[IPOL discuss] a development server in Uruguay? your opinion matters

[Nicolas Limare]

Hi all,

We want to install a new development server for IPOL; we can't host it
(plug it) in Cachan, we have an option for hosting the machine in
Universidad de la Republica, Montevideo, Uruguay.

It's good for the server stability, it's good for the uruguayan
involvement in the project and generally speaking international
collaborations, it's bad for the network connexion.

I think the upload/download will be slower from european academic
networks, but similar to the speed we get from a home DSL
connexion. It should not be a problem for external use: testing the
demos still in development on this machine, uploading/downloading some
code, ...

I think the latency will be much slower. This will only affect
real-time work (SSH) on the server from europe, but it will be
noticeable and uncomfortable, at least at the beginning.

What do you think of this option? Is it worth putting a new big
machine there, with the network effects described before?

Best,

8<----------8<----------8<----------8<----------8<----------8<----------

All the details:

We are thinking about renewing the development server. Public IPOL
demos are now running on out latest server, a Dell R910 with 4 8-core
Intel Xeon 7560 2.27GHz processors and 32GB RAM. Before they become
public and stable, the demos can be developed and tested on our
previous machine, a Dell with 2 4-core Intel Xeon E5345 2.33GHz CPUs
and 16GB RAM.

We want to test and run the demos in the same environment; this is
especially important for parallel codes, which would typically roughly
4 times faster on the recent machine. So we intend to buy a new
machine but then comes a crucial question: where to host it?

We don't feel confident with hosting in Cachan. The central DSI
serveur room has a failing cooling system and the CMLA server room had
no cooling for a few months last year. We want better reliability for
IPOL. Looking for external solutions, we found a possibility in
Universidad de la Republica, Uruguay.

The advantages of this solution are:
* we would be hosted in the main university server room, and we expect
  a good technical environment there;
* this machine would be close to collaborators from this university,
  where one person will soon be hired to work on IPOL and related
  projects.

The inconvenient is the network location. On the network map, Uruguay
is far from Europe, and we would all feel the distance:
* bandwidth: data seems to be transferred at 300kB/s between Paris
  and Montevideo
* latency: a packet of information needs 300ms to travel from Paris to
  Montevideo

The bandwith limit is involved when we upload and download large
files: upload and download big source code archives, upload some
images to a demo, receive (thus, download) the result page of a demo.
It will be noticeable from european university networks, which get
about 10MB/s (30x faster) connexion with IPOL (it depends on a lot of
factors, including the local network quality on both sides). But the
speed will be similar to the one you get from a typical home DSL line;
I did some tests from a home connexion, and downloaded at 800kB/s from
dev.ipol.im and 600kB/s from iie.fing.edu.uy; there should be even
less differences when uploading from a DSL line because the upload
speed is limited. Uploading a 2MB file (a full-size picture from a
digital camera) would take 8s à 300kB/s.

The latency will not impact transfers and web usage of this server,
but it will be noticeable when people located in europe interactively
work on the server located in Uruguay, usually via SSH: a letter will
only appear in your console half a second after you type it, it breaks
the illusion of simultaneity. I have this latency every day when I
work from Japan on machines hosted in Cachan and we can get used to
it, but it is noticeable and uncomfortable at the beginning,
especially when you only work from Paris labs on machines hosted in
other Paris labs, with less than 2 ms latency.

I started to collect some network statistics from our machines:
-> http://tools.ipol.im/munin/ipol/comparison-day.html#Network

The "Download Speed" graphs shows the bandwith between these machines
and various web servers. The same measures from our working places
would be useful. These numbers may alse be strongly affected by
various network settings on the Montevideo side and the server load.

The second line of "Ping times" graphs shows the latency between these
machines and machines in other locations: Universidad de la Republica
(Montevideo), ENS Cachan, CMLA, ENST, Paris5, ENPC, UIB (Palma de
Mallorca), and Plala (my ISP in Japan).

For you information, "fuchsia" is on the ENS network, "green" is on
the CMLA subnetwork, "black", "blue" and "gray" are on an external
non-academic network in Paris. Interestingly, these external machines
have better bandwidth and latency than the ones hosted in Cachan...

-- 
Nicolas LIMARE - CMLA - ENS Cachan    http://www.cmla.ens-cachan.fr/~limare/
IPOL - image processing on line                          http://www.ipol.im/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://tools.ipol.im/mailman/archive/discuss/attachments/20110516/631d69a8/attachment.pgp>