RFA RSSL recommended size of fragment for large RDM symbol list

@RG1

Hi Rimma,

What version of RFA C++ you are using?

As far as I understand the numinputbuffer is RSSL Channel Parameter. It's the number of input buffers available to an RSSL channel. This is the amount of data RSSL will attempt to read off of the wire with each socket read call. But the problem you are experiencing is the exception when calling MapReadIterator::forth. Normally, if the entry at this position is blank or with Invalid / Incomplete data an 'InvalidUsageException' is thrown.

• Does the problem always occur when your provider publishes the same symbolist? And does it occurs when it decodes the map from the same/specific fragment?

I think you can turn on the RSSL tracing log on both the Provider and Consumer side. So you can compare the data when the problem occurs.

With the information I have, for now, I still not quite sure why the problem may relate to the default input buffer. If the Provider can encode the fragments without any problem, the consumer should be able to handle the messages as well in my opinion.

Hi Moragodkrit,

The problem happens when provider publishes large symbol list ( map of about 6K items) . Enabled RFA tracing on both sides. RDM symbol list message is truncated on consumer side. What can be explained by a small read buffer, I suspect. Consumer is on RFA C++ version 8.1.0.L1.

Moragodkrit, in addition, smaller fragments work, consumer is able to process the full list. Question - what's the limit on size of fragment. It is clearly driven by the consumer read ability.

Regards

Rimma

Hi @RG1

Whilst I am not sure about a limit on the Fragment size from an OMM SymbolList Encoding/decoding point of view, my understanding is that whilst you can use payloads larger than 64k - the TREP infrastructure components i.e. ADS and ADH are optimised for 64k message sizes.

Therefore, purely from an ADS+ADH performance point of view, I would recommend you limit any multi fragment messages to 64k size max for each fragment.

I am also aware that making batch requests, there is a limit of 64K on the batch request message payload as well - so the same may apply to Multipart message fragments too.

Hopefully, my colleague will be able to provide a definitive answer.

Hi @umer.nalla ,

Our observation on the infra side: such long message/map is not a problem for ADH/ADS. The issue is observed when symbol list is requested for the first time, request goes to the provider ( interactive) . Infra passes through the response to the consumer, which fails to process it. When subscribing again, the symbol list comes back from ADH cache fragmented, and consumer able to process all fragments of the list correctly. Obviously, ADH does fragmentation, I suspect that ADH/ADS must have larger read buffer(s) , than the regular consumer, for better throughput, and able to swallow such payload.

My second question was if the issue relates to the read system buffer, as according to RFA configuration guide,

"RSSL Channel Parameter. Number of input buffers available to an RSSL channel. This is the amount of data RSSL will attempt to read off of the wire with each socket read call. This will be [numInputBuffers * input buffer size bytes] in length. "

Regards

@RG1

I assume that the symbolist Map contains around 6k map entry count(6k items). According to RFA Dev Guide, in theory, a Map currently has a maximum entry count of 65,535 but a recommend overall fragment message seems to be around 64k as Umer said.

• Is there any problem when connecting the consumer app to the provider app directly? Just want to make sure that it does not relate to the infra.

I did a quick test by modifying Provider_Interactive example to publish symbol list RIC which contains around 6k items/map entry. I have added key name item<count> where count starts from 1-6k. Then connecting StartConsumer_SymbolList(from RFA C++ 8.1.3 package) to the provider app. The consumer can process 6k items without the exception error. The Message size from the trace is around 178k(below is sample trace). That is a reason that I can't confirm that the problem actually relates to the default input buffer.

<refreshMsg domainType="RSSL_DMT_SYMBOL_LIST" streamId="3" containerType="RSSL_DT_MAP" flags="0x11E8 (RSSL_RFMF_HAS_MSG_KEY|RSSL_RFMF_SOLICITED|RSSL_RFMF_REFRESH_COMPLETE|RSSL_RFMF_HAS_QOS|RSSL_RFMF_CLEAR_CACHE|RSSL_RFMF_HAS_PART_NUM)" groupId="0" partNum="0" qosDynamic="0" qosRate="1" qosTimeliness="1" dataState="RSSL_DATA_OK" streamState="RSSL_STREAM_OPEN" code="RSSL_SC_NONE" text="MessageComplete"  dataSize="178910">
    <key  flags="0x7 (RSSL_MKF_HAS_SERVICE_ID|RSSL_MKF_HAS_NAME|RSSL_MKF_HAS_NAME_TYPE)"  serviceId="1" name="0#ITEMS" nameType="1"/>
    <dataBody>
        <map flags="0x1 (RSSL_MPF_HAS_SET_DEFS)" countHint="0" keyPrimitiveType="RSSL_DT_ASCII_STRING" containerType="RSSL_DT_FIELD_LIST" >
            <fieldSetDefs>
                <fieldSetDef setId="0">
                    <fieldSetDefEntry fieldId="3422" dataType="RSSL_DT_RMTES_STRING" />
                    <fieldSetDefEntry fieldId="1" dataType="RSSL_DT_UINT" />
                </fieldSetDef>
            </fieldSetDefs>
            <mapEntry flags="0x0" action="RSSL_MPEA_ADD_ENTRY" key="item1" >
                <fieldList flags="0x3 (RSSL_FLF_HAS_FIELD_LIST_INFO|RSSL_FLF_HAS_SET_DATA)" fieldListNum="0" dictionaryId="1">
                    <fieldEntry fieldId="3422" dataType="RSSL_DT_RMTES_STRING" data="appSymbol2"/>
                    <fieldEntry fieldId="1" dataType="RSSL_DT_UINT" data="3056"/>
                </fieldList>
            </mapEntry>
            <mapEntry flags="0x0" action="RSSL_MPEA_ADD_ENTRY" key="item2" >
                <fieldList flags="0x3 (RSSL_FLF_HAS_FIELD_LIST_INFO|RSSL_FLF_HAS_SET_DATA)" fieldListNum="0" dictionaryId="1">
                    <fieldEntry fieldId="3422" dataType="RSSL_DT_RMTES_STRING" data="appSymbol1"/>
                    <fieldEntry fieldId="1" dataType="RSSL_DT_UINT" data="3056"/>
                </fieldList>
            </mapEntry>
            ...
            <mapEntry flags="0x0" action="RSSL_MPEA_ADD_ENTRY" key="item5999" >
                <fieldList flags="0x3 (RSSL_FLF_HAS_FIELD_LIST_INFO|RSSL_FLF_HAS_SET_DATA)" fieldListNum="0" dictionaryId="1">
                    <fieldEntry fieldId="3422" dataType="RSSL_DT_RMTES_STRING" data="appSymbol2"/>
                    <fieldEntry fieldId="1" dataType="RSSL_DT_UINT" data="3056"/>
                </fieldList>
            </mapEntry>
            <mapEntry flags="0x0" action="RSSL_MPEA_ADD_ENTRY" key="item6000" >
                <fieldList flags="0x3 (RSSL_FLF_HAS_FIELD_LIST_INFO|RSSL_FLF_HAS_SET_DATA)" fieldListNum="0" dictionaryId="1">
                    <fieldEntry fieldId="3422" dataType="RSSL_DT_RMTES_STRING" data="appSymbol1"/>
                    <fieldEntry fieldId="1" dataType="RSSL_DT_UINT" data="3056"/>
                </fieldList>
            </mapEntry>
        </map>
    </dataBody>
</refreshMsg>

Since this case requires a further investigation to confirm the issue.

Can you please open a new support ticket for this issue?

You can open a new support case by going to RFA Section and click Contact Premium Support

Hi @RG1

Parameters such as numInputBuffers relate to the transport/network layer whereas the OMM encoding/decoding of messages and payloads is happening at the API level - therefore the changes to numinputBuffers should not impact the ability to encode and decode larger OMM messages.

Just to be clear numInputBuffers specifies the number of sequential input buffers to allocate for reading data into. This controls the maximum number of bytes that can be handled with a single network read operation. Changes to this value will affect how fast we can pull in data from a full OS receive buffer - not the size of any OMMMessag payloads being transported on the network.

Tested direct consumer-publisher connection - no issue with the large symbol list. Infra seems to be a culprit. ADH/ADS version trep3.3.0.L1

I tried to open premium support ticket, several times I get "an error occurred, Please try later.." on Submit. Please fix the service.

Thank you

Hi @RG1

Apologies for the problem you are facing raising a ticket - we are having some issues with the new developer portal.

I have created a Support Ticket in your name and email address ref 09235979 - you should get an email shortly. If not please post back here.