4.1 Overall

During the time we tested the NetApp, it had about 500-600 users. Though not as much as we'd hoped for, this still gave the NetApp enough work to do. We set up a 'peer' link to HSIJ, and a 'parent' link to the SURFNet webcache. This way, we could test any configuration that might occur within SURFNet. In this setup, the NetApp was on average capable of handling about 75 urls/second. Statistics can be found in Appendix B.

4.2 End user experience

We set up an emailadres (proxymaster@utwente.nl) for questions and bugreports. About 25 people mailed to that adres, most of them with questions on how to use the proxy. One person complained he couldn't reach some documentation sites anymore. It turned out that those were sites on which the libraries of the university had contracts with for viewing documentation. Those sites were restricted on IP basis, and because we put the C230 in a cachemesh with one neighbor (HSIJ) and one parent (SURFnet) our parent fetched the pages for us, when we didn't have them. In case of such a site, the IP address that site would see was the IP address of the SURFnet cache, which wasn't allowed access, so the user got a Access Denied message. Configuring SURFnet as a sibling solved the problem very quickly. It turned out however that site restriction wasn't that easy with NetApp.

4.3 Useability of the C230 within SURFnet cachemesh

The NetApp C230 was run standalone for the first 5 days, after which we made a parent relation with the central Dutch toplevel cache of SURFnet and a neighbor relation with a college in Deventer (HSIJ).

The links worked very well, except for a small problem with our parent, which was mentioned above.

In the month May we fetched 40 Gb of traffic through the NetApp of which 11 Gb was due to a stresstest on the C230. Even SURFnet, from which we got the central cache index of their proxy to suck empty, noticed the result in their reports over the month May.

Of all traffic we fetched, we fetched 58.88% of our traffic by parent SURFnet, 8.33% as sibling SURFnet and 0.56% as sibling from HSIJ (all mentioned in hits). In Kbytes these numbers are respectively 68.48%, 5.67% and 0.46%

See also Appendix B with statistics over the month May.

It also seems (from a glance at the stats) that the netapp generates a little more errors than a Squid proxy does at the same amount of time.

The NetApp C230 seems to perform well for small to medium sized organisations. However, for a University like Twente, the C630 would be more approriate. This had been concluded after a conversation with Peter Danzig from NetApp.