Want to see your switch or router vendors squirm? Ask how fast the “time-to-market” version of their 10G Ethernet interfaces will run.
Some vendors simply won’t know, because they haven’t yet done performance testing. Others will hem and haw because they know their interfaces go nowhere near line rate. A few honest souls may admit their first “10 Gigabit” products run closer to 3 or 4 gigabits/sec.
So where’s the disconnect? Why does the marketing collateral scream about 10-Gigabit performance when early versions deliver only a fraction of that rate? And why should network professionals pay for a given amount of performance or any other metric, and receive less in return?
The answer lies in the curious phrase “time to market.” All too often, this is a euphemism for shipping products that have no business being outside vendors’ engineering labs. These products ship because of vendors’ fears that if they don’t ship something – anything – their competitors will beat them to it.
I don’t mean to single out the 10G Ethernet crowd; there are numerous other instances where vendors have gotten it fast rather than right. Consider these examples:
– Enterasys Networks Inc. sells an excellent switch that runs at 10-Gigabit rates. The only problem is that the 10-Gigabit interfaces are proprietary. In its haste to bring product to market, Enterasys has forsaken interoperability. In fairness, Enterasys does have standards-based 10-Gigabit Ethernet products in the pipeline.
– Cisco Systems Inc.’s time-to-market version of an OC-48 line card for its flagship GSR router operated at a top speed of between 30 per cent and 40 per cent of line rate. This was in 1999 and Cisco long ago corrected the issue, but the shortcoming helped opened doors for…
– …arch-rival Juniper Networks Inc., which shipped a time-to-market OC-192 line card in spring 2000. The card had four internal paths, which caused it to reorder packets under heavy loads. Whether reordering is significant has been the subject of a lot of debate, but it’s a topic Juniper could have avoided altogether. Juniper recently shipped a newer OC-192 card that doesn’t reorder packets.
Vendors could avoid these problems by shipping products that do what they’re supposed to. Alas, it’s not realistic to expect good behaviour when there’s money on the table. Instead, here are three simple steps you can take to help protect your interests:
Get what you pay for. Buying a widget that supposedly offers 10 “dooflingies” of capacity? Ask the vendor to prove you get all 10.
Test, test, test. Don’t take a vendor’s claims at face value. Vendor-commissioned “independent” tests always seem to prove that vendor’s widget works best. Run your own evaluations or look to truly independent benchmarks.
Get a road map. If a vendor with a time-to-market product promises a better product “real soon now,” get a promise of when that product will be available and defer the purchase until then.
It’s your money vendors compete for. Whatever the claims on vendors’ data sheets, don’t settle for anything less than what you need for your network.
Newman is president of Network Test in Westlake Village, Calif., an independent benchmarking and network design consultancy. He can be reached at email@example.com.