAS IT is ramping up
> jack speeds as they go around campus. So what likely happened is that
> the switch in your basement was replaced and the jack speed was turned
> up (this is a setting on the switch which is trivial for them to
> change). As I said, the goal is gigabit, but that's not always
> possible (see below). They tend to turn things up as fast as they can.
> If you have a legit need for more bandwidth, things can also often be
> worked out. The networking group is a really nice and very capable set
> of people.
>
> Now, many buildings do have "inadequate wiring" - specifically CAT3.
> Also, some of this wiring is old, has had its cladding degraded, has
> had nails punched through it, etc. This obviously limits traffic by
> increasing packet loss and by preventing jack speeds from being turned
> up when switches are replaced. The eventual goal, I believe, is to
> rewire these buildings, but this is limited by funding and the fact
> that this often requires putting in new conduit (CAT5e is much thicker
> than CAT3), which would involve opening up the walls, making it a
> major capital project involving downtime of the building in question.
> I hear there are plans to fix this for the house renewal, but who
> knows what's happening with all the budget nonsense now.
>
> Finally, YMMV with respect to what you're doing, since there are
> congestion control policies that I know little about (the network
> architect promised a few times to tell me, but I haven't seen him
> recently to ask). I've noticed that purely UDP traffic is ratelimited
> or dropped much more than TCP traffic, but that was in pretty simple
> tests under really unusual conditions, making it unlikely that I
> really understand what's happening. There's definitely more congestion
> control over the border, so I'd avoid doing over-the-border tests to
> try to understand your bandwidth situation.
>
> Interestingly, I started playing with network testing a few weeks back
> when I noticed that the 1.0MByte/s connectivity I'd had for most of
> the year to machines in FAS IT run datacenters jumped to about
> 50.0MByte/s. I had thought my wiring in Adams was CAT3 (indeed, my
> jack is listed as having CAT3 in autoreg). Nonetheless, I can now push
> and pull TCP at over 500Mbit to HCS's new machines in Converse lab,
> which have reasonable gigabit connectivity. So I imagine that I
> actually have a decent wire. I suspect that either the switch in my
> entryway was replaced or that when I asked to be moved to a static
> VLAN (i.e. to get a static IP) they also upped my jack speed. I
> imagine I'd get better connectivity to machines that are closer to the
> core - our machines are one extra network layer down because of how
> they have our switch wired.
>
> Anyway, if you're curious about this stuff, you can come to office
> hours and we can explain and help you test accordingly. You can also
> help us figure out some of the things we don't understand or join one
> of our many network-related projects.
>
> Josh
>
> On Wed, Apr 22, 2009 at 2:56 AM, Jason Gao <jasongao.root at gmail.com> wrote:
>> IIRC, all the dorms are wired for gigabit, but FAS IT keeps them at
>> 10Mbit unless someone requests more bandwidth.
>>
>> -Jasom
>>
>> On Wed, Apr 22, 2009 at 2:48 AM, Zach Rait <rait at fas.harvard.edu> wrote:
>>> Maybe I've been living under a rock (or, more likely, only in Lamont), but I
>>> could have sworn that dorm-room Ethernet was 10Base-T as a result of
>>> purportedly "inadequate wiring."  Until recently, I had been clocking in at
>>> a little under 8 Mb/s up/down when plugged in to the jack beside my desk.
>>> In the past several days, however, my in-room speeds have jumped
>>> dramatically (see http://www.speedtest.net/result/457545991.png).  Given the
>>> lack of any news about network infrastructure/speed upgrades, I'm curious
>>> whether this is just the consequence of some sort of maintenance upgrade for
>>> Thayer or if faster Ethernet has landed everywhere.  Any reports?
>>>
>>> Zach
>>>
>>> _______________________________________________
>>> hcs-discuss mailing list
>>> hcs-discuss at lists.hcs.harvard.edu
>>> http://lists.hcs.harvard.edu/mailman/listinfo/hcs-discuss
>>>
>>>
>> _______________________________________________
>> hcs-discuss mailing list
>> hcs-discuss at lists.hcs.harvard.edu
>> http://lists.hcs.harvard.edu/mailman/listinfo/hcs-discuss
>>
>


More information about the hcs-discuss mailing list