Cache is King

Source: http://www.stevesouders.com/blog/2012/10/11/cache-is-king/

It’s important to understand where the benefits from each technique come into play. Using Ajax, for example, doesn’t make the initial page load time much faster (and often makes it slower if you’re not careful), but subsequent “pages” (user actions) are snappier. Optimizing JavaScript, on the other hand, makes both the first page view and subsequent pages faster. Better caching sits in the middle: The very first visit to a site isn’t faster, but subsequent page views are faster. Also, even after closing their browser the user gets a faster initial page when she returns to your site – so the performance benefit transcends browser sessions.

These web performance optimizations aren’t mutually exclusive – you should do them all! But I (and perhaps you) wonder which has the biggest impact. So I decided to run a test to measure these different factors. I wanted to see the benefits on real websites, so that pointed me to WebPagetest where I could easily do several tests across the Alexa Top 1000 websites. Since there’s no setting to “Ajaxify” a website, I decided instead to focus on time spent on the network. I settled on doing these four tests:

• Baseline – Run the Alexa Top 1000 through WebPagetest using IE9 and a simulated DSL connection speed (1.5 Mbps down, 384 Kbps up, 50ms RTT). Each URL is loaded three times and the median (based on page load time) is the final result. We only look at the “first view” (empty cache) page load.
• Fast Network – Same as Baseline except use a simulated FIOS connection: 20 Mbps down, 5 Mbps up, 4ms RTT.
• No JavaScript – Same as Baseline except use the new “noscript” option to the RESTful API (thanksPat!). This is the same as choosing the browser option to disable JavaScript. This isn’t a perfect substitute for “optimizing JavaScript” because any HTTP requests that are generated from JavaScript are skipped. On the other hand, any resources inside NOSCRIPT tags are added. We’ll compare the number of HTTP requests later.
• Primed Cache – Same as Baseline except only look at “repeat view”. This test looks at the best case scenario for the benefits of caching given the caching headers in place today. Since not everything is cacheable, some network traffic still occurs.

Which test is going to produce the fastest page load times? Stop for a minute and write down your guess. I thought about it before I started and it turned out I was wrong.

The Results

This chart shows the median and 95th percentile window.onload time for each test. The Baseline median onload time is 7.65 seconds (95th is 24.88). Each of the optimizations make the pages load significantly faster. Here’s how they compare:

1. Primed Cache is the fastest test at 3.46 seconds (95th is 12.00).
2. Fast Network is second fastest at 4.13 seconds (95th is 13.28).
3. No JavaScript comes in third at 4.74 seconds (95th is 15.76).

I was surprised that No JavaScript wasn’t the fastest. Disabling JavaScript removes the blocking behavior that can occur when downloading resources. Even though scripts in IE9 download in parallel with most other resources (see Browserscope) they cause blocking for font files and other edge cases. More importantly, disabling JavaScript reduces the number of requests from the Baseline of 90 requests down to 59, and total transfer size drops from 927 kB to 577 kB.

The improvement from a Fast Network was more than I expected. The number of requests and total transfer size were the same as the Baseline, but the median onload time was 46% faster, from 7.65 seconds down to 4.13 seconds. This shows that network conditions (connection speed and latency) have a significant impact. (No wonder it’s hard to create a fast mobile experience.)

The key to why Primed Cache won is the drop in requests from 90 to 32. There were 58 requests that were read from cache without generating any HTTP traffic. The HTTP Archive max-age chart for the Alexa Top 1000 shows that 59% of requests are configured to be cached (max-age > 0). Many of these have short cache times of 10 minutes or less, but since the “repeat view” is performed immediately these are read from cache – so it truly is a best case scenario. 59% of 90 is 53. The other 5 requests were likely pulled from cache because of heuristic caching.

Although the results were unexpected, I’m pleased that caching turns out to be such a significant (perhaps the most significant) factor in making websites faster. I recently announced my focus on caching. This is an important start – raising awareness about the opportunity that lies right in front of us.

The number of resource requests was reduced by 64% using today’s caching headers, but only because “repeat view” runs immediately. If we had waited one day, 19% of the requests would have been expired generating 17 new If-Modified-Since requests. And it’s likely that some of the 5 requests that benefited from heuristic caching would have generated new requests. Instead we need to move in the other direction makingmore resources cacheable for longer periods of time. And given the benefits of loading resources quickly we need to investigate ways to prefetch resources before the browser even asks for them.

Change is Good

This will be a short post. Today is my last day in the company I was working for last one year and It's a bit sad leaving some great guys behind but at the same time I am really excited about the new opportunity I have.

Understanding Third-Party Next-Hop

source:  Petr Lapukhov

Abstract

This publication briefly covers the use of 3rd party next-hops in OSPF, RIP, EIGRP and BGP routing protocols. Common concepts are introduced and protocol-specific implementations are discussed. Basic understanding of the routing protocol function is required before reading this blog post.

Overview

Third-party next-hop concept appears only to distance vector protocol, or in the parts of the link-state protocols that exhibit distance-vector behavior. The idea is that a distance-vector update carries explicit next-hop value, which is used by receiving side, as opposed to the “implicit” next-hop calculated as the sending router’s address – the source address in the IP header carrying the routing update. Such “explicit” next-hop is called “third-party” next-hop IP address, allowing for pointing to a different next-hop, other than advertising router. Intitively, this is only possible if the advertising and receiving router are on a shared segment, but the “shared segment” concept could be generalized and abstracted. Every popular distance-vector protocols support third party next-hop – RIPv2, EIGRP, OSPF and BGP all carry explicit next-hop value. Look at the figure below – it illustrates the situation where two different distance-vector protocols are running on the shared segment, but none of them runs on all routers attached to the segment. The protocols “overlap” at a “pivotal” router and redistribution is used to provide inter-protocol route exchange.

Per the default distance-vector protocol behavior, traffic from one routing domain going into another has cross the “pivotal” router, the router where the two domains overlap (R3 in our case) – as opposed to going directly to the closes next-hop on the shared segment. The reason for this is that there is no direct “native” update exchange between the hops running different routing protocols. In situations like this, it is beneficial to rewrite the next-hop IP address to point toward the “optimum” exit point, using the “pivotal” router’s knowledge of both routing protocols.

OSPF is somewhat special with respect to the 3rd party next-hop implementation. It supports third-party next-hop in Type-5/7 LSAs (External Routing Information LSA and NSSA External LSA). These LSAs are processed in “distance-vector manner” by every receiving router. By default, the LSA is assumed to advertise the external prefix “connected” to the advertising router. However, if the FA is non-zero, the address in this field is used to calculate the forwarding information, as opposed to default forwarding toward the advertising router. Forwarding Address is always present in Type-7 LSAs, for the reason illustrated on the figure below:

Since there could be multiple ABRs in NSSA area, only one is elected to perform 7-to-5 LSA translation – otherwise the routing information will loop back in the area, unless manual filtering implemented in the ABRs (which is prone to errors). Translating ABR is elected based on the highest Router-ID, and may not be on the optimum path toward the advertising ASBR. Therefore, the forwarding address should prompt the more optimum path, based on the inter-area routing information.

EIGRP

We start with the scenario where we redistribute RIP into EIGRP.

Notice that EIGRP will not insert the third-party next-hop until you apply the command no ip next-hop-self eigrpon R3′s connection to the shared segment. Look at the routing table output prior to applying the no ip next-hop-self eigrp command.

R1#show  ip route eigrp 
     140.1.0.0/16 is variably subnetted, 2 subnets, 2 masks
D EX    140.1.2.2/32
           [170/2560002816] via 140.1.123.3, 00:00:27, FastEthernet0/0

After the command has been applied to R3’s interface:

R1#show  ip route eigrp
     140.1.0.0/16 is variably subnetted, 2 subnets, 2 masks
D EX    140.1.2.2/32
           [170/2560002816] via 140.1.123.2, 00:00:04, FastEthernet0/0

The same behavior is observed when redistributing OSPF into EIGRP, but not when redistributing BGP. For some reason, BGP’s next-hop is not copied into EIGRP, e.g. in the example below, EIGRP will NOT insert the BGP’s next-hop into updates. Notice that you may enable or disable the third-party next-hop behavior in EIGRP using the interface-level command ip next-hop-self eigrp.

RIP

RIP passes the third-party next-hop from OSPF, BGP or EIGRP. For instance, assume EIGRP redistribution into RIP. You have to turn on the no ip split-horizon on R3′s Ethernet connection to get this to work:

R2#show ip route rip 
     140.1.0.0/16 is variably subnetted, 3 subnets, 2 masks
R       140.1.1.1/32 [120/1] via 140.1.123.1, 00:00:17, FastEthernet0/0

Notice the following RIP debugging output, which lists the third-party next-hop:

RIP: received v2 update from 140.1.123.3 on FastEthernet0/0
     140.1.1.1/32 via 140.1.123.1 in 1 hops
     140.1.123.0/24 via 0.0.0.0 in 1 hops

Surprisingly, there is NO need to enable the command no ip split-horizon on the interface when redistributing BGP or OSPF routes into RIP. Seem like only EIGRP to RIP redistribution requires that. Keep in mind, however, that split-horizon is OFF by default on physical frame-relay interfaces. Here is a sample output of redistributing BGP into RIP using the third-party next-hop:

R3#show ip route bgp 
     140.1.0.0/16 is variably subnetted, 3 subnets, 2 masks
B       140.1.2.2/32 [20/0] via 140.1.123.2, 00:22:13
R3#

R1#show ip route rip 
     140.1.0.0/16 is variably subnetted, 3 subnets, 2 masks
R       140.1.2.2/32 [120/1] via 140.1.123.2, 00:00:09, FastEthernet0/0

RIP’s third-party next-hop behavior is fully automatic. You cannot disable or enable it, like you do in EIGRP.

OSPF

Similarly to RIP, OSPF has no problems picking up the third-party next-hop from BGP, EIGRP or RIP. Here is how it would look like (guess which protocol is redistributed into OSPF, based solely on the commands output):

R1#sh ip route ospf 
     140.1.0.0/16 is variably subnetted, 3 subnets, 2 masks
O E2    140.1.2.2/32 [110/1] via 140.1.123.2, 00:34:59, FastEthernet0/0

R1#show ip ospf database external 

            OSPF Router with ID (140.1.1.1) (Process ID 1)

                Type-5 AS External Link States

  Routing Bit Set on this LSA
  LS age: 131
  Options: (No TOS-capability, DC)
  LS Type: AS External Link
  Link State ID: 140.1.2.2 (External Network Number )
  Advertising Router: 140.1.123.3
  LS Seq Number: 80000002
  Checksum: 0xF749
  Length: 36
  Network Mask: /32
        Metric Type: 2 (Larger than any link state path)
        TOS: 0
        Metric: 1
        Forward Address: 140.1.123.2
        External Route Tag: 200

If you’re still guessing, the external protocol is BGP, as could have been seen observing the automatic External Route Tag – OSPF set’s it to the last AS# found in the AS_PATH.

There are special conditions to be met by OSPF for the FA address to be used. First, the interface where the third party next-hop resides should be advertised into OSPF using the network command. Secondly, this interface should not be passive in OSPF and should not have network type point-to-point or point-to-multipoint. Violating any of these conditions will stop OSPF from using the FA in type-5 LSA created for external routes. Violating any of these conditions prevents third-party next-hop installation in the external LSAs.

OSPF is special in one other respect. Distance vector-protocols such as RIP or EIGRP modify the next-hop as soon as they pass the routing information to other devices. That is, the third party next-hop is not maintained through the RIP or EIGRP domain. Contrary to these, OSPF LSAs are flooded within their scope with the FA unmodified. This creates interesting problem: if the FA address is not reachable in the receiving router’s routing table, the external information found in type 7/5 LSA is not used. This situation is discussed in the blog post “OSPF Filtering using FA Address”.

BGP

When you redistribute any protocol into BGP, the system correctly sets the third-party next-hop in the local BGP table. Look at the diagram below, where EIGRP prefixes are being redistributed into BGP AS 300:

R3’s BGP process installs R1 Loopback0 prefix into the BGP table with the next-hop value of R1’s address, not “0.0.0.0” like it would be for locally advertised routes. You will observe the same behavior if you inject EIGRP prefixes into BGP using the network command.

R3#sh ip bgp
BGP table version is 9, local router ID is 140.1.123.3
Status codes: s suppressed, d damped, h history, * valid, > best, i - internal,
              r RIB-failure, S Stale
Origin codes: i - IGP, e - EGP, ? - incomplete

   Network          Next Hop            Metric LocPrf Weight Path
*> 140.1.1.1/32     140.1.123.1         156160         32768 ?

Furthermore, BGP is supposed to change the next-hop to self when advertising prefixes over eBGP peering sessions. However, when all peers share the same segment, the prefixes re-advertised over the shared segment do not have their next-hop changed. See the diagram below:

Here R1 advertises prefix 140.1.1.1/24 to R3 and R3 re-advertises it back to R2 over the same segment. Unless non-physical interfaces are used to form the BGP sessions (e.g. Loopbacks), the next-hop received from R1 is not changed when passing it down to R2. This implements the default third-party next-hop preservation over eBGP sessions. Look at the sample output for the configuration illustrated above: R1 receives R2’s prefix with unmodified next-hop.

R1#show ip bgp 
BGP table version is 3, local router ID is 140.1.1.1
Status codes: s suppressed, d damped, h history, * valid, > best, i - internal,
              r RIB-failure, S Stale
Origin codes: i - IGP, e - EGP, ? - incomplete

   Network          Next Hop            Metric LocPrf Weight Path
*> 140.1.1.1/32     0.0.0.0                  0         32768 i
*> 140.1.2.2/32     140.1.123.2                            0 300 200 i

There is a way to disable this default behavior in BGP. A logical assumption would be that using the commandneighbor X.X.X.X next-hop-self would work, and it does indeed, in the recent IOS versions. The older IOS, such as 12.2T did not have this command working for eBGP sessions, and your option would have been using a route-map with set ip next-hop command. The route-map method may still be handy, if you want insert totally “bogus” IP next-hop from the shared segment – receiving BGP speaker will accept any IP address that is on the same segment. That is not something you would do in the production environment too often, but definitely an interesting idea for lab practicing. One good use in production is changing the BGP next-hop to an HSRP virtual IP address, to provide physical BGP speaker redundancy. Here is a sample code for setting an explicit next-hop in BGP update:

router bgp 300
 neighbor 140.1.123.1 remote-as 100
 neighbor 140.1.123.1 route-map BGP_NEXT_HOP out
!
route-map BGP_NEXT_HOP permit 10
 set ip next-hop 140.1.123.100

Summary

All popular distance-vector protocols support third-party next-hop insertion. This mechanism is useful on multi-access segments, in situations when you want pass optimum path information between routers belonging to different routing protocols. We illustrated that RIP implements this function automatically, and does not allow any tuning. On the other hand, EIGRP supports third-party next-hop passing from any protocol, other than BGP, and you may turn this function on/off on per-interface basis. Furthermore, OSPF’s special feature is propagation of the third-party next-hop within an area/autonomous system, unlike the distance-vector protocols that reset the next-hop at every hop (considering AS a being a “single-hop” for BGP). Thanks to that feature, OSPF offers interesting possibility to filter external routing information by blocking FA prefix from the routing tables. Finally, BGP gives most flexibility when it comes to the IP next-hop manipulation, allowing for changing it to any value.