Cisco 871 maximum line speed

[chad patterson]

     I'm tryng to determine the maximum speed of some of the routers I have, and of some that I am considering buying. I have reviewed the Portable Product Sheets – Routing Performance many times, and i guess I cannot interperet it very well. I also reviewed this post about a Cisco 871 router speed maxing out at around 12 mbps.

     I maintain a Cisco 871 at my office, and I consistently get around 30Mbps down and 8Mbps up, but that is due to my cable companies speed limitiations applied upon me. However i did plug a Cisco 1721 into the cable connection, and maxed out at around 6Mbps, which is quite consistent with the document. This document however, is dishormonious with all other documents.

     Now all of these documents agree that we are using the Mbps notation as compared to MBps, so the discontinuity does not lie in the notation. What then could be the cause of the discord in the results? are we reading this document wrong? Even if the speeds in the document were stated at half duplex (as I have seen suggested elsewhere), I would then have to max out at around 25Mbps, but I am getting around 30Mbps as I stated earlier.

     Would some kind, enlightened soul help me to understand how to determine the maximum WAN speed that can be acheived with Cisco routers?

[Joseph W. Doherty · Accepted solution]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Perhaps the most important thing to understand, required packets per second decreases, for the same rate, as frame/packet increases in size.  Performance numbers (today) are generally provided for minimum size frame/packet.  So when you see the PPS for a 871 is noted at 25 Kpps, and knowing minimum size Ethernet uses 1.488 Kpps per 1 Mbps, 25 Kpps supports about 17 Mbps.  However, maximum size Ethernet only needs 81 PPS per 1 Mbps, so the same 25 Kpps would support about 309 Mbps.  Most routers, though, do not offer the same maximum PPS rate for larger packets, so an 871's worst case (not counting CPU for non-forwarding) should be 17 Mbps, but could range all the way up to 309 Mbps (unlikely it's anywhere nearly that high).

[chad patterson]

     I'm sorry, but it seems that the answer, summed up, is "anywhere between 17Mbps and 309Mbps". Those two values are not in same ballpark, they cannot even belong to the same router. How can I figure the absolute 'realistic', maximum routing speed for a WAN interface with a 10/100 Ethernet handoff, and no ACL's or encryption to hinder the possible speed?

     If you look at this table from Princeton University (ftp://ftp.cs.princeton.edu/techreports/2002/645.pdf)::

                             | Minimum-size Packets   |  Maximum-size Packets

                             -------------------------------------------------------------

                             | Packet Rate | Data Rate|Packet Rate| Data Rate

-------------------------------------------------------------------------------------

10 Mb/s Ethernet   |  14.88 Kpps | 7.62 Mb/s | 0.813 Kpps| 9.87 Mb/s

100 Mb/s Ethernet |  148.8 Kpps | 76.2 Mb/s | 8.13 Kpps  | 98.7 Mb/s

1 Gb/s Ethernet     |  1488. Kpps | 762. Mb/s | 81.3 Kpps  |987. Mb/s

     Then compare this to the data from the table in the previous document, which states that the Cisco 871 gets 25 Kpps for minimum size packets. It seems that the 871 should be able to route at about 17% of 100 Mb/s Ethernet: which is a minimum of 12Mbps, and a maximum of 17Mbps, roughly.

     This actually 'sounds' a lot more realistic than "anywhere between 17Mbps and 309Mbps. But, again, it is not what is happening on my router. So why am I getting almost 30Mbps on my ethernet WAN interface?

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Laugh - what I quoted is correct, even if it seems unrealistic.  Believe you're misreading the Princeton reference you've provided, which is correctly describing data throughput rates for different packet sizes.

However, using the table you provided . . .

If 100 Mbps, for minimum size packets requires a rate of 148.8 Kpps, and 870's max rate is 25 Kpps, what's our maximum capacity in Mbps?  It would be 25 / 148.8 = 16.8% of 100 Mbps or about 17 Mbps as I originally noted.

Likewise, if 1 Gbps, for minimum size packets requires a rate of 81.3 Kpps, and  our max rate is 25 Kpps, what's our maximum capacity in Mbps?  It would  be 25 / 81.3 = 30.7% of 1 Gbps or about 307 Mbps (note this is a little less than my original posted about 309 Mbps, but I had used 81 Kpps not 81.3 Kpps.

So, if you can follow the above calculations, possible performance for the 871 ranges from a low of about 17 Mbps to a high of 307 Mbps.

Now I had also noted most routers do not offer the same maximum PPS rate for maximum sized packets.  I.e. it's (very) unlikely the 871 offers 25 Kpps for maximum size packets as it does for minimum size packets.

So, what is the top rate of the 871?  Well either we need to test ourselves or we need to find some documentation for its performance at the maximum packet size.  (Ideally, other incremental packet size PPS rates are documented too.)

I have noticed one of the later Cisco performance documents (attached - see table 1) note the maximum bandwidth throughput of an 860 series router is 197 Mbps.  As this model is also rated, for minimum size packets, at 25 Kpps, like the 870 series, the 870's maximum throughput is probably nearly the same, i.e. about 197 Mbps.  So, now we have a performance range of 17 Mbps to 197 Mbps.

(BTW, showing how different models can vary much in performance, in table one, compare the maximum throughput of the 860, 880 and 890 compared to their minimum PPS rates.  Notice how the ratios don't hold between them.)

In answer to you question about finding 'realistic' maximum performance, I just answered that, but a corollary question is general realistic performance.  Here too the attached document can help as it provides performance number for IMIX traffic, see table 8.

Unfortunately again the 870 isn't documented, nor is the performance of the 860, but the (50 Kpps) 880 is documented at 30 Mbps and the (100 Kpps)  890 is documented at 45 Mbps.  So assuming the (25 Kpps) 870 is half the 880 it would be about 15 Mbps or a quarter of the 890 it would be about 11 Mbps, I would estimate it ranges somewhere between 11 to 20 Mbps.  All these numbers support your seeing performance up to about 30 Mbps.

[chad patterson]

     Well to be fair, 11 to 20 is not close to 30. What if your ISP promised you 30Mbps, and you were only getting 11Mbps to 20Mbps? You would probably drop them, or demand a refund.

     But I think I am beginning to understand your confusion. Looking at the Princeton table and using the 25Kpps statistic of the 870, I performed two calculations.

1.  25 / 148.8 = 16.8

2.  25 / 8.13   = 307.5

These match your first statement.

     But from the Portable Product Sheets – Routing Performance document I linked to in the original post, I can see that 25 Kpps only applies to 64-byte packets (small packet size), and the maximum speed the cisco 860 and 871 can get when processing all small packet sizes is 12.8 Mbps. I know that 12.8 Mbps != 17 Mbps. So your first calculation is incorrect. Also 25 Kpps only applies to small packet sizes, so your second calculation is incorrect, because the Kpps is different on packet sizes of 1500 bytes (large packets). Both of your calculations are incorrect.

     However I have learned from the document that you provided, in table six, that the PAT Performance with IMIX Traffic at 75-Percent CPU, for the Cisco 860 is 27 Mbps. The Cisco 860 and the 870 have the same Kpps with 64 bytes packets, so the difference with IMIX traffic will not vary greatly (I assume), and that explains why I am maxing out at about 27 Mbps with my Cisco 871 in a real world, PATed environment. I thank you for providing that document.

     So it turns out that the actual maximum speed of a Cisco 871, in a real world environment, is about 27 Mbps.

     I would still like to know how I can figure this out for any device I come across. Can anybody show me how to figure this out for a Cisco 1760 for instance?

[Joseph W. Doherty · Accepted solution]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Laugh! I'm not confused, although you appear to be (and I do not mean to seem argumentative when I write that). Your confusion is understandable the way performance specs are often presented.

You write "Well to be fair, 11 to 20 is not close to 30." No you're not being fair because, first you seem to totally overlook my estimation is based on insufficient performance documentation, as I noted, and second, you still don't realize how variable performance can be based on actual traffic mix and device configuration.

Do you know how well your traffic mix conforms to the IMIX standard? I don't. But as the performance documentation we do have appears to support maximum performance of close to 200 Mbps, the fact that ". . . I getting almost 30Mbps on my ethernet WAN interface." doesn't mean your observed 30 Mbps means my estimation is incorrect. (Note: nor does it mean my estimation is totally accurate, which I never claimed. However, I did document how and where my numbers came from.)

You've also now noted, you've found table 6, unlike my usage of using table 5 (which I didn't explicitly note), which does note performance for an 860 and might be more representative of your usage. This is a fine observation!  (In fact, driving in this morning, I was thinking of adding it after making my prior post.)  Again, though, it doesn't invalidate my estimation nor does it prove "So it turns out that the actual maximum speed of a Cisco 871, in a real world environment, is about 27 Mbps."

Basically, when working with router performance, your "mileage may vary". For example, take note that between tables 5 and 6 an 880's performance increases from 45 to 60 Mbps (33% increase), for an 890 performance increases from 64 to 75 Mbps (17% increase) but for a 1921 performance decreases from 110 to 80 Mbps (27% decrease). Think about how different these ratios are for the same traffic but with a different configuration!

How does an 870 truly differ from a 860? Do you know? I don't, although I noted, as you, it might be similar to an 860, but since it's not an 860 is one reason I provided an estimated practical range.

"So it turns out that the actual maximum speed of a Cisco 871, in a real world environment, is about 27 Mbps."  So you're still sure?

You've now also written "I know that 12.8 Mbps != 17 Mbps". Yup!  But the former is based on PPS computation based on the Portable Product Sheets – Routing Performance, while the latter is based on Ethernet. As most of the industry uses, as does the document reference I provided, RFC 2544 for Ethernet, my numbers are correct if that's what Cisco also intended in table 1. As it is a notable difference, it would be nice to know whether the 25 Kpps is truly computed based on what's defined in your Cisco reference or my Cisco reference. (Note: even my first post made it clear I was using 25 Kpps based on Ethernet, so my numbers are correct. So are Cisco's in your reference, for the same 25 Kpps. Basically the conflict is Cisco is providing 25 Kpps [for 860s] within both Cisco documents, which may be correct, or not.  Reason it might be incorrect, it could vary between both based on dealing with actual media traffic or moving packets across internal fabric/bus without L2.  Or, it could be 25 Kpps is correct for both, and just how "bandwidth" is measured, for minimum size packets, can vary.)

"So your first calculation is incorrect." Still think so?  You're really sure?

"Also 25 kpps only applies to small packet sizes, so your second calculation is incorrect, because the kpps is different on packet sizes of 1500 bytes (large packets)." Yes, 25 Kpps noted for minimum sized packets (whatever Cisco is really using), but it can apply to maximum size packets too, and on some devices it does (e.g. 6500 sup720 when in compact mode). You know it doesn't, for an 870, how? (Note: I also noted for this router, it was unlikely to be the same for maximum size packets. But it it was 25 Kpps, my calculation was correct.)

"Both of your calculations are incorrect." What appears to be incorrect is your understanding of what I wrote, but my writing always isn't clear to everyone.  When it's not clear, the fault is mine, but questions and/or discussion are welcome.

"I would still like to know how I can figure this out for any device I come across. Can anybody show me how to figure this out for a Cisco 1760 for instance?" I would have gladly tried to help further, but as you believe my information is incorrect, without even asking for additional explanation, so you might actually learn (as I might too), I don't want to waste your, or my, time further. However, perhaps someone else might still be able to further assist you.

[chad patterson]

Disclaimer

JosephDoherty is full of misinformation so it is no wonder that he feels the need to post a Liability Disclaimer as a preemptive form of deniability at the beginning of each and every one of his posts! Any reader taking his posts seriously does solely at reader's own risk.

             Joseph,

               your answers are very misleading and just plain wrong. Your answers are like a blind man throwing darts at a board in a pub; sometimes his dart is going to land somewhere around the board. Jospeh, you haven't been helpful in any way, wait you did provide a document from which I was able to extrapolate some useful information. But all of your words are muddling this post. You are just plain wrong.  I would thank you to refrain from attemting to answer, unless you are quite certain of the answer. I assure you are not certain of the answer in this case!

          For people stumbling on this post, I will try to articulate my question in the next reply.

[chad patterson]

     It seems my original question has been overlooked all together. My original question was this:

    Would some kind, enlightened soul help me to understand how to determine the maximum WAN speed that can be achieved with Cisco routers?

     I have learned part of the answer. For small packets, 64byte packets, there is a formula that can be applied to figure out the maximum wire speed that the Cisco 871 router, and indeed all Cisco routers can do:

• Mbps calculated by pps * 64bytes * 8bits/byte

     The Cisco 871 is capable of 25 Kpps for 64 byte packets, which is a very small packet and in the real world would indicate high fragmentation. So the maximum speed the Cisco 871 can obtain on the worst case scenario network is:

• 25 * 64 * 8 = 12.8 Mbps

     However this formula cannot be applied to large packets, or a best case scenario network. According to table 1 of the document provided earlier (white_paper_c11_595485.pdf), the best speed the Cisco 871 can obtain with this packet size is 197Mbps (I jump to this assumption because the Cisco 860 and the Cisco 870 both get 25Kpps at 64byte packets). Table 6 of this same document states the tests in an IMIX, translate 'real world', showed the Cisco 860 getting speeds of 27Mbps; this mirrors my experience with my Cisco 871.

     However I am in search of the formula that can applied to get this result. If I can figure out the max speed with 1500byte packets, then I can juxtapose the formulas to obtain a realistic, probable speed of a given router. I could estimate a maximum probable speed of somewhere between 12.8Mbps and ?Mbps.

     So again I ask, "Can some kind, enlightened soul help me to understand how to determine the maximum WAN speed that can be achieved with Cisco routers?"

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

chad patterson wrote:
     It seems my original question has been overlooked all together. My original question was this: 
    Would some kind, enlightened soul help me to understand how to determine the maximum WAN speed that can be achieved with Cisco routers?
     I have learned part of the answer. For small packets, 64byte packets, there is a formula that can be applied to figure out the maximum wire speed that the Cisco 871 router, and indeed all Cisco routers can do:
Mbps calculated by pps * 64bytes * 8bits/byte
     The Cisco 871 is capable of 25 Kpps for 64 byte packets, which is a very small packet and in the real world would indicate high fragmentation. So the maximum speed the Cisco 871 can obtain on the worst case scenario network is:
25 * 64 * 8 = 12.8 Mbps
     However this formula cannot be applied to large packets, or a best case scenario network. According to table 1 of the document provided earlier (white_paper_c11_595485.pdf), the best speed the Cisco 871 can obtain with this packet size is 197Mbps (I jump to this assumption because the Cisco 860 and the Cisco 870 both get 25Kpps at 64byte packets). Table 6 of this same document states the tests in an IMIX, translate 'real world', showed the Cisco 860 getting speeds of 27Mbps; this mirrors my experience with my Cisco 871. 
     However I am in search of the formula that can applied to get this result. If I can figure out the max speed with 1500byte packets, then I can juxtapose the formulas to obtain a realistic, probable speed of a given router. I could estimate a maximum probable speed of somewhere between 12.8Mbps and ?Mbps. 
     So again I ask, "Can some kind, enlightened soul help me to understand how to determine the maximum WAN speed that can be achieved with Cisco routers?"

"I would thank you to refrain from  attemting to answer, unless you  are quite certain of the answer. I  assure you are not certain of the  answer in this case!"

Chad, I would refrain answering if I was uncertain.

It's PPS for maximum size packets, on the WAN media, multiplied by bandwidth consumed per each packet.

"For people stumbling on this post, . . ."

What I think Chad is really trying to find, is a formula to predict a device's performance given just one PPS value such as the PPS for minimum size 64 byte packets.  Such a formula doesn't exist unless the device supports wire-speed/line-rate for all media packet sizes.  I'm unaware of any software based router that does, some hardware L3 switches do (assuming you first don't hit other hardware resource limitations).

Take note the document reference I provided earlier mentions RFC-2544 or similar performance tests and another NDR test, the operative word is test.

Perhaps this, http://www.cisco.com/en/US/products/hw/modules/ps2643/products_white_paper09186a0080091db8.shtml, (old) Cisco 7500 GEIP performance paper's graphs will help highlight how actual testing results can differ from theoretical media maximums and how the performance variations can be unexpected.  For example, if you look at figure 1 in the reference, notice how percentage of line rate dips for packets of 512 bytes.  Also notice the rapid increase in line rate percentage, for the first three tested packet sizes, but then the line rate percentage levels off.

Because device performance can vary, as the prior device's performance graphs demonstrate, given just a single PPS rate, or even PPS for minimum and maximum packet sizes, you cannot count on interpolation being accurate.

[chad patterson]

      To Joseph Doherty: I am unsure where you obtained your formula, but your answer is wrong, and you should just admit it. The answer is not "somewhere between 17Mbps and 309Mbps". I have applied Cisco's given algorithm, and found your answer to be just plain incorrect. And since the Cisco 871 uses Ethernet handoff at the WAN port, it could never exceed 98.7Mbps, even in the best case scenario. Although you seem to be finding in your subsequent replies that the max speed, in all scenarios, cannot be found with a simple algorithm. And that actually is 1/2 of the correct answer, which is what I have been looking for.

     Well it seems that I've come to the conclusion that my original question cannot be fully answered, as there seems to be no formula that can be applied to find the max speed. The only speed that can be predicted with an algorithm is the max speed in the worts case scenario (ie., smallest packet size). I've only found the formula for the best performance in the worst case scenario, which is; 25,000pps * 64bytes * 8bits/byte = 12.8Mbps

     This is the only speed listed in Cisco's performance document and I now understand why. Although in practice the router may be able to perform at greater speeds, there is no predictive algorithm which can be applied to find out exactly what those speeds may be. The only maximum speed that can be found is the maximum speed using small packet sizes (64bytes).

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Chad, I've explained several time where my formulas are "from".

We both, I believe agree, performance is documented by Cisco as 25 Kpps for the 870 series.

To convert this to bps per second, we need to agree what the packet's size is. In Cisco's older performance document, they note they're using 64 bytes for their packet. So, you're 100% correct, given that, you obtain 12.8 Mbps.

I've never said this was wrong, per se, just that most performance testing deals with real-world media, like Ethernet (as described in RFC 2544, and as noted in Cisco's later ISR performance document, which you've also been using since I provided it). Before getting back to real media, let's continue with pure L3 "raw switching performance", as Cisco notes in their older document.

Now if a 870 PPS rate is 25 Kpps, assuming that packet size isn't relevant to packet forwarding (and why should it be? - decision is based on IP destination address), then not only is 25 Kpps * 64bytes * 8bites/byte - 12.8 Mbps, we also have (as IP packets can range in size from 20 bytes to 65,535):

25 Kpps * 20 bytes * 8bites/byte = 4 Mbps

25 Kpps * 1500 bytes * 8bites/byte = 300 Mbps

25 Kpps * 65,535 bytes * 8bites/byte = 13.107 Gbps

Do I think any 870 is going to switch at 13 Gbps? No! Above are theoretical bandwidths, given constant PPS rates regardless of packet size.

Most people are interested in real-world, practical/possible performance, like can the 870 process full bidirectional 100 Mbps Ethernet? If not, what faction of that bandwidth can it handle.

So, even in my initial reply, I started using PPS rates to support line-rate for Ethernet, both for minimum size frames/packets and maximum size frame/packets.

I also noted in my first reply, I was using 1.488 Kpps per 1 Mbps for minimum size Ethernet. If we work this rate, "backwards", to obtain its effective "packet size", we get:

10 Mbps / 14.88 Kpps / 8 bits/byte = 83.96 bytes

Notice not 64 bytes! This because of all the L2 overhead. However, for 10 Mbps Ethernet (not 10 Mbps L3 packets), for minimum size frames/packets, the router still has to receive and forward at 14.88 Kpps.

If the router actually supports 25 Kpps for minimum Ethernet (real media) then the 870 can support 10 Mbps * 25 Kpps / 14.88 Kpps = 16.8 Mbps of Ethernet bandwidth or your 12.8 Mbps of L3 bandwidth.

If you do similar calculations for 1500 bytes packets, you get the multi-100 Mbps rates posted in my other posts, and bps differs for Ethernet and just L3.

I also noted in my original post, the theoretical maximum bps are usually not obtainable. You were then incredulous such rates were even possible, but Cisco's later ISR performance document notes maximum rates for the 860 and 880 series are both close to 200 Mbps, i.e. about 2/3 of the maximum theoretical rate.

So, again, my formulas are correct, just read and account for what I'm basing them on.

On the subject of being wrong, or, perhaps more likely, really not understanding, let's take even a few statements from your last posting.

"And since the Cisco 871 uses Ethernet handoff at the WAN port, it could never exceed 98.7Mbps, even in the best case scenario. "

Again, you're confusing effective L3 payload transmission rate. FastEthernet is 100 Mbps. The 98.7 Mbps applies to standard 1500 byte packets; effective L3 transfer rate is less with smaller packets, effective transfer rate is more with jumbo packets (i.e. never exceed 98.7 Mbps is incorrect with larger jumbo frames).

Additionally when working with modern Ethernet, it's often full duplex. So even using your 98.7 Mbps, handoff can be in and out, or twice your rate. Also, when looking at PPS, you need to account for duplex, if being used. For example, how much pps do you need to guarantee 10 Mbps Ethernet, duplex? It's 2x 14.88 PPS, which is more than an 870 is documented being guaranteed to provide.

"The only speed that can be predicted with an algorithm is the max speed in the worts case scenario (ie., smallest packet size)."

Also incorrect. All you need to know is PPS rate. What you can't predict is PPS rate for ANY packet size. Either you or someone else has to document values; often via testing (with traffic generators).

"The only maximum speed that can be found is the maximum speed using small packet sizes (64bytes)."

Again, incorrect, although today, that's often the only value provided because it's the worst case. BTW, years ago, when devices were so much, much slower, often "best" case performance was only documented.

The generic formula is # pps * packet-size (in bytes) * 8 bits/byte = # bps

Given all values but one, you can solve for the missing one. For example, in the newer Cisco ISR reference document, table 1 provides Kpps for 64 byte packets, and Mbps for 1500 byte packets, but assuming Ethernet or pure L3 packets size, you can convert the Kpps to Mbps; and the provided Mbps to Kpps. I.e. what Cisco did for this table is confusing, but sometimes they make it clearer, as their tables in the 7500 GEIP paper do.

I'll leave you to consider "found your answer to be just plain incorrect." and "Well it seems that I've come to the conclusion that my original question cannot be fully answered," yet a couple of my replies have been VIP Endorsed and of the 250+ views of this thread no one else has disagreed with what I'm trying to explain to you; why is that? (Once again, last is not to be taken as intended to be argumentative, but try to open your mind, read my posts carefully. If you have questions, ask.)

[Joseph W. Doherty]

Disclaimer

The  Author of this posting offers the information contained within this  posting without consideration and with the reader's understanding that  there's no implied or expressed suitability or fitness for any purpose.  Information provided is for informational purposes only and should not  be construed as rendering professional advice of any kind. Usage of this  posting's information is solely at reader's own risk.

Liability Disclaimer

In  no event shall Author be liable for any damages whatsoever (including,  without limitation, damages for loss of use, data or profit) arising out  of the use or inability to use the posting's information even if Author  has been advised of the possibility of such damage.

Posting

Chad, have you seen (this also old)?  http://www.cisco.com/web/partners/downloads/765/tools/quickreference/switchperformance.pdf

The reason I ask, unlike the (old) router performance document (http://www.cisco.com/web/partners/downloads/765/tools/quickreference/routerperformance.pdf), note packet size isn't described and the only documented bandwidth is for fabric bandwidth.  So, how can anyone use this document to determine if a switch's performance is adequate for how they might use it?

Well, since these are Ethernet switches, let's assume we're looking at Ethernet PPS rates and Ethernet bandwidths.  Let's also assume, PPS rates are for minimum size Ethernet.

If I needed a wire-speed/line-rate 48 gig ports switch, I "know" I need 48 Gbps * 1.488 Mpps = 71.424 Mpps; and 48 Gbps * 2 (duplex) = 96 Gbps fabric bandwidth.

If you examine the specs for a 4948, it's listed as 72,000,000 Mpps (Cisco really meant 72,000,000 pps or 72 Mpps; i.e. column label and column values don't correctly agree) with a 96 Gbps fabric.  Looks to me, according to my formulas for Ethernet, this is a wire-speed/line-rate capable switch.  Cisco's 4948 data sheet appears to agree as it says ". . . providing wire-speed routing and switching performance . . . ".

However if you go by minimum packet size is 64 bytes, then 72 Mpps * 64 bytes * 8 bits/byte = 36.864 Gbps.  That doesn't seem to match well with our 48 gig ports, does it?

Again, using 64 bytes per packet isn't wrong, per se, it's just not what is commonly used when dealing with device performance for Ethernet.

PS:

If you still feel I'm wrong, post a new (thread) similar question and I promise I won't post an answer to it.