Speeds and Feeds are Boring - Until You Have Service Downtime
For most people speeds and feeds are boring.
You may be thinking that that’s a bold statement from someone at Juniper Networks – a company built upon high performance routing. So, why did I include High Performance in the four network essentials for the future (Automated, Open, Secure, and High Performance)?
Speeds and feeds may be boring for most people, but they are essential and must not be overlooked.
Today, all hardware can deliver almost unbelievable scale, and performance keeps getting better every year. For example when the T640 launched around 15 years, there would be stories in the industry press, such as this one in Light Reading about the launch of the first core router from Juniper. But now journalists are less interested in raw performance than topics such as the interoperability of virtual functions.
Despite this, performance is still critical to all businesses. For most businesses, any downtime in a service is lost revenue or reputation.
But, high performance is not just about the raw throughput of a platform. It also needs to be delivered with the appropriate cost base. That’s why Juniper continues to invest in world-leading networking silicon. Recent examples include the Q5 ASIC in the QFX10k series of switches. Merchant silicon is great for many applications, but for networks requiring the best price/performance ratio, custom silicon has continued to show leadership.
Other dimensions to performance that need to be considered include: reliability, availability and serviceability. Networks need to embrace technologies such as in-service software upgrades (ISSU), hot swapping of hardware and hitless failover. There will be hardware failures in your network – you will need to design around them to minimise their impact.
So, as you architect the network for your business, you need to ensure that it is able to scale as the business grows in a simple way that is also cost-effective.
Borrowing a theme from my previous blogs in this series, I would define a network with score of 10 out of 10 for being high performance as one where:
All parts of the network scale to 10x today’s capacity without major forklift upgrades
The network is designed with reliability, availability and serviceability so that any single site or node being taken out of service for upgrades, even at peak times, without impacting end user services
This doesn’t mean that you run a network at 10% utilisation, but it does mean that you pick platforms which have the ability to scale significantly. That could be via new top-of-rack switches or additional line cards. But you don’t want to change the whole platform you’ve built the network on.
A 10x growth in capacity may seem like a lot, but with traffic often growing at 50% a year you’re going to hit this limit within just over 5 years. That’s without the impact of any new services or unexpected spikes in demand (e.g. a new iPhone launch). Despite the typical 5-year depreciation cycle, most hardware stays in a network for a lot longer than 5 years.
As I finish this blog series; I have a few questions for you:
How would you score your own network on a score of 1 to 10 for high performance?
Where would you like to be in 2-3 years’ time?
What are the immediate next steps you would like to take to close your performance gap?
If you want to learn more about what Juniper is delivering for high performance networks check out the links below: