Presentation on theme: "Top 10 Ways to Optimize Network Performance Carrie Higbie, Siemon Global Network Applications Market Manager Ask the Expert, TechTarget – SearchNetworking,"— Presentation transcript:
Top 10 Ways to Optimize Network Performance Carrie Higbie, Siemon Global Network Applications Market Manager Ask the Expert, TechTarget – SearchNetworking, SearchEnterprise Voice, SearchDataCenter President, BladeSystems Alliance
Top 10 ways to optimize performance Understand your performance What is harming optimal performance? Physical layer Routing and switching Applications Environmental concerns How to fix performance What does poor performance cost?
1. Bandwidth bandits – Find ‘em, slay ‘em! Begin with auto-discovery Start at the physical layer Common to all systems Can cause the most problems According to LAN Magazine – 70% of all downtime! Can be 80% of all troubleshooting costs Only 20% of troubleshooting time is spent fixing the problem Documentation of the physical layer may be required for compliance Absolute necessity for security reasons
Top offenders Improperly terminated cables Improperly terminated patch cords Lengths exceeded specified maximum Cabling was improperly or not labeled (troubleshooting problems) Electronics and closets in bad locations (humidity, EM, RF) Cables bunched to tightly causing the pairs to be flattened Cables tied to electrical conduits or run too close to power panels Cabling that did not pass testing due to various issues Closet spaghetti CAT 3 cables terminated to CAT 5 – 100M switches Bends exceeding bend radius Cables not to spec Racks not grounded
Active electronics and networking layers Flat networks with large address ranges Subnets are helpful Poor VLAN management Intruders and security breaches Build a historic reference Duplex mismatches Ports forced to low speeds Half-duplex connections May be electronics; may be cabling!
2. Building in resiliency Load balancing and load sharing will help performance Proper subnets and/or VLANs can help If you build a monster, it will be a monster! Use the electronics to your advantage Price/performance ratios are key! Interoperability must be tested, not just stated Failover must be tested
3. Pay attention to applications Application sizes double every 18 months More capability and application bloat Process increasingly larger file sizes Shared applications and server-side processing can enhance or deplete resources QoS attributes can be set in applications Don’t let them step on each other Increased demands on storage, backups and network resources
With growth comes problems According to Nucleus Research (2004) Average storage budget = $2.3 Million Average storage capacity = 115 TB Troubleshooting = Average 14 hours per month Among companies with budgets > $1B 85% report service degradation 51% said poor application performance 82% said problems impact employee productivity 79% said customer service quality suffers! Study from Network World (2003) According to Gartner, downtime will increase 200% in Q % of all IT expenditures are for things that DON’T work
Sample views - SNMP
NodeIndexMtuSpeedIn OctetsIn Errors ,609,426,1926,975, ,116,050,9175,761, ,841,840,3121,507, ,369,4762,302, ,099,071,3714,227, ,274,947,5023,155, ,490,061,1695,247, ,847,4343,437, ,306,4898,258, ,130,881,1547,166, ,190,70511,145, ,094,5031,315, ,346,0036,514, ,077,239,3331,913, ,455,568,62814,413, ,461,3641,789, ,564,4443,115, ,670,1141,112, ,618,6941,109,967 Actual audit data Note: Proprietary And confidential information removed
4. Examine utilization Snapshots do little good Must be done over time Sample period should include all normal business functions Closing out accounting End-of-month processes Payroll runs High-traffic periods such as high customer service demands Forget averages – watch peak periods!
What is utilization really? What do I look for? Make sure that periods viewed are consistent with hours worked Averages do you little good For real-time applications always plan on highest utilization numbers Includes VoIP/IPT, video, etc. Group utilization needs by class of user, not department
Environmental conditions Temperature and humidity variations EM and RF interference High network traffic Outdated, slow PCs and NICs Poor installation Inferior patch cords Damaged cable due to pulling, bending Too many splices Poor cable management Inferior network cabling ACCORDING TO ESTIMATES GIVEN TO IEEE, OVER 50% OF ALL CAT 5E WON’T PASS 5E TESTING! What causes slow response
The cost of a slow network Examples: Company A: Number of Employees: 500 Average Hourly Wage: $15.00 Hours of Productivity Lost per Year: 30 Network Slow Cost = $225, Company B: Number of Employees: 1,000 Average Hourly Wage: $18.00 Hours of Productivity Lost per Year: 52 Network Slow Cost = $936, Company C: Number of Employees: 5,000 Average Hourly Wage: $20.00 Hours of Productivity Lost per Year: 20 Network Slow Cost = $2,000, Calculate network slow cost: Cost = P x W x E P = Total Number of hours lost Productivity per year (weekly minutes/60 x 52) W = Average hourly Wage E = Number of Employees on the network
Formulas Revenue per hour Total revenue / 2080 hour work week Revenue per employee per hour Total revenue / Number of employees / 2080 Salary expense per hour (weighted) Average hourly wage * 1.4 (to include overhead) / 2080 Salary expense plus lost revenue Total revenue per hour + weighted salary expense per hour * % of workforce down at any given time (we used 15%)
5. Performance optimization: Tricks of the trade Audit your infrastructure – cabling, electronics, etc. Use a Certified Infrastructure Auditor Trained to understand relationships between electronics and physical layer Omitting either one is only half an audit Audit your ports, not the entire switch Use RMON to help determine bottlenecks
Adding new applications BEWARE – minimums are dangerous! Utilize a test bed to help understand needs Multiply all results by the number of end users Assume concurrent operations for all on the same shifts Don’t forget additional loads for replication, redundancy and backups There is a fine line between not enough and too much Aim for a little too much Don’t forget to account for growth!
Products to help Bandwidth managers Layer 7 products Application classification Routing based on need Dynamic bandwidth allocation Don’t stop at Layer 3 Courtesy of Packeteer
6. Trends: Know your bandwidth needs Courtesy of Packeteer
Determine your need for speed 10G products are currently available, with new copper-based products expected this year Your network may be a combination of speeds to the desktop If a user can’t fill an entire gig channel but you have discards at 100 Mbps, gig should be used Same applies to 10G Watch power users CAD/CAM/CAE Imaging and graphics Video
7. Predict the future – The crystal ball Examine your past 5 years if possible Application changes Speed changes Hardware upgrades Server upgrades Memory upgrades OS upgrades
Trends for ‘05 Gates’ Law revisited “640k ought to be enough for anybody” -- Bill Gates, 1981
Application growth/bloat Office 95Office XPWindows 95Windows XP 386 – 486 recommended PIII recommended 133MHz min 486 recommended 233MHz min 300MHz rec. 8 MB RAM 128 MB RAM + 8MB for each program 4 MB RAM, 8 recommended 128 MB RAM+ recommended 64MB min 40 MB min disk space 128 all features 245MB disk space + 115MB +50 MB MB1.5 GB Disk space
8. Know your security challenges #$%^&* hackers! Spyware and malware Breaches Reporting structure Weaknesses Logging Compliance
IT management and security management Shift in duties Other spending is often derailed in lieu of security expenditures Consumes many resources Patch management adds overhead WAN links can be hindered or halted Utilize the best tools ROI based on cost avoidance Beware of target size!
9. Revisit and revise Any and all measures mentioned Quarterly health checks a must Downtime is not the only factor Slow performance is also very costly Put in the best if you expect the best The level of support you can give is equal to the level of support you can receive
10. Evaluate products: What are the extras in a top-of-the-line system? R&D Standards participation Support on a global scale Tried-and-true support Product expediting Value-added services Warranty Partner as opposed to seller
Evaluate your vendors properly Check non-vendor references Pose some questions and validate the answers Training and transfer of knowledge Quality of training Certification levels Training for their resellers Back-end support – knowledge bases Pricing is only one factor and should never be the deciding factor MTBF/MTTR – every single component! Upgrade paths and future capabilities
Have a bake-off Invite your potential vendors in for testing If they state it will work – make them show you Verify fail-over/redundancy/resiliency Look at the management tools Play with the technology If you find a weakness in one product, see how/if it is addressed in another
MOST IMPORTANT! Understand what you are getting for your money Front end Back end In between Understand market share and marketing information Understand where the standards are going. Today’s investment could kill support tomorrow!
Thank you Carrie Higbie, Siemon Global Network Applications Market Manager Ask the Expert, TechTarget – SearchNetworking, SearchEnterprise Voice, SearchDataCenter President, BladeSystems Alliance